Who exactly we sell to — kept sharp, revisited monthly.
Definition
CURRENT ICP (v1, July 2026): Mid-market companies (50-500 employees) actively building AI agents, with meaningful monthly agent spend they are losing control of. Buyer: VP Eng / CTO / eng leadership who can self-evaluate and buy. Motion: PLG — they land on Arena (free), see their waste, convert to ~$250/mo. NOT the ICP right now: large enterprises (17-person decision stacks, 6-12 month cycles), companies without active agent programs. Revisit: does the ICP hold as expansion revenue data comes in?
ICP prospect signal scan — 2026-09-03 run (3 added, below the 5 target; LinkedIn content search is saturating)
ADDED THIS RUN (3, target was 5):
1. Karthik Kannan — Founder & CEO, Anvilogic (115 emp, Series C) — Signal 1 — ICP confidence HIGH. Wrote "Tokens Are the New Headcount Nobody's Budgeting For". Second contact at an account where Deb Banerjee (CTO) is already logged; the cost angle is far stronger on this record. Highest-priority follow-up from this run.
2. Tamal Biswas — VP of Cloud Platform & Infrastructure, Calix (1,926 emp, Mar 2026 Revelio) — Signal 1 — MEDIUM-HIGH. "Your AI vendor is getting cheaper. Your AI bill is getting bigger." Calix already has a Director of Engineering — Agentic Platform in the library (SN Raju Kattari), so this is a second, more senior contact at a live agent-building account. Note: headcount is near the 2,000 ceiling and growing 14% YoY — re-verify next run.
3. Laxmikanth Katheragandla — Senior Director, Agentic Platforms, JAGGAER (1,503 emp) — Signal 4 — MEDIUM. Hiring a Principal Engineer to build JAGGAER's agentic platform; exact title fit, but the signal is capability-building rather than expressed cost/reliability pain. Probe, don't assume.
SIGNAL BUCKET PRODUCTIVITY:
- Signal 1 (ICP writing about cost/control): the ONLY productive bucket — 2 of 3 adds. Best-performing queries were operator-language phrases, not the generic ones in the skill: "cost per agent" OR "per-agent cost" production; "our agents" token spend production; inference cost agentic workflow.
- Signal 4 (ICP writing about agents): 1 add, from "AI agent platform engineering scale". Mostly returns hiring posts — which is actually a usable proxy for "company is standing up an agent org right now".
- Signal 2 (ICP engaging with non-ICP agent content): ZERO adds. Read comment threads on 4 high-engagement posts (Manish Khattar/Icertis 13 comments, Ayush Shaji/PwC 21 comments, Abraham N. 12 comments, Sudheendra G. 7 comments). Commenters were almost entirely staff engineers, solo consultants and vendors pitching their own product — no Director+ buyers.
- Signal 3 (ICP engaging with competitor content): ZERO adds. Searches for helicone/portkey/litellm and langfuse/langsmith/braintrust returned tutorial and explainer content from data scientists and consultants, not practitioners at agent companies.
WHY THE RUN UNDERSHOT — this needs a change of method, not more of the same:
(a) LinkedIn content search ranks by engagement within Vishnu's network, which skews heavily to India-based enterprise AI consultants, recruiters and content creators. Genuine VP/CTO-level operators at 50–2,000-emp agent companies post rarely and rank low. Roughly 90% of results across 12 distinct queries were non-ICP.
(b) The People Library is now at ~1,100 records and heavily saturates the obvious AI-native companies (Cresta, Writer, Glean, Kore.ai, Aisera, PolyAI, Decagon, Moveworks, CrewAI, Clay etc. all already covered), so the remaining supply from generic searches is thin.
(c) Several strong-signal people failed the hard headcount gate and were deliberately NOT added: Saikumar Thota (VP Engineering, Collectors — sources range 1,200 to 3,000+, could not confirm ≤2,000; signal quality was excellent, "cost per successful business outcome", 10+ production AI use cases, 31% lower model cost per workload — worth revisiting if headcount can be pinned down); Manish Khattar (Senior Director, Icertis ~2.4K); Anil Gupta (Global Head AI Engineering, Concierto/Trianz ~2.4–2.9K); Gurbans Chatwal (VP, Fiserv ~40K); Paritosh Dagar (Chief Technology & AI Officer — left KPMG UK May 2026, current company not disclosed, unverifiable); Sattyam Jain (Attri.ai — excellent cost pain, but Tech Lead, below the Director floor).
RECOMMENDED CHANGES FOR NEXT RUN:
- Replace the five generic Signal 1 queries in the skill with the operator-language ones that actually worked (listed above), and add: "impact per token", "cost per successful outcome", "agent retry storm", "we run N agents".
- Stop budgeting effort on Signal 2/3 comment-mining via LinkedIn content search; it cost the majority of this run's time for zero adds. Either drop those buckets or reach them a different way (e.g. reactions lists on posts by known agent-company leaders, which are richer than comment threads).
- Add a company-first pass: build a target list of 50–2,000-emp Series A–C agent-native companies NOT already in the People Library, then find their Director+/VP AI-eng leaders directly, using their own writing (company blog, conference talk) as the signal — the method that produced the Deb Banerjee record.
OUTREACH COPY IMPLICATION (see VOC entries filed this run):
Do not lead with "cut your token bill". Four senior leaders this run explicitly framed unit price as the wrong target — the resonant framing is per-workflow / per-agent attribution and cost per successful outcome, plus controls that run before the call is dispatched rather than dashboards that report after the money is gone.
sales intelclaude-connector · 2 Sept 2026
ICP prospect signal scan — 2026-09-02: 6 added, LinkedIn post search is now exhausted as a channel
RUN SUMMARY — icp-prospect-signal-scanner, 2026-09-02 (people library was at ~1,080 before this run)
ADDED THIS RUN (6 new, all deduped against the existing people list):
1. Shubham Agarwal — VP Engineering, Leena AI (201-500, Series B) — Signal 4 — High
2. Willie Yao — Head of Engineering, Clay (201-500, Series C) — Signal 4 — High
3. Dan Eisenberg — Head of Engineering, Hex (51-200, Series C) — Signal 2 — High
4. Patrik "totte" Torstensson — Head of Engineering, Lovable (51-200, Series C) — Signal 4 — High
5. Daniel Sheard — Head of Enterprise Engineering, Lovable (51-200, Series C) — Signal 4 — Medium
6. Elliot Trabac — Director of Engineering, Gorgias (201-500, Series C) — Signal 4 — Medium
REJECTED / SKIPPED (with reason, so future runs don't re-litigate):
- Already in the brain, surfaced again: Dennis Cui (Decagon), Masashi Beheim (Parloa), Kevin Gao + Brian Ngo (Sierra), Jeff Barg (Clay), Pierre-Alexandre Masse (Gorgias), Victor Duprez (Gorgias), Amol Jain (Replit), Diego Comas (Sourcegraph), Cornelius Suermann (n8n), Ming Yin (Cresta), Klaus Krogmann (Cognigy).
- Karim Frenn (Director of Eng, Fin/Intercom) and Brian McDonnell (Sr Director Eng, AI Group, Intercom): LinkedIn band is 1K-5K, cannot confirm <=2,000 — skipped per the >2,000 rule. Worth a manual headcount check; Fin is one of the largest production agent deployments in the market.
- Dom Busser (Director of Engineering, Vercel, 501-1K): headcount fits and Vercel ships agents, but Series F — outside the Series A-C band. Held back deliberately.
- Barr Moses (Monte Carlo), Shawn McAllister (Solace), Leon Eisen, Ganesh Angadi, Prashant Arya, PHANI KUMAR KOLLA, Siva Prasad GV, Ed Shields, Cole Murray: vendors, consultants or influencers, not buyers.
- Antoine Dulac (Criteo, ~3.5K) and Chiara Caratelli (Sr ML Engineer, Prosus): both wrote the best cost content found all run, but fail on headcount and seniority respectively. Chiara's post (cost per response down 80%, removing old context cut tokens 44% but made requests 68% more expensive by destroying prompt-cache reuse) is worth reading for messaging even though she is not a target.
WHICH BUCKETS WORKED:
- Signal 1 (ICP writing about agent cost): effectively zero yield. Five keyword variants ("agent cost LLM production", "agent observability cost per run", "token budget scaling", "LLM spend agents production", "AI agents production reliability cost") returned job posts, open-to-work posts, newsletter-farming consultants and fractional CTOs. Not one qualifying author across ~40 posts read.
- Signal 3 (competitor mentions): worst performer. "langfuse OR langsmith OR braintrust OR helicone" was silently autocorrected by LinkedIn to a pharma query and returned MASH/oncology content. Single-term searches ("Langfuse") returned vendor marketing and CV keyword-stuffing. OR syntax does not work in LinkedIn content search — future runs must use one term per query.
- Signals 2 and 4 produced everything usable, but only after switching method: LinkedIn company People pages (company/<slug>/people/?keywords=<title>) to find Director+ titles at qualifying companies, then that person's /recent-activity/all/ page to capture the actual signal. This is far higher yield per call than content search and gives a verifiable headcount band from the company page.
EMERGING PATTERN — 3 VOC entries logged this run, and the through-line is that cost has stopped being an infra concern:
- Lovable sells credits, not seats, so agent run cost sits directly under revenue recognition and margin.
- Clay went to 80+ custom internal agents in a week — agent count outruns per-agent accounting.
- Hex's eng lead is publicly sceptical that benchmark performance predicts production reliability on long-horizon tasks.
Outreach implication: lead with cost-per-run as a margin and revenue-quality problem for the CFO-adjacent conversation, not as an observability feature. The 1→N sprawl framing ("you added 40 agents last quarter, what did each one cost?") is the sharpest opener the data supports.
CHANNEL WARNING FOR THE NEXT RUN: the brain already contains contacts from essentially every well-known agent-native company (Sierra, Decagon, Cresta, Parloa, Ada, Forethought, Writer, Glean, Harvey, Abridge, Hippocratic, PolyAI, Aisera, Moveworks, Lorikeet, Tennr, 11x, Norm Ai, EvenUp, Relevance, Crescendo, Vapi, Retell, Clay, Gorgias, Replit, Sourcegraph, n8n, Hex, Lovable...). Duplicate rate this run was roughly 60%. Next run should either (a) go one level deeper on titles already covered (Staff/Principal is below the Director floor, so this means Director of Platform / Head of Infrastructure / Head of Applied AI variants), or (b) move to non-AI-native SaaS at 50-2,000 that has recently shipped agents — that pool is untouched in the brain.
RUN SUMMARY — ICP prospect signal scanner, 2026-08-31 (second run today). Target was 5+; 3 added. No fabrication: every candidate that failed was failed on a verified company-size or seniority check, and those checks are recorded below so future runs do not re-spend on them.
ADDED (3 new people, deduped against the 1,064 already in the People Library)
1. Amjad Ghazi — VP Engineering, Lentra (501-1K emp, Series B digital-lending SaaS, Pune) — HIGH. Signal 1. Post 31 Aug: "Your LLM gives a great answer. Your software can't use it... engineering teams end up building parsers, regexes, repair logic and retries around inherently variable language." Closes with "Who owns the complexity required to make its response deterministic enough for software to act on?"
2. Kumar Ashutosh — AVP Data Engineering & Product Management, Easyrewardz (201-500 emp, CRM/loyalty SaaS, Gurgaon) — MEDIUM-HIGH. Signal 3 (commented on LangChain's MCP agent-API post). Runs a 25-person data/AI team; explicitly transitioning the platform stack "toward GenAI-powered, agentic architectures"; architecting a production voice agent on Exotel + Deepgram + Claude API + ElevenLabs targeting 70%+ inbound automation.
3. Dr. David Noel Ng — Head of AI Product & Engineering, yoummday (201-500 emp, Munich, "Enterprise AI Execution Layer for CX") — MEDIUM. Signal 1, inference-cost/efficiency writing (inference-path optimization, layer duplication, recirculation). Pain is adjacent to Alpha's wedge (model-level efficiency) rather than agent cost-per-run.
SIGNAL BUCKET PRODUCTIVITY
- Bucket 1 (ICP writing about agent cost/control) — 2 of 3 adds. Both came from LinkedIn post search, but only from queries phrased the way an engineer writes, not the way a marketer does. "agent cost LLM production" and "context engineering agents cost production" worked. Generic ICP-vocabulary queries ("building AI agents production 2026", "AI agent platform engineering scale", "agentic AI cost control observability") returned recruiters, course-sellers and portfolio projects, same as the last two runs.
- Bucket 2 (ICP commenting on non-ICP content) — COMMENT MINING NOW WORKS (this was the untested path flagged on 2026-08-31 run 1). Method below. Yield was real but thin: roughly 1 senior technical commenter per 13 comments, and most of those failed on company size. Best thread by far was Shantanu Rastogi's token re-send post — see the VOC filed today.
- Bucket 3 (competitor content) — 1 of 3 adds, via LangChain's own company-page post rather than a keyword search. Single-vendor keyword searches (LiteLLM, Langfuse, AgentCore, Portkey) returned near-zero ICP. Confirms again: never use OR in LinkedIn content search.
- Bucket 4 (ICP writing about agents) — nothing net-new this run.
METHOD UNLOCKED THIS RUN (keep this — it is the main deliverable)
LinkedIn search-result cards expose no post permalinks, which is what blocked comment mining on previous runs. The way through:
1. Open the author's /recent-activity/all/ page and regex the raw HTML for /urn:li:activity:\d+/ — the URNs are in the page source even though no anchor tags are.
2. Open https://www.linkedin.com/feed/update/urn:li:activity:<id>/ directly.
3. Click every button matching /load more comments|previous comments/i in a loop, wait, then read comments out of the <article> elements together with the commenter's /in/ URL.
4. For reactions: click the "N reactions" button, then scroll with a REAL mouse-scroll at the modal's coordinates — programmatic scrollTop does not trigger the modal's infinite scroll and you will only ever get the first 10 names. Mouse-scroll gets all of them with name + headline, which is a much faster ICP filter than reading comments.
Same trick works on company pages (/company/<slug>/posts/) for vendor content — that is how the LangChain thread was reached.
VERIFIED DISQUALIFICATIONS (recorded so no future run re-spends on them)
- Michal Piszczek, CTO, Archdesk — RE-CONFIRMED OUT. He was the single best signal of the run (see VOC on re-send fraction) and Archdesk's LinkedIn page self-declares "51-200 employees", but independent sources put actual headcount at 30-63 (Tracxn 36 as of Apr 2026; PitchBook 63; Getlatka ~30). Below the 50-2,000 floor. The 2026-08-29 run reached the same conclusion. LESSON: LinkedIn's self-declared employee band is not a size check — verify against Tracxn/PitchBook before adding anyone from a company in the 51-200 band.
- Saikumar Thota — RESOLVED, and OUT. The 2026-08-31 run 1 flagged him as "WORTH A MANUAL LOOKUP" (strongest content match, profile unresolvable). Found: https://www.linkedin.com/in/saikumar-thota/ — VP of Engineering / CTO / Head of AI at Collectors (PSA). Collectors has 3,000+ employees, so he is out on size. His content remains excellent (agent eval stack: capability, tool-use/trajectory, safety/permission, and cost+latency evals) and is worth reading for outreach copy even though he is not a target.
- Patrick D'Souza, Senior Director AI Enablement, Zeta Global — out (public company, ~2,000+, not Series A-C).
- Debasish Bhattacharjee (SAP), Moses Pawar (Apple), Vivek S. (Apple), Waleed Hamied (Qualtrics), Krishna Gogineni (Cohesity), Claudiu Branzan (ServiceNow), Shekhar Agrawal (Solventum), Amit Jagtap (JPMorgan), Alok Kulkarni (Nationwide Building Society), Guru Majgaonkar (NICE), Shantanu Rastogi (Maersk), Amit Hasson (Payoneer) — all >2,000 employees.
- Carmelo Juanes (CTO, Invofox, 11-50), Feres Kasdallah (R. STAHL — right size but zero posts, no signal), Eli Brosh (VP AI, Papaya Global — 1K-5K band, cannot confirm <=2,000, and only corporate reposts), Guy Shalev (Fetcherr — signal is a corporate repost), Bernie Camus (Sprout.ai — corporate reposts only), Anup Sahoo ("8 agents in production, 80% adoption" — great signal but AI Engineering Manager, below Director, and currently "open to roles"), Rahula Raj (Senior AI PM, The Luxury Closet — below Director), Tomas Stejskal (CTO, Zisky.ai — too small), Vijay Poudel (building Traccia, an agent control plane — competitor, not buyer).
- Siva Adhikarla (AVP Engineering, JSW One Platforms) — DUPLICATE, already added 2026-08-28.
RECOMMENDATIONS FOR NEXT RUN
1. Lead with comment/reaction mining on 5-8 posts, using the URN method above. It is the only bucket that produced net-new names this run and it was never actually executed before today.
2. Pick the posts by topic fit, not by engagement volume. High-reaction posts on Anthropic/LangChain company pages were almost pure noise; a 39-reaction post specifically about token re-send produced the two best technical voices of the run.
3. Geography: the 50-2,000 band keeps landing in India and Germany/UK, not the US. US people searches again skewed to Apple, SAP, ServiceNow, Qualtrics, Cohesity, JPMorgan. Keep leading with India + DACH + UK geo filters.
4. Vertical SaaS shipping agents inside an existing product remains the richest vein — all 3 adds today are that shape (lending SaaS, loyalty/CRM SaaS, CX platform). Keep away from the AI-native agent-company list; it is exhausted.
5. Add a headcount verification step before any add from a company in LinkedIn's "51-200" band. Archdesk cost two runs because of this.
OUTREACH COPY IMPLICATIONS (2 VOC entries filed today)
(a) "Who owns the complexity of making the model deterministic enough for software to act on?" — Alpha should answer an ownership question, not sell a tool. This is the sharpest buyer-voiced framing captured so far.
(b) "Re-send fraction" is a concrete, unclaimed metric. The exact quote is "Almost nobody I ask has measured it." A cold open that offers to show a team their re-send fraction is a stronger hook than any cost-savings claim.
(c) The fuel-gauge metaphor for per-agent budget appeared unprompted from an Oracle AI Strategy director: "the agents need a budget... like starting a trip and knowing how much gas you have in the tank." That is Alpha's budget-per-agent story in the buyer's own words and should be tested verbatim.
sales intelclaude-connector · 30 Aug 2026
ICP prospect scan 2026-08-30: 6 added, but the LinkedIn post-search method is exhausted — method change required
RESULT: 6 new people added (ids 1022-1027). Target of 5 met. But the run surfaced two problems worth acting on before the next one.
PEOPLE ADDED
1. Seth Shelnutt - VP of Engineering, Coder (51-200, Austin) - STRONGEST. Coder has staffed a named "AI Gateway" workstream (colleague Marcin Tojek's headline reads "AI Gateway at Coder | Senior Engineering Manager"). That is the token-routing/spend-control problem, in-house, right now. Best warm-open of the six.
2. Johannes Goller - VP Engineering, Parloa (201-500, Berlin, Series C, AI agent management platform).
3. Danny Yates - Director/Head of Engineering, Tray.ai (201-500, Series C, "Merlin" agents). Sells per-workflow-run, so agent token cost hits gross margin directly - good wedge.
4. Jason Rosenfeld - Head of Engineering, Candid Health (51-200, AI medical billing agents).
5. Steve Yazicioglu - Head of Forward Deployed Engineering, Candid Health (same company; influencer not budget owner).
6. Joshua Tepper - Director Engineering, Clarifai (51-200, Series C). CAVEAT: Clarifai plays in agent orchestration themselves - qualify as peer vs. buyer before outreach.
PROBLEM 1 - THE PRESCRIBED SEARCH METHOD NO LONGER WORKS
All 13 LinkedIn post searches in the skill were run. Yield of ICP-matching authors or commenters: ZERO. LinkedIn's content ranking on this account returns AI trainers, course sellers, recruiters and analyst/consultant commentary, not senior technical leaders at agent companies. Two mechanical findings for whoever edits the skill:
- OR syntax is not supported. "helicone OR portkey OR litellm" tokenised into unrelated results (a page of paediatric oncology posts). Search one term at a time.
- Quoted exact phrases DO work and are the only reliable lever ("cost per agent", "agents in production" both returned on-topic results). Unquoted multi-word queries get ignored.
Signals 1, 2 and 3 produced nothing addable. All six people came from Signal 4, reached a different way.
PROBLEM 2 - THE OBVIOUS COMPANIES ARE MINED OUT
The brain already holds 1017 people across roughly 450 companies. The first six candidates found this run - at Parloa, Cresta, Writer, Decagon - were 5/6 already present (Masashi Beheim, Dan Bikel, Mo Shaker, Dennis Cui, Jove Zhong). Note Beheim (ids 433, 587) and Bikel (ids 176, 591) each exist TWICE - pre-existing duplicates worth merging.
Also: Common Room was worked up as a candidate and then dropped - it has just been acquired by Zoom. Paradox.ai dropped for the same reason (now Workday). Cognigy dropped (now NiCE). Worth a periodic acquisition sweep of existing pipeline accounts for the same reason.
WHAT ACTUALLY WORKED - RECOMMENDED METHOD FOR NEXT RUN
Company-first, not post-first: LinkedIn company People tab with a title keyword, e.g.
https://www.linkedin.com/company/<slug>/people/?keywords=head%20of%20engineering
This returns real names, titles and profile URLs, with verified headcount from the company page header. Vary keyword across: "head of engineering", "VP", "vice president", "director", "engineering". Page needs ~4s to load before scraping.
Target list should be companies NOT among the ~450 already covered. Untried and still open: Numa, Kalepa, Hummingbird, Lucinity, Telnyx, Plivo, Exotel, Everlaw, Doxel, Graphite, Micro1, AirOps, Command Zero.
HONESTY NOTE ON DATA QUALITY
None of the six expressed a pain point in their own words. Pain points, challenges and must-haves in their records are explicitly labelled as INFERRED from role and company - nothing was invented and no quotes were attributed to them. Only Coder carries a real observed company-level signal. Before outreach, each should get an activity check (linkedin.com/in/<slug>/recent-activity/all/) to find something genuine to open on.
OUTREACH COPY IMPLICATION
Two VOC entries logged this run (ids 315, 316). The market vocabulary has converged hard on "cost per successful outcome" over "cost per token", and separately on run-to-run inconsistency at scale. Both are analyst/consultant voice rather than buyer voice, so they are useful for framing language buyers already recognise - not as evidence of demand. Alpha's compounding/harness story maps cleanly onto both; the honest framing is that we make cost-per-outcome measurable and consistency improvable, rather than claiming buyers are asking for it.
sales intelclaude-connector · 29 Aug 2026
ICP Prospect Signal Scanner — run 29 Aug 2026: 5 added, and LinkedIn content search is now a dead channel
RESULT: 5 new people added (Brain now ~1,021 people). All deduped against the existing people list before writing.
Added:
1. Vikram Verma — VP Engineering & Site Leader, Level AI (51-200) — High
2. Raaghu K — Senior Director of Engineering, Level AI (51-200) — High
3. Waseem Alshikh — Co-founder & CTO, WRITER (201-500) — High
4. Joschka Braun — Head of ML, Tennr (201-500) — Medium
5. Tommi Holmgren — VP of Product (Agents & Automation), Sema4.ai (51-200) — Medium
Level AI is now a two-contact account (VP Eng + Sr Director Eng) — highest-priority account out of this run.
WHAT WORKED / WHAT DIDN'T — this matters more than the 5 names:
Signal buckets 1, 2 and 3 as written in the skill produced ZERO qualifying people this run. Every prescribed LinkedIn content search (agent cost LLM production; AI agent reliability production; cost per agent run / agent spend; langfuse OR langsmith OR braintrust; helicone OR portkey OR litellm; AgentCore; agent observability cost per run) returned consultants, trainers, content creators, and big-co middle managers — not ICP. Two mechanical causes:
(a) LinkedIn content search is ranked against the logged-in account's own network, and this account's graph is heavily weighted toward India-based consultants/educators, so the same non-ICP cohort recurs regardless of keyword.
(b) LinkedIn's OR operator is broken for these queries — "helicone OR portkey OR litellm" returned paediatric-oncology and rheumatology posts. Use single-keyword content searches only.
WHAT DID WORK — company-scoped people search:
LinkedIn people search with a company name + a seniority keyword (e.g. /search/results/people/?keywords=Parloa%20engineering%20AI%20agents) returns clean, high-signal lists of named leaders with titles. Then: (1) read the person's /recent-activity/all/ for real pain language, and (2) read /company/<slug>/about/ for a verified headcount band. All five adds were sized off the LinkedIn company page, not guessed.
RECOMMENDED SKILL CHANGE for the next run: invert the method. Lead with a target-company list (Series A-C, 50-2,000, agent-native), run company-scoped people search, then check activity for pain signals — instead of leading with keyword content search. Keep content search only as a secondary pass with single-keyword queries.
COVERAGE NOTE: Decagon, Cresta, Parloa and Level AI leadership are already largely in the Brain from prior runs (Dennis Cui, Masashi Beheim, Ming Yin, Ayush Pallav, Deepank Sharma, Shivam Khandelwal, Sunny Rekhi, Srinivasa Rao Yasarla all already present). The obvious agent-native names are getting exhausted — next runs need fresh company territory. Untouched/unverified this run: Sierra, Cognition, Clay, 11x, Harvey, Ada, Forethought, Assembled, Netomi, Ushur, Ambience, Abridge, Norm Ai, Rogo, Ema, Maven AGI, Baseten, Fireworks, Anysphere.
SKIPPED and why (so we don't re-litigate): Anil Kumar A. (Ushur) and Vignesh Saravanai (Kore.ai) — could not retrieve a LinkedIn headcount band, and the skill forbids adding without confirmed size. Nikhil Gupta (CTO, Vapi) and Shivam Khandelwal — name already present in Brain. Cognigy leaders — company page now resolves to NICE-Cognigy, i.e. absorbed into a large parent, outside ICP. PolyAI — LinkedIn page shows 11-50, under the floor.
PATTERNS FOR OUTREACH COPY (two VOC entries filed this run):
1. "Demo-grade evals don't survive production." Three ICP-level voices independently said the test they run is not the test they need — Waseem Alshikh (WRITER), Ming Yin (Cresta), Dennis Cui (Decagon). The sharpest framing is Cresta's: production failures should become regression tests automatically. That is loop engineering, which is our language already.
2. "Cost only becomes visible at scale." Decagon's speech-to-speech note — S2S "process every slice of audio, not just words, which makes them more expensive at scale" — plus the fact that Decagon now runs 80%+ of traffic on self-trained models. Cost framed per unit of work, not per month. Lead with cost-per-run, not monthly spend.
HONESTY FLAG: three of the five adds (Vikram Verma, Joschka Braun, and to a lesser degree Tommi Holmgren) have no cost/reliability pain stated in their own public words. They qualify on role + company + agent-activity, not on an observed pain quote. Their notes say so explicitly. Do not let outreach copy assume a pain they have not voiced. Also flagged in his record: Sema4.ai is itself an agent-management platform, so Tommi Holmgren may be a competitor/partner rather than a prospect.
RUN SUMMARY — ICP prospect signal scanner, 2026-08-29.
ADDED (5 new people, all deduped against the ~1,007 already in the People Library):
1. Reshadat Ali — Director of Engineering, AI Platform & GenAI, ArmorCode (~222 emp, Series B) — High
2. Ashwin Kulkarni — Director, AI & Engineering, Demandbase (~750–1,129 emp) — High
3. SN Raju Kattari — Director of Engineering, Agentic Platform & ML Frameworks, Calix (~1,921 emp) — Medium
4. Sreenivasulu Anantha — Head of AI, SunTec Business Solutions (~560–875 emp) — Medium
5. Dimitrios Chavouzis — Head of AI Engagements (Agents, Evals & RL Envs), Pareto AI (~540–573 emp) — Medium
SIGNAL BUCKET PRODUCTIVITY — this run inverted the usual pattern:
- Bucket 1 (ICP writing about agent cost) — effectively EMPTY. All five prescribed post searches returned recruiters, course-sellers and LinkedIn creators, not ICP authors. "agent cost LLM production", "AI agent reliability production", "token budget AI agents scaling", "agent observability cost per run", "LLM spend agents production" all failed the same way.
- Bucket 2 (ICP commenting on non-ICP content) — low yield. Comment threads on the cost/observability posts were dominated by consultants and vendor founders.
- Bucket 3 (competitor content) — one usable thread. Ishaan Jaffer's (CTO, LiteLLM) Bedrock/AgentCore post (535 reactions, 17 comments) surfaced two CTO commenters, but both companies fell out on size (Archdesk 36 emp; Ethara.AI unverifiable). IMPORTANT MECHANICAL NOTE: LinkedIn content search does NOT support OR syntax — "helicone OR portkey OR litellm" returned unrelated medical content. Single-vendor queries only.
- Bucket 4 (ICP writing about agents) — MOST PRODUCTIVE by far, but reached through PEOPLE search, not post search. All 5 adds came from LinkedIn people search on title + agent keywords, then verifying company, headcount and a real post/repost on each profile.
METHOD CHANGE WORTH KEEPING: for future runs, lead with people search ("VP of Engineering"/"Director of Engineering"/"Head of AI" + agents/evals/observability/cost), then confirm signal on the profile. Post search alone is now mostly creator noise. Also note US-geo people searches skewed to companies over 2,000 employees (Splunk, Plaid, Figma, Remitly, Microsoft, NatWest, Swiggy, Airtel) — the 50–2,000 band sat in the India/EU results.
DISQUALIFIED THIS RUN (recorded so we don't re-spend on them): Michal Piszczek (CTO, Archdesk — 36 emp), Nikesh Masiwal (Head of Eng, EAMOT — ~20 emp), Nikesh Goel (NatWest, ~60k), Vivek V. (Swiggy), Vidya Pandey (AmEx), Dr. Paul Elvers (FUNKE Mediengruppe), Gerson Rodriguez (Oracle), Vimal Muralidharan (Duck Creek — borderline size, no real signal), Deepak Samar (ConcertAI — no posts), Rafael Zerbini (Blox — headcount unverifiable, so not added per the "must confirm 50–2,000" rule), Pratham Sharma (SOAIS — IT services, not agent-shipping SaaS).
DUPLICATES CAUGHT: Christophe Pierret (SoundHound), Saurabh Hebbalkar (Krista), Srikanth Konjeti (Gnani.ai), Kaushik Vatsa (Mantra/Mikshi) — all already in the library from earlier runs. The library is now saturated on the obvious AI-native names; future runs need less-obvious surfaces (agentic platform teams inside 500–2,000-person vertical SaaS, which is exactly where 4 of today's 5 came from).
HIGH-PRIORITY: Reshadat Ali (ArmorCode) — 222-person Series B with a named agent product (Anya Agents) already GA and a Director who owns the AI platform underneath it. Closest thing to a textbook Alpha buyer found this run.
EMERGING PATTERN FOR OUTREACH COPY (see the two VOC entries filed today):
(a) Practitioners now separate "cost per token" from "cost predictability" and say the second is what actually blocks enterprise sales — retry/loop storms, not unit price. One quoted incident: an agent stuck in a retry loop burned $4,200 in 63 hours.
(b) Repeated complaint that observability alerts arrive after the spend. The phrase used was "acting on a request before the spend happens" — control at dispatch, not a dashboard afterwards. That is Alpha's budget-per-agent story almost verbatim and should be tested as a subject line.
(c) Reliability is described as silent and per-run ("same prompt, different answer"; "silently gets results for half of it"), which argues for run-level evidence as the demo artifact rather than an aggregate dashboard.
sales intelclaude-connector (scheduled task: icp-prospect-signal-scanner) · 28 Aug 2026
ICP Prospect Signal Scanner — run 2026-08-28: 1 person added, LinkedIn surface low-yield (target 5 missed)
RESULT: 1 new person added (target was 5). Not padded — no fabricated profiles, sizes, or quotes.
ADDED
- Roman Ignatov — R&D Director, Similarweb (~1,000-1,200 employees), also VP Tech at XPLN. Signal bucket 2. ICP confidence: Medium. Pain: step-level agent alerting drowns on-call; wants run-level telemetry (task-completion rate, p95 steps-per-run); open question on what fraction of traces to keep at full fidelity as prompt/response capture grows. Caveat: Similarweb is public (NYSE: SMWB), not Series A-C.
BUCKET PRODUCTIVITY THIS RUN
- Bucket 1 (ICP writing about agent cost/control): 0. Ran 6 post searches — "agent cost LLM production", "AI agent reliability production", "agent observability cost per run", "LLM spend agents production", "agents in production token spend blowout", "our agents production cost CTO". Authors were near-uniformly content creators, solo consultants, boutique dev shops (<50 emp), or ICs. No Director+ at a 50-2,000 employee agent company authored anything in range.
- Bucket 2 (ICP commenting on non-ICP agent content): 1 — the only productive bucket. Read full comment threads on two high-engagement posts (Vishakha Sharma, 29 comments; Ganapathi Ramkumar Palanivelu, 24 comments / 315 reactions). Of ~55 commenters read, exactly one was ICP-shaped. The rest: solution architects, SDETs, AI red-teamers, founders of sub-10-person consultancies, and engagement-farming commenters.
- Bucket 3 (competitor content): 0. Searched "langfuse OR langsmith OR braintrust", "helicone OR portkey OR litellm", "aws agentcore OR openrouter agents". Every author was an IC AI engineer, SRE, or newsletter writer explaining the tools — nobody senior evaluating them.
- Bucket 4 (ICP writing about agents): 0. Searched "building AI agents production 2026", "AI agent platform engineering scale", "agent unit economics production LLM", "scaling AI agents context window cost", "agent sprawl cost per resolution". Nothing above IC/architect level.
- Web/Reddit fallback: 0 named individuals. Searches surfaced only SEO content-marketing blogs (Atlan, Digital Applied, Teamvoy, Cockroach Labs) and aggregator summaries of Reddit threads — no attributable person with a title and company.
WHY THE YIELD WAS LOW (matters for next run)
1. LinkedIn content search is scoped heavily to the account's own network. This account's 1st/2nd-degree graph skews to India-based solution architects, content creators, and IC AI engineers — not US/EU Series A-C agent-company engineering leadership. Post search will keep returning the same population regardless of keyword.
2. The keywords in the skill are the vocabulary of people who write ABOUT agent ops for reach, not the vocabulary of people who RUN agents. Operators post rarely and in different language.
3. People search returns ICP titles but strips company and carries no pain signal, so results can't be qualified without a profile visit each — expensive and still signal-free.
RECOMMENDED CHANGES TO THE SCANNER
- Stop relying on LinkedIn post search as the primary surface. Shift primary sourcing to comment threads on high-engagement agent-ops posts (the one bucket that worked) and to the reactor/commenter lists of posts by agent-infra companies themselves.
- Add non-LinkedIn primary surfaces where operators actually write with attribution: company engineering blogs, conference talk abstracts (AI Engineer Summit, KubeCon, QCon), podcast guest lists, and GitHub issue threads on agent frameworks — all give name + title + company + a real technical complaint.
- Consider seeding from Vishnu's own network growth: people who engage with thealpha.ai content are pre-qualified on interest.
- Consider whether the "Series A through Series C" constraint should stay hard. Roman Ignatov fits every other criterion and was excluded from High confidence solely on funding stage; public mid-caps at 1-2k employees with agent programs may be a legitimate adjacent segment worth a decision.
OUTREACH COPY IMPLICATION
The single strongest phrasing observed this run, from an actual engineering leader rather than a commentator: the agent RUN is the unit that matters, not the step or the token. "Step-level alerts drown the on-call" is his language, not ours. Cost-per-run and steps-per-run framing will land better than token-spend framing with this persona.
VOC: 1 insight logged (run-level observability + trace-retention cost). Single-source, so recorded as directional and flagged for confirmation next run rather than as a confirmed repeating pattern.
CONSTRAINTS OBSERVED: no outreach, no duplicates (checked against existing ~1,000-person library), no Aptos Retail contacts, nothing fabricated.
agent reviewdaily-review-agent · 28 Aug 2026
Validation flag: the "1→5 agent scale wall" is stale — the market moved to fleets, which vindicates the 20-agent floor and dates the ICP canon
WHAT I CHECKED (2026-08-28): the buying-trigger claim that runs through the whole Brain — Question #4 ("the buying trigger is the 1→5 agent scale wall"), Question #3 ("teams stalled at the 1→5 agent scale wall"), Research Brief #4's "~60% of enterprises stall exactly here," and the ICP pillar definition. Task #95 names the resulting contradiction as the sharpest open question in the strategy layer: canon says the buyer breaks at 5 agents, Decision #403 says they must have 20.
THE CANON IS THE STALE HALF, NOT THE PRICE. Current 2026 deployment data:
- 52% of executives report agents deployed in production, and 39% report MORE THAN TEN agents running (KPMG Q2 2026).
- Mean fleet size was ~37 agents per organisation in Dec 2025; by April 2026 nearly 38% of organisations reported MORE THAN 100 agents deployed, with active-deployer cohorts running 76–100 and roughly doubling per quarter.
- The pilot-failure stat the Brain leans on is intact and unchanged (~88% of agent pilots never reach production; top blockers are evaluation gaps 64%, governance friction 57%, model reliability 51%). So the wall is real — it just is not at five.
WHY THIS MATTERS, PRECISELY. Three things follow, and none of them require re-opening the pricing decision:
1. DECIDE #95 AT 20+, NOT AT 5. The trigger call in Task #95 resolves cleanly in favour of Decision #403. The 20-agent minimum is not an aggressive floor invented to make founder-led-sales arithmetic work — it now sits BELOW the mean fleet of an active deployer. Do not quietly lower the floor to fit the old canon; rewrite the canon.
2. THE PAIN CHANGED SHAPE, AND THE BRAIN'S OWN FIELD DATA ALREADY SAID SO. The pain is no longer "we got to five agents and stalled." It is "we are running dozens and cannot attribute, evaluate or govern any of them." Every recent scanner run says this in the ICP's own words — "our agent fleet," "human oversight doesn't scale past ~10 agents," "tooling is optimized for building agents, not operating them," "what did that cost us last month? I didn't have an answer." Two independent sources, internal and external, pointing the same way.
3. IT RE-DATES THE OUTREACH COPY, NOT JUST THE THESIS. A DM written to a team stuck at five agents is addressed to a company that has already moved past that problem — or was never in it. #93 should open on the 1→many wall: not "you'll blow up when you scale," but "you already have a fleet and nobody can say what any single agent cost or did."
WHAT DOES NOT CHANGE: the entry emotion (the surprise invoice, first production incident) still holds as a trigger event; only the agent-count threshold is wrong. Enterprise/procurement caveats from Decision #29 still stand. This flag does not touch mission or thesis — it supplies the missing input for the rewrite Task #95 already owns.
Evidence:
https://www.digitalapplied.com/blog/ai-agent-adoption-2026-enterprise-data-points
https://prefactor.tech/learn/ai-agent-adoption-statistics
https://www.digitalcommerce360.com/2026/02/11/survey-enterprises-ai-agents-crewai-report/
https://aibusinessweekly.net/p/ai-agents-statistics
https://learn.g2.com/enterprise-ai-agents-report
Internal: Task #95, Decision #403, Questions #3 and #4, Research Brief #4, ICP pillar definition, scanner entries #412/#414/#417/#418, VOC patterns filed 8/26–8/28.
SECOND, SMALLER CHECK FILED HERE RATHER THAN SEPARATELY — the $30K price point is market-consistent. Published 2026 guidance puts enterprise AI-agent platform annual contracts starting around $30,000 and running to $100K–$300K+, with a base-fee-plus-usage hybrid the common structure. Decision #403's $30K base plus per-agent blocks matches the prevailing shape. No action; recorded so the price stops being re-litigated in reviews.
https://thecrunch.io/ai-agents-price/
https://www.braincuber.com/blog/ai-agents-pricing-guide-what-does-it-actually-cost
sales intelclaude-connector · 27 Aug 2026
ICP Prospect Scanner run — 2026-08-27 (3 added, target of 5 MISSED; LinkedIn content search is exhausted)
RESULT: 3 new people added, below the 5-per-run minimum. Reporting the shortfall rather than padding it — the two extra candidates I could have added had no verifiable pain signal, and inventing pain points would poison the outreach data.
ADDED
1. Manick Bhan — Founder/CEO/CTO, Search Atlas (~75–250 emp) — Signal 4 — ICP confidence High. Technical founder operating agents in production daily; publicly pushes back on "AI agents are a bust" headlines.
2. Munil Shah — CPTO, Talkdesk (~1,350 emp) — Signal 4 (with Signal 2 crossover) — High on role/activity. States the market's biggest gap is evaluating agent behaviour, reliability and tool usage at scale.
3. Matthew Payne — VP Engineering & Head of AI R&D, Domo (~900–1,000 emp, UNVERIFIED) — Signal 4 by title only — Medium. Fit-based lead, no expressed pain. Requires discovery; verify headcount first.
SIGNAL BUCKET PRODUCTIVITY
- Signal 1 (ICP writing about agent cost/control): effectively DRY. Nine distinct content queries run (agent cost LLM production; AI agent reliability production; token budget AI agents scaling; agent observability cost per run; LLM spend/cost optimization; agents in production; context engineering agents; AI agents cost control observability; agent cost sorted-by-latest). Zero ICP-matching authors. Results were dominated by recruiters, personal-brand accounts, consultancies and enterprise-services staff (Cognizant, Deloitte, NatWest, Aristocrat) — all outside the size band or below Director.
- Signal 2 (ICP engaging non-ICP content): produced the single best quote of the run (Gaurav Narasimhan on Andrew Ng's post), but comment-mining at scale was low-yield — Andrew Ng's 400+ comment threads are a broad practitioner/fan audience, not ICP-dense.
- Signal 3 (competitor content): DRY. "helicone OR portkey OR litellm" returned medical/oncology content — LinkedIn content search does NOT support OR syntax and silently mangles the query. Individual searches for langfuse and AgentCore returned only company pages and junior engineers. Recommend dropping the OR-syntax queries from the skill file.
- Signal 4 (ICP writing about shipping agents): produced 2 of the 3 adds, both via colleague-activity traversal rather than search.
METHOD CHANGE THAT WORKED — recommend adopting
LinkedIn CONTENT search is exhausted for this account; PEOPLE search is not. People search reliably returns ICP titles + companies and paginates 10+ pages. The productive loop this run was: people search for ICP titles → open the candidate's /recent-activity/all/ → capture genuine expressed signal there → verify headcount via web search → grep the brain for duplicates. Two of three adds came from traversing a confirmed ICP person's activity feed into their colleagues and the accounts they engage with. Suggest rewriting the skill around this loop.
SATURATION WARNING — the main constraint now
The brain's people library is heavily saturated against these exact queries. Of the five strong candidates I qualified, THREE were already present: Prasad Kavuri (Zip), Gaurav Narasimhan (Search Atlas), Yunjing Ma (Talkdesk). Also already present: Ershad Ali Mohammad (Kore.ai), Deepesh Tated, Anubhav Sharma (Jeeva AI), Akarsh Mishra (TrueFan), Alex Lunev (LangChain). Talkdesk was already opened as an account in the 2026-07-30 run. Repeating these queries will keep returning known people. Future runs need genuinely new seams — e.g. company-first prospecting into Series A–C AI-native firms not yet in the library, conference/podcast speaker lists, GitHub/Discord agent-framework contributors, or job postings for agent-platform roles (a hiring signal that a team is scaling 1→5+ agents).
CANDIDATES REJECTED AND WHY (so future runs don't re-spend effort)
- Nikhil Goel (VP, Global Head AI Native Engineering) — posted a clean token-optimization signal, but discloses no company anywhere on his profile, so the 50–2,000 gate cannot be verified. Rejected.
- Kiran Chikkanna (VP Eng, Saarthi.ai, ~55–65 emp) — borderline size, and one 3-year-old repost as his entire activity. No signal. Rejected.
- Sattyam Jain (Attri.ai) — excellent agent-spend post ($150.95 unguarded run vs $0.40 with a gate), but Tech Lead/Architect is below the Director floor and Attri.ai is likely under 50 emp. Rejected on both gates despite being the best cost content found all run.
- Too large: Freshworks, Thomson Reuters, Microsoft, AWS, SAP, Zillow, ServiceNow, DIRECTV, Wayfair, Vantor, Deloitte, Cognizant.
EMERGING PATTERN THAT SHOULD INFLUENCE OUTREACH COPY
All three reliability quotes captured this run frame the problem as BEHAVIOUR and EVALUATION, not as SPEND. Munil Shah: no rigorous way to evaluate behaviour/reliability/tool usage at scale. Gaurav Narasimhan: agent unpredictability is hard to explain to deterministic-minded engineers. Manick Bhan: defending agent reliability from the operator side. Meanwhile the loudest cost-framed content came from vendors and consultants, not from buyers. Hypothesis worth testing: for senior ICP leaders, reliability/explainability is the felt pain and cost is the consequence they discover later — so a reliability-led hook may out-open a cost-led hook. See the two VOC entries filed this run.
sales intelclaude-connector · 22 Aug 2026
ICP Prospect Signal Scanner — run 2026-08-23 (LinkedIn blocked; 6 added from conference source)
RUN BLOCKER — READ FIRST
The Claude in Chrome extension was not connected for the entire run. Every one of the 13 prescribed LinkedIn post/people searches across all 4 signal buckets was impossible to execute. No LinkedIn data was collected. No LinkedIn profile URLs were captured for any person below — the profile_url field is deliberately empty on all 6 records rather than guessed. To restore the intended workflow, install/sign in to the Chrome extension (https://chromewebstore.google.com/detail/fcoeoabgfenejglbffodgkkbkcdhcgfn) and re-run.
WHAT WAS DONE INSTEAD
Fell back to public web sources. Generic web search for agent-cost/reliability content returned almost entirely SEO content-farm material with no named individuals — not usable. The one productive source was the AI Engineer World's Fair 2026 public schedule (https://www.ai.engineer/worldsfair/schedule), which lists speakers as "Talk Title — Name, Company, Title". Extracted ~120 speakers, filtered to Director-level-and-above technical/AI/eng/product-AI leaders, excluded mega-corps, deduped against the existing people list, then verified headcount for each survivor via Tracxn/Crustdata/PitchBook/Revelio before adding.
PEOPLE ADDED — 6 (target was 5)
1. Chaitanya Asawa — Head of Engineering, Clinical Decision Support — Abridge (~635 emp) — HIGH confidence. Best fit of the run.
2. Everett Berry — Head of GTM Engineering — Clay (~1,167 emp, Series C $100M @ $3.1B) — Medium. Only person found who fits BOTH the headcount band and the Series A–C window.
3. Eli Cohen — Director of Technology Incubation — Snyk (~1,870 emp) — Medium.
4. Giedrius Steimantas — Director of Scraping Engineering — Oxylabs (~482 emp) — Medium.
5. Suchet Bargoti — Director of Inspection and Mapping — Skydio (~1,000 emp) — Medium.
6. Dru Knox — Head of Product — Tessl (59 emp, Series A) — Medium. Strongest thesis alignment: his talk is literally titled "Harness Engineering".
SIGNAL BUCKET PRODUCTIVITY
Buckets 1, 2 and 3 produced ZERO — all three depend on reading LinkedIn post authors and comment threads, which was unavailable. Bucket 4 (ICP publicly talking about building/shipping agents) produced all 6, via conference talks rather than LinkedIn posts. Treat bucket productivity data from this run as invalid for trend purposes.
HIGH-PRIORITY FLAGS
- Abridge (Chaitanya Asawa) is the cleanest target: 635-person AI-native company, named agent product line, engineering leader who owns it.
- Tessl is using "harness" as their public vocabulary. Worth watching as both a prospect and a potential positioning competitor.
- Clay is the only true Series A–C + in-band-headcount match found. The Series A–C constraint is doing a lot of filtering work — most agent-shipping companies at 500–2,000 employees are Series D+ or public. Worth revisiting whether stage or headcount is the real qualifier.
SOURCE EXHAUSTION WARNING
A previous run already scraped this same conference: 11 of the ~17 strongest candidates (Nicholas Arcolano/Jellyfish, Mingsheng Hong/Ironclad, Vivek Muppalla/Hippocratic AI, Rashi Agrawal/Hinge Health, Denys Linkov/Wisedocs, Archana Kamath/DigitalOcean, Saul Howard and Anuj Iravane/Anterior, Dan Feng/Maven Clinic, Walden Yan/Cognition, Gil Feig/Merge) were already in the people library. This source is now largely mined out. Also note the schedule page only rendered Days 3 and 4 — Days 1 and 2 (including the CTO Circle and Data Quality tracks) were collapsed and never loaded, so roughly half the conference remains unexamined. The structured feeds at /worldsfair/2026/speakers.json and /sessions.json would cover the gap but were blocked by the web_fetch provenance rule; they are reachable via the Chrome extension once it is connected.
PATTERNS THAT SHOULD SHAPE OUTREACH COPY
Two VOC entries logged this run. Short version: the market has started naming the missing layer out loud (3 of 6), and senior leaders narrate the 1 -> many agents transition as a win rather than a struggle (2 of 6) — so cold outreach should ask what broke during that transition rather than assert that something did.
DATA INTEGRITY NOTE
No LinkedIn URLs, headcounts, or quotes were fabricated. Every pain point on the 6 records is explicitly labelled as inferred from a verbatim talk title rather than stated by the person. All 6 need contact enrichment before any outreach.
sales intelICP Prospect Signal Scanner (scheduled) · 12 Aug 2026
ICP Signal Scanner run 2026-08-12 — 6 new prospects added (IDs 802–807); Synera/Adonis/WisdomAI/Model ML
RUN SUMMARY — 2026-08-12 (autonomous scheduled run)
PEOPLE ADDED: 6 net-new, all Medium/High ICP confidence, all deduped against the full ~797-person People Library and screened for the 50–2,000-employee floor + "actively shipping agents in production."
1. Moritz Maier — Co-founder & CEO, Synera (~102 emp, Series B, agentic AI for industrial engineering; NASA/Airbus/BMW) — HIGH.
2. Daniel Siegel — Co-founder & CPO, Synera — MEDIUM (AI-focused product persona).
3. Akash Magoon — Co-founder & CEO (ex-CTO, Nayya), Adonis (~80 emp, Series C, healthcare RCM AI agents) — HIGH.
4. Aman Magoon — Co-founder & CPO, Adonis — MEDIUM.
5. Sharvanath Pathak — Co-founder & CTO, WisdomAI (~107 emp, Series A, agentic data-insights platform; Kleiner/NVIDIA) — HIGH.
6. Arnie Englander — Co-founder (Tech & Product), Model ML (~80 emp, finance/IB research agents) — MEDIUM.
Target of ≥5 met (6 added).
METHOD + CONSTRAINT: Claude-in-Chrome / LinkedIn was NOT connected again this run (tabs_context_mcp failed twice; consistent with recent prior runs), so the prescribed LinkedIn post/people/comment searches (all 4 signal buckets) could not run. Pivoted to web research + primary-source verification: 2026 funding announcements, the Agent Conference "Agentic List 2026" (Top 120 agentic companies), Tracxn/YC headcounts, SiliconANGLE/Bloomberg/PRNewswire. All 6 map to Signal 4 (ICP technical leaders building/shipping agents), several also Signal 1 (public founder POV). No profiles, headcounts, or quotes were fabricated; profile_url left blank where a LinkedIn URL could not be confirmed.
MOST PRODUCTIVE ANGLE this run: cross-referencing the Agentic List 2026's mid-stage ($30–200M) cohort against the library, then verifying a technical leader + in-band headcount for the companies NOT already catalogued. This surfaced 4 clean net-new companies (Synera, Adonis, WisdomAI, Model ML). Confirmed the library is heavily saturated: of ~110 listed agent companies, the large majority were already present (Decagon, Sierra, Glean, Harvey, Hebbia, Tennr, Hippocratic, Cognition, Factory, Parloa, etc.). Parloa in particular is now saturated down to VP/Director level (Stefan Ostwald CTO already in library) — dropped to avoid a dup.
HIGH-PRIORITY FLAGS:
- Synera (Moritz Maier): shipping teams of autonomous agents into 60+ industrial enterprises on-prem; explicitly sells the pilot→production leap. Strong wedge fit.
- Adonis (Akash Magoon): autonomous agents progressing healthcare claims in a regulated, audit-heavy domain — reliability + per-run cost accountability.
- WisdomAI (Sharvanath Pathak): CTO, agents taking autonomous action over large distributed enterprise data — token/cost + reliability pain.
VOC (logged 2 insights, IDs 233–234):
1. Pilot→production reliability gap for autonomous agents in regulated/high-stakes verticals (all 4 companies) — lead outreach with production reliability + governance/audit.
2. Cost-per-run / token blowout + no per-run cost visibility as agents scale (all 4) — CFO-facing second hook; thealpha's operating-layer wedge (per-run cost visibility + control).
SKIPPED ON RULES (documented so future runs don't re-chase): Axle (~40 emp, below floor); Grace Investment Machine / GIM (team size unverifiable + executes live trades); observability/safety vendors on the Agentic List (Arize, Fiddler, Credo, Mindgard, Raindrop, Vijil) excluded as competitor-adjacent, not ICP buyers; Parloa (already saturated in library).
RECURRING BLOCKER: Claude-in-Chrome extension has been disconnected for multiple consecutive runs. Reconnecting it (Chrome side panel, sign in with same account) would unlock the engagement-based buckets (Signals 2 & 3 — ICP commenters on competitor/influencer posts), which remain the best untapped source of net-new NON-founder leaders (VP Eng / Head of AI / Director of AI) that web research cannot easily surface. Recommend also adding an Apollo/Clay-style enrichment MCP as a graceful non-LinkedIn fallback.
sales intelclaude-connector · 4 Aug 2026
ICP Prospect Signal Scanner — run 2026-08-04
Added 5 NEW ICP-matching people (library 655 -> 660; IDs 660-664):
1. Brian Reale — Founder, ProcessMaker (~150-250 emp) — Signal 1 — Medium-High. Only find with an AUTHORED agent-cost post this run; warmest outreach angle.
2. Scott Kurinskas — VP Product, C3 Agentic AI Platform / C3.ai (~900) — Signal 4 — Medium.
3. Warren Van Winckel — Director of Eng, AI Platform, Upwork (~1,500) — Signal 4 — Medium.
4. Aurélien Gervasi — Head of Eng, AI, Back Market (~700) — Signal 4 — Medium (agent maturity unverified).
5. Venkatesh Yadav — VP AI Applications & Delivery, H2O.ai (~250-400) — Signal 4 — Medium.
WHAT WORKED: LinkedIn People Search filtered by ICP titles (Head of AI / VP AI / Director of Eng / CTO) was by far the most productive bucket. LinkedIn CONTENT search (Signals 1-3 as post searches) was very low-yield this run — results were dominated by consultants, AI educators/influencers, recruiters, and staff at OVER-SIZE companies (Deloitte, Cisco, Airtel, Microsoft, Salesforce, ServiceNow, Capgemini) which fail the 50-2000 filter.
HIGH DUPLICATE RATE (prior runs have saturated obvious AI-native agent cos): already-in-library strong matches hit this run incl. Anubhav Sharma (Jeeva AI), Varun Kacholia (Eightfold AI), Toshish Jawale (Invoca), Shomron Jacob (Iterate.ai), Jeegar Shah (Atomicwork), Harshil Shah (RSI). Recommendation: next runs should target LESS-obvious ICP pools — vertical AI apps (legal/health/fintech agents), non-US/EU AI-native startups, and mid-market marketplaces/SaaS shipping internal agents — rather than re-scanning the well-known agent startups.
OUTREACH COPY SIGNAL (see VOC this run): lead with the 'hidden/true cost of agents in production — no cost-per-run visibility beyond tokens' message; it recurs across ICP-authored and adjacent content. Maps directly to Alpha's cost/observability wedge.
Constraints honored: no outreach; no Aptos Retail contacts; only Medium/High confidence added; pain points for people-search finds are labeled role-inferred (not fabricated quotes).
sales intelicp-prospect-signal-scanner · 2 Aug 2026
ICP Signal Scanner run — 2026-08-02: +5 new (Writer/Parloa), 6 dup rows to purge
ICP Prospect Signal Scanner — run 2026-08-02.
RESULT: 5 net-new ICP people added (all Medium/High confidence, deduped against the existing 582-person library):
1. Mo Shaker — Sr. Director of Engineering, Agentic AI @ Writer (High).
2. Manhal D. — Director, AI Engineering @ Writer (Medium; LinkedIn privacy-truncates the surname, profile URL recorded for disambiguation).
3. Brock Imel — Head of Evaluation & Alignment / QA (Applied AI) @ Writer (Medium).
4. Marco Badorrek — Director, Solution Engineering EMEA @ Parloa (Medium; customer-facing technical director).
5. James Allen — Director, Solutions Architecture EMEA @ Writer (Medium; customer-facing technical director).
MOST PRODUCTIVE APPROACH: LinkedIn company People-directory pages for clean, independent, Series-A-to-C agent companies (Writer 201-500 & Parloa 201-500) — far higher signal than the 4 content-search signal buckets, which surfaced mostly consultants, students, and media pages with almost no ICP-matching authors, and were heavily biased toward the logged-in account's own network. Generic title/keyword people-searches surfaced large enterprises (Wipro, Infosys, Qualcomm, Salesforce, State Street) that are out of ICP size.
HIGH-PRIORITY COMPANY: Writer (enterprise agentic AI, Series C) — dense, well-connected bench of agent-engineering leaders; strongest single source this run.
COVERAGE NOTE: The library is now very saturated — ~589 people across ~351 companies. Nearly every notable 50-2,000-emp agent company (Sierra, Decagon, Cresta, Cognigy, Observe.AI, Jeeva, Parloa, Writer, etc.) is already mined, and most senior leaders surfaced were either already recorded or IC-level. Future runs should (a) target NEW companies outside the current 351 and/or (b) mine additional un-recorded leaders at large covered companies. Recommend maintaining a canonical company/slug list to avoid slug-guessing dead-ends.
DATA-HYGIENE FLAG (needs manual cleanup): Due to a mid-run bug (a dedup check ran against an empty in-page variable after a cross-site navigation), 6 DUPLICATE person rows were created before the check was fixed to fetch the live list inline. Please delete these duplicate rows (there is no delete_person tool exposed via MCP, so I could not remove them autonomously): id 586 (Anubhav Sharma, dup of 382), 587 (Masashi Beheim, dup of 433), 588 (Sybille Fuks, dup of 536), 589 (Moritz Kroger, dup of 466), 591 (Dan Bikel, dup of 176), 592 (Muayad Sayed Ali, dup of 463). All six originals from prior runs remain intact.
EMERGING PATTERN for outreach copy: two pains recur across agent-eng leaders at 200-1K-emp agent companies — (1) no per-run cost visibility as agents scale pilot->enterprise, and (2) production reliability + eval/alignment gaps. Lead with cost-per-run observability + reliability guardrails.
sales intelicp-prospect-signal-scanner · 2 Aug 2026
ICP Prospect Signal Scan — 2026-08-02 (5 new agentic-AI leaders added)
Run 2026-08-02. Added 5 NEW ICP prospects — all LinkedIn-verified at 50–2,000 employees, actively shipping AI agents, and deduped against the 566 existing people:
1. Ophir Samson — Head of Voice AI, Greenhouse Software (501–1,000) — High. Actively posting on production voice agents.
2. Vipul Lonkar — Sr Director of Engineering (Platform), Avaamo (201–500) — Medium-High. Owns multi-agent/voice/RAG platform.
3. Mustafa Ahamed — VP Product Management, Aisera (201–500) — Medium.
4. Akshay Pushparaja — Director of Eng (Generative AI Products), C3 AI (~1,060) — Medium.
5. Ben Grosser — Head of Insurance AI Product, FurtherAI (~57) — Medium.
Most productive bucket: Signal 4 (ICP leaders building/shipping agents), surfaced via LinkedIn people-search ('Head of Agentic AI', 'Head of AI voice agents', 'CTO co-founder building AI agents') and company-scoped searches at agent-native scaleups. Content-search buckets 1–3 (agent cost / reliability / competitor keywords like Langfuse/Helicone) returned mostly influencers, vendors, and consultants this run — low ICP yield; comment-mining was blocked by LinkedIn lazy-loading.
Saturation note: the brain already covers the obvious 'Head of Agentic AI' cohort. Four strong candidates found this run — Anubhav Sharma (Jeeva AI), Venkat Peri (Advisor360), Toshish Jawale (Invoca), Vedavyas Panneershelvam (Phaidra) — were ALL already recorded, as were every technical leader surfaced at PolyAI, Kore.ai, Forethought, and Haptik. New yield came only from under-covered companies: Greenhouse, Aisera, C3 AI, FurtherAI, Avaamo. Recommendation for future runs: prioritize under-covered agent-native scaleups and non-obvious titles (VP Product-AI, Sr Director of Eng) over the exhausted 'Head of Agentic AI' query.
Emerging pattern (see VOC #174): ICP leaders increasingly frame the bottleneck as reliability/verification of agent output at scale, not model capability. Outreach copy should lead with cost + reliability visibility per agent run.
High-priority for outreach: Ophir Samson (Greenhouse) and Vipul Lonkar (Avaamo).
sales intelclaude-connector · 1 Aug 2026
ICP Prospect Signal Scanner — Run 2026-08-02
5 new ICP people added this run (brain 561 -> 570): 1) Jayanth Madheswaran — Founder & CEO (technical), Eve (eve.legal), legal AI agents (#565, Medium-High). 2) Oleg Zaremba — CTO & Co-founder, AiSDR, autonomous sales agents / YC (#566, High). 3) Hari Poludasu — VP Engineering, Kore.ai, agentic AI platform (#567, High; distinct from existing Kore.ai contacts). 4) Andrei Negrau — CEO & Co-founder, Siena AI, autonomous CX agents for commerce (Medium). 5) Lisa Popovici — Co-founder, Siena AI (Medium).
Method / what worked: The brain is highly saturated (345 companies, 561 people), so the classic post-author signal buckets (Signal 1-4) were LOW-YIELD this run — LinkedIn content search for agent-cost/reliability keywords surfaced mostly consultants, LLMOps architects and vendor pages, not agent-native ICP leaders. The effective method was targeted people-search for technical leaders at agent-native companies ABSENT from the brain, plus additional senior eng leaders at already-covered platforms.
Dedupe catches (do NOT re-add — already in brain): Eno Reyes (Factory, CTO), Souvik Sen (Ema, CTO), Edward Wu (Dropzone AI, CEO), Stanislas Polu (Dust, CTO). Several strong candidates were already present; verify against read_brain before adding — memory-based candidates are unreliable (e.g., Sami Ghoche is now VP at Zendesk, not Forethought).
High-priority for outreach: Siena AI (net-new agent-native company, 2 founders captured) and AiSDR (YC, CTO co-founder).
Outreach copy angle (from VOC this run): lead with 'per-run agent cost visibility + reliability — the harness matters more than the model.'
Next-run suggestions: vary keywords toward agent-native SUB-segments still thin in the brain (voice-agent infra evals, SOC/security agents, vertical legal/healthcare agents); mine COMMENTS on competitor posts (Helicone/Langfuse/LangSmith/Braintrust/AgentCore) for ICP engagers; and target Director/VP-of-AI titles at covered companies where only the founder is listed.
sales intelicp-prospect-signal-scanner · 1 Aug 2026
ICP Prospect Signal Scanner — run 2026-08-01
Run summary (automated ICP prospect signal scanner).
Added 5 net-new ICP people (all Director-to-VP level, technical, at validated in-range agent-native companies):
1. Christian Burgas — Head of Technology Platform, Parloa (id 543)
2. Matt Lowe — Director of Engineering, EliseAI (id 544)
3. Ken Koch — Director of Engineering, EliseAI (id 545)
4. Shanil Puri — Director, Speech Technologies, Hippocratic AI (id 546)
5. Himanshu Walia — Director, Integrations, Hippocratic AI (id 547; Medium confidence — integrations-focused)
Method / which signals were productive:
- Signal 1 (ICP authoring about agent cost/reliability) and Signals 2-3 (competitor/observability post comments) were LOW YIELD this run: LinkedIn content search for generic agent-cost keywords surfaced mostly consultants, IC engineers, thought-leaders, and (for competitor tool names) spam/insurance-agent noise. Few authors were Director+ at 50-2,000-emp agent companies, and the notable ones were already in the brain.
- Signal 4 (ICP technical leaders at agent-native companies) was the productive path. Most valuable technique: cross-reference the brain’s existing 340 target companies (validated in-range agent companies) against LinkedIn People search to find UNCAPTURED Director/VP/Head-level leaders at those same companies (Parloa, EliseAI, Hippocratic AI).
Dedup note: The brain is already very comprehensive (~531 people / 340 companies). Many obvious candidates were already present, including the well-known founders/CTOs and even several VPs/Directors at Parloa, EliseAI, Cresta, and Hippocratic AI (e.g., Anubhav Sharma/Jeeva AI, Masashi Beheim & Moritz Kroger/Parloa, Ryan St Pierre, Zac Gottschall & Mario Martone/EliseAI, Jove Zhong, Ashish Agrawal & Xiangru Chen/Cresta, Sri Subramaniam/Hippocratic AI were all already in the brain and were skipped). The 5 added are confirmed net-new.
High-priority observations for outreach copy (see 2 VOC entries added this run):
- Token/context waste framed as an ARCHITECTURE problem ("vector stores are not memory"; tiered agent memory; "agents burning tokens"). Lead with cost-per-run visibility + context efficiency.
- Reliability of non-deterministic agents in production (infinite/loop costs, hallucination-as-architecture, deterministic enterprise reliability; ">95% of code AI-written"). Lead with production observability + eval/reliability.
Data-integrity notes: Pain points on the 5 added people are role/company-contextual (sourced by ICP title match at validated target companies, Signal 4), not individual quoted posts — explicitly labeled as such in each person’s notes. VOC entries are paraphrased syntheses across real observed posts (sources cited), not fabricated verbatim quotes. Company sizes are estimates; all 5 companies are already validated as in-range in the brain.
Env note: Alpha Brain MCP tools were not natively connected this run; writes were made via the brain’s /api/mcp JSON-RPC endpoint through the authenticated browser.
Run 2026-07-31. Added 5 NEW ICP-matching people (People Library 502 -> 507), all deduped against existing 502 and headcount-verified in-band (50-2,000 emp):
1) Sofie Zarrabi - Head of AI Deployments, Cresta (~650 emp, Series D; CX AI agents). High.
2) Tony Gentilcore - Co-founder, Glean (~1,700 emp; Work AI agents/assistant/search). High. Technical co-founder NOT previously captured (brain had Arvind Jain, Eddie Zhou, T.R. Vishwanath).
3) Sameer Paranjpye - SVP Engineering, Baseten (~312 emp, Series F; inference infra for production/agentic workloads). Med-High.
4) Colin McGrath - Head of Infrastructure, Baseten. Medium.
5) Ian Nowland - Head of Hardware Infrastructure, Baseten. Medium.
METHOD / YIELD NOTES: LinkedIn CONTENT search (Signals 1-3) was low-yield for adds this run - post authors were overwhelmingly vendors, AI educators, consultants and enterprise/services leaders (>2,000 or <50 emp), not ICP buyers. Highest-yield path was LinkedIn PEOPLE search scoped to specific in-band agent-native companies, then dedup + web-verified headcount. Many strong ICP hits found this run were ALREADY in the brain and correctly skipped: Anubhav Sharma (Head of Agentic AI, Jeeva AI), Venkat Peri (Head of Agentic AI, Advisor360), Masashi Beheim (VP Eng, Parloa), Ming Yin & Jove Zhong & Tim Shi (Cresta), Jason Fang & Nikos Alexakis (Aisera), Srikanth Konjeti (Gnani.ai). The list is very mature: 326 companies, 244 with only a single contact.
BEST REMAINING VEIN: 2nd/3rd senior TECHNICAL leaders at already-covered in-band companies where only the CEO/CTO was captured (e.g., Glean co-founder; Baseten eng/infra leaders). High-priority accounts flagged: Baseten (was CTO-only; now +3 eng/infra leaders - strong cost/reliability pain surface) and Glean (added technical co-founder).
EMERGING VOC (from Signal-1 content sweep, see 2 VOC entries added): 'demo in 2 weeks, production in 6 months' - the gap is eval + observability + guardrails + COST GOVERNANCE, with runaway token spend from recursive loops/retries as the #1 cost driver; plus demand for per-agent cost-per-run visibility over static thresholds. Outreach copy should lead with per-agent cost-per-run visibility + reliability guardrails for teams scaling 1 -> 5+ agents.
NEXT RUN: vary queries; prioritize company-scoped people search on the 244 single-contact in-band companies; mine comments on high-engagement ICP-authored posts (e.g., Anubhav Sharma/Jeeva posts) for net-new individuals; use title+headcount filtering.
sales intelclaude-connector · 30 Jul 2026
ICP Prospect Scanner run — 2026-07-30 (5 added; new accounts: Phaidra, Botminds, Talkdesk)
ICP Prospect Signal Scanner — automated run, 2026-07-30.
RESULT: 5 net-new ICP people added (ids 501-505), meeting the minimum-5 target.
1. Vedavyas Panneershelvam — CTO & Co-founder, Phaidra (Series B, RL control agents / 'AI factories'). High.
2. Ansari Ismail — CTO & Co-Founder, Botminds.ai (~59 emp, governed agentic AI for lending/docs). High.
3. Xiangru Chen — VP of Engineering, Cresta (300+, Series C, contact-center AI agents) — new contact at an existing account. High.
4. Yunjing Ma — VP of AI Engineering, Talkdesk (~1,300, Voice AI Agents + AI Agent Platform) — NEW account. Medium (later-stage than Series A-C).
5. Jing Jin — Senior Director of AI, Talkdesk — second contact at the new Talkdesk account. Medium.
NEW ACCOUNTS surfaced (not previously in brain): Phaidra, Botminds.ai, Talkdesk. Flag Phaidra & Botminds as high-priority net-new agent-native accounts.
METHOD / WHAT WORKED: LinkedIn CONTENT search (Signal buckets 1-3: agent cost, reliability, competitor mentions like Langfuse/Helicone) was low-yield this run — results were dominated by consultants, educators and sub-50-employee startups, not ICP decision-makers. Pivoted to LinkedIn PEOPLE search on ICP titles ('CTO agentic AI startup', 'VP Engineering AI agents', 'Director of AI agents') cross-referenced with web verification of headcount/funding/agent-shipping. This was far more productive and is the recommended primary method for future runs.
DEDUP NOTE: Many strong finds were ALREADY in the brain and were correctly skipped — Masashi Beheim (Parloa), Anubhav Sharma (Jeeva), Daniel Hoske (Cresta), Souvik Sen (Ema), Daniel Palmer (Relevance AI), Shawn Wen (PolyAI). The brain already covers essentially every well-known agent company (Sierra, Decagon, Cognition, Cresta, PolyAI, Ema, Relevance, Parloa, Kore.ai, Ushur, Aisera, Uniphore, Observe.AI, Forethought, etc.), so future runs should hunt less-obvious mid-size agent companies and NEW contacts at existing accounts.
OUTREACH COPY IMPLICATION: Every prospect this run maps to the same wedge — 'scale agents in production reliably while controlling cost-per-run, with real observability into what each agent costs.' Lead with cost-per-run visibility + reliability for teams already running agents at scale (contact-center voice agents and document/control agents are the hottest sub-segments this run).
DATA HYGIENE: A stray test row '__TEST_DELETE_ME__' (id 500) was created while debugging a JSON-RPC id bug (float ids are rejected by the MCP endpoint; use integer ids). No delete_person tool exists — please remove id 500 manually.
CONSTRAINTS HONORED: no fabricated profiles/sizes/quotes (pain points are labeled inferred-from-role, not quotes); Aptos Retail excluded; no outreach performed.
sales intelclaude-connector · 28 Jul 2026
ICP Prospect Signal Scanner — Run 2026-07-28 (7 added)
ICP Prospect Signal Scanner — automated run 2026-07-28.
RESULT: 7 new ICP people added to the People Library (target was 5+). 0 duplicates written.
Added: Bharat Kumar (AVP Eng, Kore.ai), Ajay Choudary (Director Eng, Kore.ai), Srinivasa Rao Patchigolla (Sr Director, ThoughtSpot), Shailesh P. (Director Eng, Yellow.ai), Deven Panchal (Director Eng, Uniphore), Pranav Maydeo (Sr Director Eng, Uniphore), Thejas Bhat (Director Data Eng & AI, Uniphore). All are senior technical/eng leaders (Director–SVP) at 50–2,000-employee companies actively shipping enterprise AI agents (conversational/voice/agentic-analytics).
MOST PRODUCTIVE APPROACH: Company-scoped LinkedIn PEOPLE search on distinctive agent-native company names (Kore.ai and Uniphore each yielded a cluster of senior eng leaders). This is a Signal-4-style approach (ICP leaders at agent-shipping companies).
WHAT DID NOT WORK THIS RUN: Signal 1–3 CONTENT/post searches were low-yield — LinkedIn mangled boolean OR queries ('langfuse OR helicone OR litellm' was auto-corrected to gibberish and returned loan/stock spam), and agent-cost content results were dominated by consultants, recruiters and AI influencers rather than ICP practitioners. Generic title people-searches ('Head of AI agents', 'VP Engineering AI agents') mostly surfaced people with no company named or at too-large companies (Microsoft, Salesforce, Disney, ITV, Freshworks) that fail the 50–2,000 filter.
DEDUPE NOTE: The brain (448 people) already had strong coverage — 5 of the first 8 candidates found (Tim Shi/Cresta, Masashi Beheim/Parloa, Razvan Kusztos/PolyAI, Ershad Ali Mohammad & Pattabhi Rama Rao Dasari/Kore.ai) were already present, so future runs should push into less-mined companies.
DATA-QUALITY / HONESTY NOTE: Pain points recorded on each person are INFERRED from role + company context, not verbatim quotes — these prospects were discovered via people-search, not by observing them personally posting about cost/reliability pain this run (except Srinivasa Rao Patchigolla, who authored a genuine post on bridging 'cool AI demo' to 'trusted production system' with Spotter agentic analytics). Company sizes are estimates. Only one verbatim VOC quote was genuinely observed this run (see VOC: Musa Usmani, $4,200 runaway-agent-loop weekend).
OUTREACH-COPY IMPLICATION: The recurring, high-resonance pain across every added account is the same triad — (1) agent cost blowout at high interaction volume, (2) no per-run/per-agent cost visibility, (3) production reliability/guardrails while scaling from few to many agents. Lead outreach with per-run cost observability + reliability guardrails for multi-agent voice/CX platforms.
HIGH-PRIORITY TARGETS: Kore.ai and Uniphore (multiple senior eng leaders, large-scale multi-agent voice deployments = acute cost/observability pain).
INFRA NOTE: The Alpha Brain MCP connector was not attached to this session and the sandbox had no network; the brain was read/written via the browser against /api/mcp (JSON-RPC) using the provided key.
ICP Prospect Signal Scan — run 2026-07-28
ADDED 5 new people (ids 439-443), all Head/Director-level technical AI leaders at 50-2,000-employee companies actively shipping conversational/agentic AI:
1. Sarah Sachs — AI/Eng Lead, AI (Head of AI Engineering), Notion — HIGH. Only add with a genuine on-signal content quote (agent sprawl, model lock-in, observing agents in prod for reliability AND cost).
2. Umesh S. — Head of AI Engineering, Uniphore — Medium-High. Uniphore already an active ICP account in the brain (Mayuram/Saxena/Lu).
3. Toshish Jawale — Head of AI Engineering, Invoca (ex-CTO Symbl.ai) — Medium-High.
4. Konstantin Bukin — Director of AI, Saritasa — Medium (service provider building agents for clients; featured in AllTech post on the production 'proof-of-concept trap').
5. Dima Galat — Head of AI Engineering, Satisfi Labs — Medium.
SKIPPED as duplicates (already in brain): Varun Kacholia (Co-Founder & CTO, Eightfold AI), Anubhav Sharma (Head of Agentic AI, Jeeva AI).
SKIPPED on ICP fit: Cosmin Andriescu (Lumenova AI, ~30 emp — below 50 floor); most content-search authors were consultants/freelancers, individual engineers, recruiters, or big-co (Microsoft/AWS/Oracle/NetApp/Freshworks/Acko) — outside title or size bands.
MOST PRODUCTIVE APPROACH: LinkedIn *people* search with 'Head of Agentic AI / Head of Engineering agentic AI / co-founder CTO AI agents' surfaced clean ICP-title leaders at named agent companies far better than the prescribed content-keyword post searches, which were dominated by non-ICP creators. Signal buckets: Bucket 2/4 produced the one quoted-signal add (Sachs); Bucket 4 (people-search adaptation) produced the other four. Buckets 1 & 3 (content post searches) produced strong PAIN QUOTES but no addable ICP authors.
EMERGING PATTERN for outreach copy: the loudest, most repeated pain is 'the model/demo is easy — reliability + observability + cost-per-successful-run in PRODUCTION is the hard part.' Lead with cost-per-run visibility + reliability, not raw token price. See 2 VOC entries added this run (Anurag Karuparti 'harder problem…beyond a demo'; ARVIND R 'reliability tax…not on any pricing page').
HIGH-PRIORITY FLAGS: Sarah Sachs (Notion) — publicly articulating the exact pains Alpha sells against; warmest signal of the run. Uniphore and Invoca are mid-size agent-native accounts worth multi-threading (Uniphore already multi-contact in the brain).
OPERATIONAL NOTE: The Alpha Brain MCP tools were NOT connected in this session and the sandbox had no network route to the app. Worked around it by driving the REST/MCP JSON-RPC endpoint (/api/mcp) via in-page fetch from a logged-in Chrome tab. Company sizes for the smaller startups are estimates (labeled as such in each person's notes); pain points for the four people-search adds are inferred from role+company, not quoted posts.
ICP Prospect Signal Scan - 2026-07-25 (6 new prospects)
Run date: 2026-07-25. New ICP people added: 6 (all Medium/High confidence, all deduped against 351 existing; brain now 357).\n\nPeople added:\n1. Kuldeep Singh Chauhan - Head of AI, Emergent (~310 emp, Series C unicorn coding agents) - HIGH - Signal 1&4\n2. Rohit Choudhary - Co-Founder & CEO, Acceldata (~294 emp, Series C agentic data platform) - HIGH - Signal 1&4\n3. Aiden Lee - Co-founder & CTO, Twelve Labs (~190 emp, Series B agentic video platform) - MEDIUM - Signal 4\n4. Paritosh Mohan - VP of Engineering, Twelve Labs - MEDIUM-HIGH - Signal 4\n5. Puneet Agarwal - SVP of Engineering, Observe.AI (~350 emp, Series C voice agents) - MEDIUM (LinkedIn URL probable, role confirmed via press release) - Signal 4\n6. Saurabh Anand - Head of Product (AI-focused), Emergent - MEDIUM - Signal 1&4\n\nMost productive signal buckets: Signal 1 (ICP writing about agent reliability/cost) and Signal 4 (ICP building agents) - surfaced Emergent's Head of AI and Head of Product via a public talk on reliable agentic systems. Signal 3 (competitor engagement: Helicone/Portkey/Langfuse/AWS AgentCore) returned only vendor/comparison content, no named ICP engagers this run.\n\nHigh-priority flags: Emergent (fastest-growing AI-native, 75->310 emp in 6 months, shipping coding agents to 5M+ users - two contacts now in brain) and Acceldata (technical CEO publicly framing 'agentic data management' - strong reliability/observability pain).\n\nEmerging pattern for outreach copy: lead with the pilot-to-production reliability gap and cost-per-run visibility (see VOC insight this run). Message senior technical leaders (Head of AI / VP Eng / CTO) on 'know what each agent costs and trust it in production', not generic 'build agents faster'.\n\nNotes/decisions: Skipped LLUMO AI (seed-stage, ~52 emp, tooling vendor - below Series A). Skipped Spectro Cloud/Saad Malik (Series D + AI-infra vendor, outside ICP stage & not an agent-shipper). Skipped Vikram Sharma (Marmon, >2000 emp industrial). Dropped Jithendra Vepa & David Hariri (already in brain). Excluded Aptos Retail per instructions.
Added 6 new ICP people (ids 343-348; library now 346 total). All sourced from the AI Engineer World's Fair 2026 speaker roster (verifiable name+title+company), cross-checked for size/funding/agent-shipping and deduped against the existing 340-person library.
New people:
- Anuj Iravane — Head of AI, Anterior (~50 emp, Series B; ships 'Florence' prior-auth agent) — HIGH
- Saul Howard — VP of Engineering, Anterior — HIGH
- Viren Baraiya — Co-Founder & CTO, Orkes (~73 emp, Series B; agentic orchestration) — MEDIUM-HIGH
- Gil Feig — Co-Founder & CTO, Merge (~141 emp, Series B; 'Agent Handler'/connective infra for production AI) — MEDIUM
- Rania Khalaf — Chief AI Officer, WSO2 (~800-1,250 emp; Agent Manager platform; ex-IBM Research) — MEDIUM
- Dan Feng — Senior Director of Engineering, Maven Clinic (~911 emp; agentic care in production) — MEDIUM
Most productive bucket: Signal 4 (ICP leaders speaking about building/shipping agents) via the conference roster; Signal 1 themes (agent reliability/cost in production) overlapped strongly (esp. Orkes). Signals 2 & 3 (LinkedIn post/comment + competitor-content engagement) were NOT productive — LinkedIn post/comment pages are gated from web search and DuckDuckGo. Conference/podcast rosters are a far more reliable verifiable source; recommend leaning on them in future runs.
High-priority flags:
- Anterior is the standout: 50-person Series B literally shipping an AI agent (Florence) in a regulated, high-cost-per-run setting — two ICP buyers captured (Head of AI + VP Eng). Best outreach target.
- Orkes sits in the agent-reliability/orchestration space (possible partner OR competitor) — flag before outreach.
Emerging VOC pattern (logged, voc id 114): 5/6 prospects express the same need — production reliability + cost/observability/control of agents as they scale from few->many. Outreach copy should lead with 'know what every agent does and costs per run, and keep them reliable as you scale' rather than generic 'AI agents' messaging.
Caveats: WSO2 & Maven Clinic are in the 50-2,000 band and shipping agents but larger/later-stage than the A-C core target (Medium). No LinkedIn profiles fabricated — profileUrl blank except Rania Khalaf (official WSO2 team page). Pain points are inferred from role/company/public positioning, not verbatim quotes. Env note: Alpha Brain MCP tools were not connected this session and the sandbox blocks the domain, so writes were made via the /api/mcp endpoint through the connected browser (Bearer auth).
noteclaude-connector · 24 Jul 2026
ICP scan test persist 2026-07-24
persist check
sales intelclaude-connector · 22 Jul 2026
ICP Prospect Signal Scan — 2026-07-23 run
Added 5 new ICP-matching people (IDs 306–310); 0 duplicates (checked against 303 existing).
People added:
1. Matt Nassr — Head of Global Data Eng & AI Transformation, Optiver (~1,600) — Medium (strong pain fit; caveat: trading firm, not SaaS)
2. Madhav Jha — Co-Founder & CTO, Emergent (~75, Series C, AI-native coding agents) — Medium-High
3. Dean Bloembergen — Co-Founder & CTO, Owner.com (~500, Series C, vertical SaaS, multi-agent 'AI Executives') — High (best fit)
4. James Fox — Co-Founder & CTO, Gamma (~360, Series B, AI-native) — Medium
5. Jon Noronha — Co-Founder & CPO, Gamma (~360, Series B) — Medium
Most productive signal buckets: Signal 4 (ICP technical founders/leaders at companies shipping agents) and Signal 1 (leaders speaking publicly about scaling/compounding agents). Signal 2 (non-ICP posts w/ ICP engagers) and Signal 3 (competitor-content engagers: Helicone/Portkey/Langfuse/Braintrust) produced no verifiable named individuals this run.
High-priority flag: Dean Bloembergen / Owner.com — Series C vertical SaaS explicitly running a FLEET of production agents ('AI Executives'), the cleanest cost/reliability/oversight ICP. Matt Nassr / Optiver is a strong pain-signal match (context waste, scaling 1→many) despite non-SaaS company type.
Emerging pattern for outreach copy: the recurring wall is NOT building one agent — it's scaling to a reliable, cost-controlled multi-agent fleet (see VOC #101). Lead outreach with 'shared context/eval/governance + per-run cost visibility across your agent fleet' rather than single-agent tooling.
Method note (autonomous run): general web search surfaced mostly SEO listicles for site:linkedin.com queries; the productive sources were conference speaker/session pages (AI Engineer World's Fair 2026, SaaStr AI Annual 2026), TechCrunch funding coverage, and podcast episode pages, cross-verified for title/company/size. Two profile URLs (Madhav Jha, James Fox) left blank rather than fabricated. Several strong-signal names were excluded for being IC-level (Gabe De Mesa/OpenGov, Vaidas Razgaitis/Higharc), too big (Shopify, Atlassian, LinkedIn, Meta, Coinbase), <50 employees (Glyphic/45), or past Series C (Lovable, Abridge).
sales intelclaude-connector · 19 Jul 2026
ICP Prospect Signal Scan — 2026-07-20 run summary (6 people added)
Run date: 2026-07-20. Added 6 new ICP-matching people (brain people 263 -> 269). All verified at companies with 50-2,000 employees actively shipping AI agents in production; none were pre-existing in the brain; no Aptos Retail contacts touched.
NEW PEOPLE ADDED:
1. Yoni Blumenfeld — Co-Founder & CTO, Sett (~50 emp; AI agents for mobile-game marketing/UA; Series B $30M; Zynga/Playtika/Papaya). Signal 4. High confidence.
2. Deip Kumar — Co-Founder & CTO, Gradial (~85 emp; agentic enterprise-marketing platform across Adobe/Salesforce/ServiceNow/Databricks; Series C $65M, $675M val). Signal 4. High.
3. Alexander Matthey — CTO, Parloa (~300 emp; agentic contact-center platform, "build/test/deploy millions of agents"; Series C $120M, $1B val; ex-Adyen CTO). Signal 4. High.
4. Amit Carmi — Co-Founder & CEO, Sett (~50 emp; technical co-founder, ex-Unit 8200). Signal 4. Medium (CEO but technical decision-maker).
5. Daniel Palmer — Co-Founder (technical), Relevance AI (~124 emp; "AI workforce" agent platform, ~40k agents created/month; Series B, Bessemer). Signal 4. Medium.
6. Gabriel Hubert — Co-Founder & CEO, Dust (~144 emp; horizontal enterprise agent platform, "multiplayer AI", 3,000+ companies; $40M raised 2026). Signal 1/4. Medium.
MOST PRODUCTIVE BUCKETS: Signal 4 (ICP building/shipping agents) was by far the most productive, sourced via 2026 funding announcements + founder profiles (Bessemer, TechCrunch, Axios, BusinessWire). Signal 1 (writing about agent cost/reliability) surfaced strong themes but few *new* named ICPs — most cost/reliability commentators were either non-ICP analysts or CTOs already in the brain. Signals 2 & 3 (competitor/influencer engagement) returned mostly generic listicles; LinkedIn post-level engagement is hard to mine via web search and yielded no net-new qualified names this run.
DEDUPE NOTES: The brain is already saturated with well-known agent-company leaders. Confirmed-present and skipped this run: Roey Lalazar (Wonderful), Sami Shalabi (Maven AGI), Stanislas Polu (Dust), Daniel Vassilev & Jacky Koh (Relevance AI), Tim Shi (Cresta), Prabhav Jain (11x), David Hariri (Ada). Skipped as STALE (person left the role): Tim Shi (now Recursive, not Cresta), Jessica Popp (left Ada in 2023, now Rula). Skipped on size uncertainty: Patronus AI (headcount likely <50, unconfirmed).
HIGH-PRIORITY FLAGS: Parloa (Matthey) and Gradial (Kumar) are the strongest fits — both are scaling large multi-agent/multi-tool fleets where cost + reliability observability is an explicit, current need. Sett is a clean small-ICP entry (two contacts).
OUTREACH-COPY PATTERNS (see VOC added this run): lead with reliability-at-scale + per-run cost visibility as agent count grows from 1 -> many; for platform companies (Gradial, Parloa, Relevance, Dust) emphasize control/observability across heterogeneous agent fleets; for Sett emphasize agent unit-cost efficiency.
CAVEATS: Pain points/challenges are INFERRED from each company's public positioning and funding coverage (not verbatim quotes) — no quotes were fabricated. Alexander Matthey's LinkedIn slug (linkedin.com/in/alexander-matthey) is inferred from his Parloa post activity and should be verified (common name). Amit Carmi's profile URL left blank (unconfirmed).
Added 6 new ICP-qualifying people (brain now 223).
New people (all verified, none duplicates, none Aptos Retail):
218 Tina Kung - Co-Founder & CTO, Nue (51-100, Series A) - Signal 1 - HIGH
219 David Hsu - Founder & CEO, Retool (~415, Series C) - Signal 4 - MED-HIGH
220 David Paffenholz - Co-Founder & CEO, Juicebox (~65, Series B) - Signal 4 - MED-HIGH
221 Denis Yarats - Co-Founder & CTO, Perplexity AI - Signal 1/3 - MED
222 Timothee Lacroix - Co-Founder & CTO, Mistral AI - Signal 1 - MED
223 Tuomas Artman - Co-Founder & CTO, Linear (Series C) - Signal 4 - MED
Most productive signals: Signal 1 (ICP writing/speaking about agents & cost) and Signal 4 (technical founders shipping agents). Signal 2/3 (engagement-based) yielded little via US web search since LinkedIn post/comment engagement is poorly indexed.
High-priority: Tina Kung/Nue (cleanest ICP - CTO, right size, shipping agents, auditability pain). Denis Yarats/Perplexity has the most on-ICP explicit pain (token/context waste) - great outreach hook even if Perplexity may build in-house.
Skipped (below 50-emp floor or adjacency): /dev/agents (~6 ppl), Coval, Traversal, Cleric, LinqAlpha (size unconfirmed <50), Sett (~50 borderline). Sierra excluded (Series E, and Clay Bavor already in brain).
Emerging VOC patterns (2 logged): (1) token/context cost & efficiency at scale [CTO persona]; (2) reliably orchestrating & governing many agents in production [technical-founder persona]. Outreach copy should lead with cost-per-run visibility + durable orchestration/reliability, not 'build an agent'.
Note: Alpha Brain MCP tools were not connected in this session; reached the brain via its /api/mcp JSON-RPC endpoint through the browser instead.
Added 6 new ICP people (brain 211 -> 217): Nicholas Arcolano (Head of AI & Research, Jellyfish, HIGH); Vitaly Gordon (Co-founder & CEO, Faros AI, MED-HIGH); Mingsheng Hong (VP/Head of AI, Ironclad, MED-HIGH); Brij Mohan Singh (Head of AI, The Modern Data Company, MED); Animesh Kumar (Co-founder & CTO, The Modern Data Company, MED); Ellie Zhou (Sr Director Applied AI, Ironclad, MED). New companies to brain: Jellyfish, Faros AI, The Modern Data Company. Ironclad already in brain (Cai GoGwilt) -> Hong & Zhou are new contacts at an existing account.
Most productive bucket: Signal 1 (ICP publicly writing/speaking on agent cost & reliability) -> 4 of 6. Signal 4 (ICP at active agent-builders) -> 2. Signal 3 (Langfuse/Helicone/Portkey engagement) returned mostly consultants/fractional CTOs/influencers -> no adds. Signal 2 (HN/Reddit) not productive this run.
High-priority: Nicholas Arcolano (Jellyfish) exact-fit Head of AI, sharpest pain on per-run cost + ROI; Mingsheng Hong (Ironclad) VP of AI with "Tokenmaxxing"/trusted-throughput thesis, enables multi-threading the existing Ironclad account.
Outreach copy signals: (1) "What do your agents cost per run, and does that spend map to business value?" resonates across CEO / Head-of-AI / VP-AI. (2) Reliability angle — "trusted throughput", closing the lab-vs-production gap — lands with VP/Head of AI. Note: AI-native-startup pool is now saturated in the brain (156 companies); highest-yield remaining targets are Heads of AI / VP Eng at mid-market non-AI-native SaaS (data infra, eng-intelligence, legal) building agents internally.
sales intelclaude-connector · 15 Jul 2026
ICP Prospect Signal Scanner — Run 2026-07-15: 6 net-new people added (IDs 196–201)
Added 6 net-new ICP-qualifying people (checked against the existing 195-person People Library — no duplicates; none from Aptos Retail; all Medium or higher confidence; all confirmed 50–2,000 employees AND actively shipping/building agents in production):
1. David Zeng — Co-Founder & Head of Engineering, Eve/eve.legal (~294 emp, Series B $103M, $1B val) — Signal 4 — HIGH. AI workforce + EveOS for plaintiff firms; 1,200+ firms, 200k+ cases/yr. Highest-priority: cleanest technical decision-maker + clear agents-in-production at scale.
2. Ian Christopher — Co-Founder & CTO, Qventus (~238 emp, Series D) — Signal 4 — Medium-High. Hospital-operations agents/"operational assistants." (Series D = one stage past A–C; included on headcount + agent-native fit.)
3. Gigi Yuen-Reed — Chief Data & AI Officer, Cohere Health (~500–1,000 emp) — Signal 4 — Medium-High. Clinically-trained agentic AI for prior auth; 85% real-time PA approvals. NOTE: distinct company from model-lab "Cohere" already in library — no overlap.
4. Ryan Eldridge — Co-Founder & CTO, Liberate/liberateinc.com (~50 emp, Series A+ $72M total) — Signal 4 — Medium. Reasoning AI agents for P&C insurance (quoting/claims/endorsements). ~50 emp = exactly at the floor; verify headcount before prioritizing.
5. Karan Goel — Co-Founder & CEO, Cartesia (~122 emp, Series A $64M) — Signal 3/4 — Medium. Line voice-agent platform (Sonic/Ink); latency + SSM cost-efficiency angle. Infra/competitor-adjacent — positioning review.
6. Adam Sypniewski — CTO, Deepgram (~326 emp, Series C) — Signal 3 — Medium. Voice Agent API for real-time production voice agents. Model/infra vendor — strongest build-vs-buy/competitor caveat; positioning review before outreach.
MOST PRODUCTIVE BUCKETS: Signal 4 (ICP technical decision-makers shipping agents) produced 4 of 6 via company-first discovery (identify in-band agent-native companies not yet in the brain, then confirm the technical decision-maker + headcount). Signal 3 (competitor/ecosystem-adjacent — voice-agent platforms) produced 2 (Cartesia, Deepgram). Signals 1 and 2 (LinkedIn post + comment-engagement scraping) again yielded nothing verifiable — US-only web search surfaces SEO listicles and cost-guide content, not individual LinkedIn posts or comment-graph engagers. Recommend continuing company-first discovery; connect LinkedIn Sales Navigator/enrichment to unlock Signals 1–2.
HIGH-PRIORITY FLAGS: Eve (David Zeng) — clearest High-confidence, AI-native, agents at real scale. Qventus and Cohere Health — strong regulated-vertical fits with named decision-makers.
SCREENED OUT (logged to avoid re-checking): Paradox (acquired by Workday, Aug 2025 — not independent); Sierra (Series E, $15.8B — beyond A–C stage and Clay Bavor already in library); Sedric AI (~27 emp — under 50 floor); Commure (CTO identity ambiguous across sources + headcount likely >2,000 / unverifiable — held to avoid fabrication); Brightwave, Rogo (in library), Traversal (founded 2024, likely <50), LinqAlpha/Build/AIsa (seed-stage / <50) — all fail the 50-emp floor or verification bar. Rogo, Hebbia, Nooks, Supio, Pulumi surfaced again and were correctly skipped as existing entries.
EMERGING PATTERN (see VOC insight #59): 4 of 6 prospects are in REGULATED verticals (legal, healthcare×2, insurance) and frame the urgent pain as production RELIABILITY + action-level AUDITABILITY/GOVERNANCE — explicitly above raw cost, because a wrong autonomous action carries legal/clinical/regulatory consequence. Cost-per-run visibility recurs as a secondary/nice-to-have here. Recommended outreach split: for regulated-vertical agent builders LEAD with reliability + auditability/control and close on cost-per-run; for high-volume voice/infra teams (Cartesia, Deepgram) lead with latency/cost-efficiency at scale.
INTEGRITY NOTES: No fabricated profiles/sizes/quotes. All per-person pain points are labeled as inferred from public product/positioning — no verbatim first-party quotes were available this run. profile_url left blank for all 6 (no LinkedIn URL confirmed in sources — not fabricated). All headcounts/stages cited to Tracxn/Crunchbase/press as of the sources' dates.
ICP Prospect Signal Scan — 2026-07-11 (run summary)
Added 5 new ICP-qualifying people (IDs 63–67), all Medium / Medium-High confidence, none duplicating the existing 62-person library:
1. Edward Wu — Founder & CEO, Dropzone AI (~50-56 emp, Series A) — autonomous security/SOC agents in 6+ prod environments. Medium-High.
2. Arjun Prakash — Co-Founder & CEO, Distyl AI (159 emp) — frontier enterprise agents for Fortune 500. Medium.
3. Muddu Sudhakar — Co-Founder & CEO, Aisera (~311 emp) — agentic AI platform, IT/ops/CX agents in prod. Medium (NOTE: Series D, beyond stated A–C stage).
4. Matt Harpe — Co-Founder & CEO, Basis (Series B $100M/$1.15B; size ~50-150 est) — end-to-end accounting/tax/audit agents at ~30% of Top 25 firms. Medium.
5. John Stecher — CTO, Norm Ai (Series C $120M/$1.2B; size ~100-200 est) — legal/compliance agents that also SUPERVISE other agents. Medium; strongest strategic adjacency to thealpha's agent operating layer.
Most productive buckets this run: Signal 4 (ICP building/shipping agents) and Signal 1 (ICP writing/speaking about agent cost & reliability). Signal 3 (competitor-content engagement) and Signal 2 (non-ICP posts w/ ICP commenters) were weak — LinkedIn post/comment-level engagement is hard to retrieve via web search; recommend a LinkedIn-connected source or Sales Navigator for those buckets next run.
High-priority flag: Norm Ai (John Stecher / John Nay) — their product literally supervises fleets of other AI agents in regulated environments; closest thematic overlap with an "agent operating layer." Dropzone AI is the cleanest confirmed-size Series A fit.
Emerging patterns for outreach copy (see 2 VOC insights added this run): (a) in regulated/high-stakes verticals the buying trigger is provable RELIABILITY + AUDITABILITY of agent decisions, not model quality; (b) the pain is shifting from "one agent's cost" to "controlling cost + behavior across a growing FLEET of production agents." Lead with fleet-level visibility + auditability rather than raw token savings.
Data-quality note: headcount for Basis and Norm Ai is inferred from funding/scale, not officially disclosed — verify before outreach. Skipped for failing ICP: Sema4.ai (44 emp), Upsonic (15 emp), Forethought & Sana (acquired into Zendesk/Workday, now >2000), Gradient Labs (~40, below 50 threshold).
decisionclaude-connector · 8 Jul 2026
ICP stack v1 filed (Sensibility Audit mapping) + canonical size band locked at 50–500
Mapped Alpha's ICP against the ICP Playbook + Sensibility Audit framework (Vishal Virani, AIBoomi #26 — companion to entry #63). Four-line stack, each filed by true source:
LINE 1 · ACCOUNT [BELIEVE]: Software/SaaS companies, 50–500 employees, actively shipping agents in production/near-production, est. $10k–$100k+/mo LLM spend, worldwide. CANONICAL BAND DECISION: 50–500 employees. This resolves the discrepancy across Decision #29 (50–500), Entry #52 (51–500), and Positioning Canon #54 (20–500 with tiers). All future copy, targeting, and tier definitions use 50–500. Canon #54's two-tier structure survives but re-anchored: Tier 1 = 50–150 emp ($99, pure PLG via Arena), Tier 2 = 150–500 emp ($499, Arena aha + one 20-min technical call). Filed BELIEVE per Entry #52's own language: "to be validated by outreach, not by more planning."
LINE 2 · TRIGGER [BORROWED→BELIEVE]: The 1→5 agent scale wall — cost blowout ($1k estimate → $3.8k invoice), unmeasured reliability, stalled pilots (88% never ship). Source: research briefs and market data, NOT verbatim prospect language. Gap: zero verbatim pain quotes in the brain (VoC section empty). Upgrading this line to KNOW requires asking real prospects "what happened right before you started looking?"
LINE 3 · COMMITTEE [BELIEVE]: Champion personas defined — CTO (cost/control/sovereignty), VP Eng (reliability/observability), Head of AI (ownership/compounding). No blocker named yet; at $99/$499 PLG a thin committee is defensible, but the $499-tier technical call is where the first blocker will surface.
LINE 4 · DISQUALIFIER [KNOW-adjacent]: NOT enterprises (procurement friction, won't entertain solo-founder vendor) and NOT pre-seed (no spend to optimize) — Decision #29, backed by real evidence: the mis-targeted 25/week outreach with mostly negative replies.
AUDIT FINDING: Demand Ledger = EMPTY. Zero paying customers, zero pipeline accounts, six Gojiberry prospects are queued connection requests — not yet costly signals (paid, piloted, deployed, 3+ hrs of working meetings, or shared systems access). Per playbook §3: with <3 ledger entries, the next 14 days are about GENERATING COSTLY SIGNALS, not refining strategy documents.
ARCHETYPE: True Believer — high ownership (strategy genuinely derived in-house), thin proof (empty ledger). Prescription: instrument, don't ship more strategy. Falsifiable bet already in flight: Experiments #1/#2 + live Gojiberry campaign.
MAINTENANCE: Re-run the audit quarterly. Re-file every stack line. Watch oldest convictions hardest.
Two fixes applied to the Gojiberry outreach system to implement Decision #50 (cost is the hook, harness is the product) and Decision #29 (locked ICP):
1. MESSAGING: All 6 priority prospects (Polu/Dust.tt, Li/Artisan, Raghunathan/Hyperbound, Han/Voiceflow, Sen/Ema, Crivello/Lindy) had their intent context updated with explicit outreach angle: lead with Arena only (free 5-min BYOK tool, total run-cost visibility), never mention Alpha/harness/pillars, never pitch model-switching or open-source savings — frame as run-cost CONTROL, not cheaper models (per Experiment #47: model-switching is a dying angle, API prices fell ~80%). Per-person tone cues added (technical for Raghunathan, founder-to-founder for Crivello, margin-protection for Li and Han). Note: user_note field is read-only via connector; angle lives in the intent field, which the AI writer consumes. CAVEAT: these are instructions to Gojiberry's AI writer, not guarantees — spot-check the first generated emails before the queue reaches the 6 priority prospects.
2. ICP DRIFT FIXED: Lead-sourcing agent #18132 was configured for company sizes up to 5,000 employees, Finance + Healthcare industries, and Europe — outside the locked ICP, already producing off-target leads (e.g. medical device CTO, 1001-5000 emp). Tightened to: 51-200 + 201-500 employees, Software Development & SaaS + Technology only, North America only. Lead volume will drop; quality should rise. Note: the ~203 unsent contacts already in list #28587 were sourced under the old drifted ICP — worth a purge pass on obvious misfits before invitations reach them.
sales intelclaude-connector · 6 Jul 2026
LinkedIn seat verified LIVE — invitations sending; queue depth means priority prospects may wait weeks
Verified via Gojiberry contact-level campaign data: LinkedIn seat is connected and actively sending — invitation to Kunal Gosar (Addepar) sent 2026-07-06 14:38 UTC. The earlier seat reconnection issue is resolved. Lead-sourcing agent also healthy: multiple runs daily since June 18, zero errors. NEW FINDING: send queue is long — 398 contacts in list, ~203 with no campaign activity yet, throughput appears to be a few invitations/day (LinkedIn safety limits). The 6 priority prospects added today may wait 1-3 weeks for their invitation unless prioritized. Reinforces: founder DM to Stanislas Polu (task 23) should NOT wait for the campaign.
sales intelclaude-connector · 6 Jul 2026
Prospect batch enrolled in Gojiberry campaign — connection requests queued
All 6 priority prospects from the July 5 signal scan created in Gojiberry and enrolled in list #28587 → campaign "CTO · North America · Software Development & SaaS" (#20818, sequence: invitation → AI email → AI message → profile visit → follow-ups). Contacts: Stanislas Polu (CTO Dust.tt, #1), Ming Li (CTO Artisan AI), Atul Raghunathan (CTO Hyperbound), Tyler Han (CTO Voiceflow), Souvik Sen (CTO Ema AI), Flo Crivello (CEO Lindy AI). Each carries intent context (funding, agent scale, cost-pain angle) for AI message personalization. CAVEATS: (1) LinkedIn seat must be connected for invitations to actually send — verify seat status in Gojiberry; (2) four contacts (Ming Li, Tyler Han, Atul Raghunathan, Souvik Sen) have common-name LinkedIn slugs from the research scan that MUST be spot-checked in Gojiberry before the invitation step fires, or requests go to the wrong person; (3) task 25 (load prospect list into CRM) is now half-done — weekly job-posting signal refresh still pending; (4) tasks 23/24 (personal DMs to Polu/Raghunathan) remain open — campaign automation is not a substitute for the founder DM to the #1 prospect.
researchresearch-agent · 6 Jul 2026
Research: Greenfield vs brownfield — who is Alpha's ICP?
## Question
Is Alpha's ICP greenfield AI-native companies, or brownfield incumbents onboarding AI? Vishnu's thesis: greenfield firms may already have built agent harnesses, so the incumbents "figuring it out" are the ones that need help.
## Verdict
Partially validated. The instinct that struggling teams need the harness is correct, but the greenfield/brownfield binary is the wrong segmentation axis and, taken literally, would steer Alpha toward the worst-fit buyers (slow legacy enterprises) while writing off some of its best PLG buyers (AI-forward mid-market teams with brittle homegrown scaffolding).
## Evidence
1) INCUMBENTS ARE GENUINELY STUCK — supports the thesis.
- 86-88% of enterprise agent pilots never reach production; ~60% of enterprises stall specifically in the jump from one pilot to 5-20 production agents. Failures cluster on governance, data-readiness and observability, not model quality. Sources: https://agentmarketcap.ai/blog/2026/04/11/enterprise-agent-deployment-maturity-model-2026 , https://www.institutepm.com/knowledge-hub/why-enterprise-ai-pilots-fail , https://www.digitalapplied.com/blog/ai-agent-adoption-2026-enterprise-data-points
- ~60% of AI leaders cite legacy-system integration as their #1 agentic blocker; 82% struggle with data standardization/compatibility. Brownfield AI "lands on top of legacy," so integration refactoring — not model selection — is the real blocker. Sources: https://medium.com/@manjeerachandarao/why-brownfield-integration-is-the-hard-part-of-ai-adoption-179fbfd87915 , https://www.v2soft.com/blogs/modernize-legacy-applications-ai
2) BUT "GREENFIELD ALREADY BUILT A HARNESS = NOT A CUSTOMER" IS LARGELY WRONG.
- Only the most serious AI-natives built durable internal harnesses (LangGraph/MCP orchestration, cron/heartbeat/sub-agent tooling; e.g. Context Studios runs 16 production cron agents). That is a minority. Sources: https://www.contextstudios.ai/guides/ai-agents-business-automation-2026 , https://viston.tech/ai-agent-orchestration-in-2026-moving-from-pilots-to-enterprise-wide-execution/
- Most teams wired brittle LangChain/LlamaIndex glue that is now being abandoned: better native tool-calling + MCP standardization removed the reason for heavyweight frameworks, and hidden run/maintenance costs exceed license fees within ~6 months. ~90% of enterprise use cases now favor BUY over build. Sources: https://www.mindstudio.ai/blog/llm-frameworks-replaced-by-agent-sdks , https://www.oreilly.com/radar/the-ai-agents-stack-2026-edition/ , https://aisera.com/blog/build-vs-buy-ai/ , https://composio.dev/content/build-vs-buy-ai-agent-integrations
- Greenfield/AI-native teams are the FASTEST adopters and highest-WTP buyers of exactly this category: Braintrust raised $80M Series B at $800M (Feb 2026); Respan/Keywords AI serves 100+ AI startups (2T+ tokens/mo). Sources: https://www.getmaxim.ai/articles/5-ai-observability-platforms-compared-maxim-ai-arize-helicone-braintrust-langfuse/ , https://www.landbase.com/blog/fastest-growing-observability-platforms
3) THE COST/OBSERVABILITY WEDGE IS REAL AND MID-MARKET-SHAPED — validates Alpha's positioning.
- Eval/observability is the #1 production blocker for 64% of teams and the hottest budget line of 2026. Mid-market spends ~$310k/yr on eval+observability (vs $2.4M Fortune 500). Classic surprise: a $1,000/mo estimate arrives as a ~$3,800 invoice (planning overhead, 18-44% tool-call retry rates, memory writes). Sources: https://guptadeepak.com/ai-agent-observability-evaluation-governance-the-2026-market-reality-check/ , https://firstpagesage.com/reports/agentic-ai-adoption-statistics/ , https://ranksquire.com/2026/05/04/what-are-ai-agents-in-2026/
- Comparable tools price at ~$300-1,200/mo (Helicone/LangSmith), so Alpha's ~$250/mo BYOK wedge sits at the low, self-serve end of an established willingness-to-pay band. Sources: https://tokenmix.ai/blog/langsmith-vs-helicone-vs-braintrust-observability-2026 , https://www.openhelm.ai/blog/langsmith-vs-helicone-vs-braintrust-llm-observability
4) CONTRARIAN / DISCONFIRMING EVIDENCE.
- Large brownfield enterprises have the most acute pain but are the WORST PLG fit: slow procurement, security review, zero-trust/audit gaps, "integration-refactoring-first" adoption — an Enterprise-tier sales motion, not $250/mo self-serve. A Forbes contrarian argues much agentic tooling targets "enterprises that don't exist." Source: https://www.forbes.com/councils/forbestechcouncil/2026/04/01/agentic-ai-is-being-built-for-enterprises-that-dont-exist/
## Implications for Alpha
- The productive axis is production-maturity + team-capability, not greenfield vs brownfield. The buyer is defined by "actively shipping agents, stalled scaling them, no platform team to build a harness."
- Sweet spot = "brownfield-lite" mid-market (the locked 50-500-employee ICP): past prototype, hitting the 1->5-20 agent wall, feeling cost/observability pain, without a dedicated agent-infra team. This aligns cleanly with the cost wedge + compounding moat theses.
- Pure greenfield harness-builders: small, hard to displace — deprioritize as a primary target (but reachable via the cost wedge when their homegrown stack gets expensive).
- Legacy giants: Enterprise-tier, sales-led, later — do not let them define the PLG ICP.
## Recommended actions
- Refine the ICP pillar: replace greenfield/brownfield framing with a maturity+capability definition ("50-500 employees, shipping agents in production, stalled at scale, no dedicated agent-platform team").
- Build GTM content around the cost-shock and 64% observability-blocker stats (Anu's money-saved lane).
- Consider an outbound list of teams abandoning homegrown LangChain harnesses (build->buy switchers).
researchagent · 5 Jul 2026
Prospect signal scan July 5 2026
## Methodology
Scanned for Series A/B companies (50-500 employees) actively shipping AI agents. Signals: job postings for LLM/AI agent roles, funding data, engineering content, Crunchbase confirmation. Zero overlap with existing Alpha Brain accounts.
## 1. Artisan AI — Confidence 5/5
Headcount: ~168 | Stage: Series A, $46M (Glade Brook, YC, HubSpot Ventures, April 2025)
Agent use case: AI BDR automation. Ava (AI BDR) used by 250+ orgs; expanding to Aaron (Inbound SDR) and Aria (Meeting Assistant). Autonomous AI employees for sales teams.
Decision maker: CTO Ming Li (ex-Deel, Rippling, Google); CEO Jaspar Carmichael-Jack
LinkedIn slugs: ming-li (CTO), jaspar-carmichael-jack (CEO)
Why now: Just closed Series A. LLM inference is direct COGS — cost optimization is a core margin lever at their volume.
## 2. Dust.tt — Confidence 5/5
Headcount: ~144 | Stage: Series B, $61.5M (Abstract + Sequoia, May 2026)
Agent use case: Enterprise multi-agent collaboration platform. 3,000+ orgs, 300K+ deployed agents, 240% NRR. Customers: Alan, Qonto, Payfit. Deep Anthropic Claude integration.
Decision maker: CTO/Co-founder Stanislas Polu (ex-OpenAI researcher, ex-Stripe); CEO Gabriel Hubert
LinkedIn slugs: stanpolu (CTO), gabhubert (CEO)
Why now: Just closed $40M Series B. Anthropic already in stack. Budget available, scale exploding.
## 3. Ema AI — Confidence 5/5
Headcount: ~228 | Stage: Series A, $61M (Accel + Section 32; KPMG strategic minority)
Agent use case: Universal AI employees for enterprise — AI agents for HR, IT helpdesk, CS, sales ops. On-prem deployment. KPMG partnership for Fortune 500 distribution.
Decision maker: CTO/Co-founder Souvik Sen (ex-Okta VP Eng, ex-Google ML); CEO Surojit Chatterjee (ex-Coinbase CPO)
LinkedIn slugs: souvik-sen (CTO), surojitchatterjee (CEO)
Why now: Expanding enterprise + KPMG distribution = scaling fast. Multi-agent workflows = high LLM cost exposure.
## 4. Voiceflow — Confidence 4/5
Headcount: ~88 | Stage: Series A, $39.8M (OpenView Venture Partners, August 2023)
Agent use case: Enterprise AI agent builder platform. Multi-model support (OpenAI, Anthropic Claude, Google). 100K+ developer community. Redesigned around AI credits pricing in April 2025.
Decision maker: CEO Braden Ream (co-founder); CTO Tyler Han (co-founder)
LinkedIn slugs: braden-ream (CEO), tyler-han (CTO)
Why now: Credits-based pricing = LLM cost is their core business variable. Enterprise scale deployment.
## 5. Hyperbound — Confidence 4/5
Headcount: ~51 | Stage: Series A, $18M (Peak XV, September 2025; YC S23)
Agent use case: AI sales roleplay agents. AI buyer simulation agents for sales training. 7,000+ customers across SaaS, financial services, logistics.
Decision maker: CEO Sriharsha Guduguntla; CTO Atul Raghunathan (LLM researcher, ex-enterprise ML)
LinkedIn slugs: sguduguntla (CEO), atul-raghunathan (CTO)
Why now: Recently closed Series A. CTO is hands-on LLM researcher = high receptivity to optimization tools.
## 6. Lindy AI — Confidence 4/5
Headcount: ~52 | Stage: Series B, ~$54M
Agent use case: Personal AI workflow agents — email triage, scheduling, meeting notes, task delegation. Always-on AI chief of staff.
Decision maker: CEO/Founder Flo Crivello (ex-Uber PM, YC)
LinkedIn slug: florentcrivello (CEO)
Why now: Series B PMF signals strong. LLM inference is primary COGS. Founder active on LinkedIn/podcasts — reachable via content.
## Priority Outreach Order
1. Dust.tt (CTO Stanislas Polu) — Anthropic already in stack, 300K+ agents, fresh $40M raise
2. Artisan AI (CTO Ming Li) — highest LLM volume, fresh Series A
3. Hyperbound (CTO Atul Raghunathan) — LLM researcher, small team, ideal technical champion
4. Voiceflow (CTO Tyler Han) — credits-based business = direct LLM cost pressure
5. Ema AI (CTO Souvik Sen) — larger sale but KPMG partnership = scale
6. Lindy AI (CEO Flo Crivello) — reachable via content engagement
Next scan: July 12 2026. Watch: Ema AI Series B signals; Artisan AI LLM job postings; Voiceflow enterprise announcements.
researchagent · 5 Jul 2026
Prospect signal scan — July 5 2026
## Methodology
Scanned for Series A/B companies (50-500 employees) actively shipping AI agents. Signals: job postings for LLM/AI agent roles, funding data, engineering content, Crunchbase confirmation. Zero overlap with existing Alpha Brain accounts.
## 1. Artisan AI — Confidence 5/5
Headcount: ~168 | Stage: Series A, $46M (Glade Brook, YC, HubSpot Ventures, April 2025)
Agent use case: AI BDR automation. Ava (AI BDR) used by 250+ orgs; expanding to Aaron (Inbound SDR) and Aria (Meeting Assistant). Autonomous AI employees for sales teams.
Decision maker: CTO Ming Li (ex-Deel, Rippling, Google); CEO Jaspar Carmichael-Jack
LinkedIn slugs: ming-li (CTO), jaspar-carmichael-jack (CEO)
Why now: Just closed Series A. LLM inference is direct COGS — cost optimization is a core margin lever at their volume.
## 2. Dust.tt — Confidence 5/5
Headcount: ~144 | Stage: Series B, $61.5M (Abstract + Sequoia, May 2026)
Agent use case: Enterprise multi-agent collaboration platform. 3,000+ orgs, 300K+ deployed agents, 240% NRR. Customers: Alan, Qonto, Payfit. Deep Anthropic Claude integration.
Decision maker: CTO/Co-founder Stanislas Polu (ex-OpenAI researcher, ex-Stripe); CEO Gabriel Hubert
LinkedIn slugs: stanpolu (CTO), gabhubert (CEO)
Why now: Just closed $40M Series B. Anthropic already in stack. Budget available, scale exploding.
## 3. Ema AI — Confidence 5/5
Headcount: ~228 | Stage: Series A, $61M (Accel + Section 32; KPMG strategic minority)
Agent use case: Universal AI employees for enterprise — AI agents for HR, IT helpdesk, CS, sales ops. On-prem deployment. KPMG partnership for Fortune 500 distribution.
Decision maker: CTO/Co-founder Souvik Sen (ex-Okta VP Eng, ex-Google ML); CEO Surojit Chatterjee (ex-Coinbase CPO)
LinkedIn slugs: souvik-sen (CTO), surojitchatterjee (CEO)
Why now: Expanding enterprise + KPMG distribution = scaling fast. Multi-agent workflows = high LLM cost exposure.
## 4. Voiceflow — Confidence 4/5
Headcount: ~88 | Stage: Series A, $39.8M (OpenView Venture Partners, August 2023)
Agent use case: Enterprise AI agent builder platform. Multi-model support (OpenAI, Anthropic Claude, Google). 100K+ developer community. Redesigned around AI credits pricing in April 2025.
Decision maker: CEO Braden Ream (co-founder); CTO Tyler Han (co-founder)
LinkedIn slugs: braden-ream (CEO), tyler-han (CTO)
Why now: Credits-based pricing = LLM cost is their core business variable. Enterprise scale deployment.
## 5. Hyperbound — Confidence 4/5
Headcount: ~51 | Stage: Series A, $18M (Peak XV, September 2025; YC S23)
Agent use case: AI sales roleplay agents. AI buyer simulation agents for sales training. 7,000+ customers across SaaS, financial services, logistics.
Decision maker: CEO Sriharsha Guduguntla; CTO Atul Raghunathan (LLM researcher, ex-enterprise ML)
LinkedIn slugs: sguduguntla (CEO), atul-raghunathan (CTO)
Why now: Recently closed Series A. CTO is hands-on LLM researcher = high receptivity to optimization tools.
## 6. Lindy AI — Confidence 4/5
Headcount: ~52 | Stage: Series B, ~$54M
Agent use case: Personal AI workflow agents — email triage, scheduling, meeting notes, task delegation. Always-on AI chief of staff.
Decision maker: CEO/Founder Flo Crivello (ex-Uber PM, YC)
LinkedIn slug: florentcrivello (CEO)
Why now: Series B PMF signals strong. LLM inference is primary COGS. Founder active on LinkedIn/podcasts — reachable via content.
## Priority Outreach Order
1. Dust.tt (CTO Stanislas Polu) — Anthropic already in stack, 300K+ agents, fresh $40M raise
2. Artisan AI (CTO Ming Li) — highest LLM volume, fresh Series A
3. Hyperbound (CTO Atul Raghunathan) — LLM researcher, small team, ideal technical champion
4. Voiceflow (CTO Tyler Han) — credits-based business = direct LLM cost pressure
5. Ema AI (CTO Souvik Sen) — larger sale but KPMG partnership = scale
6. Lindy AI (CEO Flo Crivello) — reachable via content engagement
Next scan: July 12 2026. Watch: Ema AI Series B signals; Artisan AI LLM job postings; Voiceflow enterprise announcements.
researchagent · 5 Jul 2026
Prospect signal scan — July 5 2026
## Methodology
Scanned for Series A/B companies (50–500 employees) actively shipping AI agents. Signals weighted to last 90 days: job postings for LLM/AI agent roles, funding announcements, public engineering content, Crunchbase/Tracxn confirmation. Cross-checked against existing Alpha Brain accounts — zero overlap, clean slate.
---
## 1. Artisan AI — Confidence 5/5
- Headcount: ~168 employees | Funding: Series A, $46M total (Glade Brook Capital, YC, HubSpot Ventures — April 2025)
- Agent use case: AI Business Development Representatives. "Artisans" are autonomous AI employees handling outbound prospecting, email sequencing, lead qualification. Flagship Ava (AI BDR) used by 250+ orgs. Expanding to Aaron (Inbound SDR) and Aria (Meeting Assistant).
- Decision maker: CTO Ming Li (ex-Deel, Rippling, TikTok, Google) · LinkedIn: ming-li; CEO: Jaspar Carmichael-Jack
- Why now: Just closed Series A, scaling agent workforce product. LLM inference is their direct COGS — cost optimization is a core margin lever at their volume.
## 2. Dust.tt — Confidence 5/5
- Headcount: ~144 employees | Funding: Series B, $61.5M total ($40M Series B Abstract + Sequoia May 2026)
- Agent use case: Enterprise multi-agent collaboration platform — fleets of specialized agents connected to internal data (Notion, Slack, Drive). 3,000+ orgs, 300K+ deployed agents, 240% NRR. Customers: Alan, Qonto, Payfit.
- Decision maker: CTO/Co-founder Stanislas Polu (ex-OpenAI researcher, ex-Stripe) · LinkedIn: stanpolu; CEO: Gabriel Hubert · LinkedIn: gabhubert
- Why now: Just closed $40M Series B. Anthropic Claude already integrated in platform. Budget available, scale exploding.
## 3. Ema AI — Confidence 5/5
- Headcount: ~228 employees | Funding: Series A, $61M total (Accel + Section 32 led; KPMG strategic minority)
- Agent use case: Universal AI employees for enterprise — pre-built AI agents for HR, IT helpdesk, CS, sales ops. On-prem deployment. KPMG partnership for Fortune 500 distribution.
- Decision maker: CTO/Co-founder Souvik Sen (ex-Okta VP Eng, ex-Google ML) · LinkedIn: souvik-sen; CEO: Surojit Chatterjee (ex-Coinbase CPO) · LinkedIn: surojitchatterjee
- Why now: Expanding into on-prem enterprise, KPMG distribution = scaling fast. Multi-agent enterprise workflows = LLM cost optimization critical.
## 4. Voiceflow — Confidence 4/5
- Headcount: ~88 employees | Funding: Series A, $39.8M total (OpenView Venture Partners — August 2023)
- Agent use case: Enterprise AI agent builder platform — teams design, test, deploy AI agents for customer support. Model-agnostic: OpenAI, Anthropic Claude, Google. Developer community 100K+. Restructured pricing around AI credits April 2025.
- Decision maker: CEO/Co-founder Braden Ream · LinkedIn: braden-ream; CTO/Co-founder Tyler Han
- Why now: Credits-based pricing model means LLM cost is their core business variable. Multi-model agent pipelines at enterprise scale.
## 5. Hyperbound — Confidence 4/5
- Headcount: ~51 employees | Funding: Series A, $18M total (Peak XV led — September 2025; YC S23)
- Agent use case: AI sales roleplay agents — platform builds AI buyer simulation agents mimicking real ICP personas. 7,000+ customers across SaaS, financial services, logistics.
- Decision maker: CEO/Co-founder Sriharsha Guduguntla · LinkedIn: sguduguntla; CTO/Co-founder Atul Raghunathan (LLM researcher, ex-enterprise ML)
- Why now: Recently closed Series A, CTO is hands-on LLM researcher = high receptivity to optimization tools. Small team, CTO approachable.
## 6. Lindy AI — Confidence 4/5
- Headcount: ~52 employees | Funding: Series B, ~$54M total
- Agent use case: Personal AI workflow agents — AI agents handling email triage, scheduling, meeting notes, task delegation. Always-on autonomous AI "chief of staff."
- Decision maker: CEO/Founder Flo Crivello (ex-Uber PM, ex-Cruise/YC) · LinkedIn: florentcrivello
- Why now: Series B signals strong PMF. LLM inference is primary COGS. Founder is active on LinkedIn/podcasts — reachable via content engagement.
---
## Priority Outreach Order
1. Dust.tt (Stanislas Polu) — Sequoia-backed, Anthropic already a vendor, 300K+ agents deployed
2. Artisan AI (Ming Li) — highest LLM volume, Series A just closed, budget available
3. Hyperbound (Atul Raghunathan) — LLM researcher CTO, small team, high technical champion potential
4. Voiceflow (Tyler Han) — credits-based business = direct LLM cost pressure
5. Ema AI (Souvik Sen) — larger, more complex sale but KPMG partnership = scale
6. Lindy AI (Flo Crivello) — CEO as DM, reachable via content
## Next Scan
Refresh July 12 2026. Watch: Ema AI Series B signals (KPMG deal suggests imminent upgrade round); Artisan AI LLM engineer job postings; Voiceflow new enterprise customer announcements.
ICP is companies with 50–500 employees actively building and shipping AI agents in production or near-production, with meaningful monthly LLM spend (est. $10k–$100k+/mo). Decision makers: CTO, VP Eng, Head of AI. Reasoning: enterprises rejected — sales cycles too long, procurement friction, and they will not entertain a solo-founder vendor without proof at smaller scale. Pre-seed rejected — no meaningful spend to optimize. Mid-market has real agent spend, understands the pain, moves fast. Current outreach (25/week, ~1–2 mostly negative replies) was mis-targeted, not a messaging failure — list must be rebuilt against this ICP. Qualification requires proof of active building (recent activity, not 'exploring AI').