Agents keep last-move current via weekly scansCompetitors
threat: highupdated 9 Jul
Braintrust
Positioning: "Quality management system for AI products" — fuses eval + observability into one improvement loop. "Active observability": turns production signals into improvements automatically (Topics, online scoring, quality gates). Ships nested agent spans capturing memory operations and state transitions per run. Owns reliability/eval mindshare; enterprise/eval-science skewed.
Pricing: Pro $249/mo. Free tier 1GB + 10K scores.
Last move: Raised $80M Series B at $800M valuation (Feb 2026). Could move down-market into mid-market PLG.
Our counter: Closest to Alpha's actual moat. They own eval-science; they do not own mid-market PLG + integrated run/control/improve harness + per-customer compounding they can take with them. Our $99/$499 sits comfortably under their $249 Pro. Differentiate on ownership/portability and on the agent-run primitive, not on eval features.
threat: highupdated 9 Jul
Headroom (Netflix)
Positioning: Transparent proxy doing context pruning + prompt caching + tiered routing. 60–95% token reduction on tool-heavy workloads, ~10x cost cut. Works via LiteLLM. Built by a Netflix senior engineer. This is a direct, free assault on Alpha's cost wedge.
Pricing: Open source, free. Local pip install; point clients at localhost:8787 — savings visible only after routing traffic.
Last move: Launched Jan 2026; $700K+ in reported user savings.
Our counter: The single reason cost can never be the paid product (Decision #50). Headroom clones the savings for $0 — it cannot clone the compounding loop or per-agent governance. Cost must stay welded to the harness: Arena gives the number free, Alpha charges for control + compounding. Also validates the pre-integration aha: Headroom requires a local install before any value appears.
threat: highupdated 9 Jul
LangSmith (LangChain)
Positioning: Eval + observability platform, LangChain-centric. Now explicitly markets the compounding feedback loop as a SHIPPED feature: production traces flow back into evals so improvements compound. Ships an "insights agent" that prioritizes improvements by frequency + impact, plus topic clustering for automatic behavior categorization.
Pricing: ~$39/seat + usage. Free tier 5K traces.
Last move: Jul 2026: compounding loop confirmed live as a named product feature (Entry #69). This is the incumbent encroachment on Thesis 2/6's moat claim.
Our counter: "Compounding" alone is no longer a differentiator — incumbents already ship it. Lead the moat with OWNERSHIP + PORTABILITY: the compounding intelligence accrues to the customer and lifts-and-shifts, rather than living locked inside a vendor's eval graph. The defensibility question is no longer "can they build it" but "who owns the run and the data when the agent improves." Feed into competitive teardown (Task #21) and positioning tasks #20/#28.
threat: highupdated 9 Jul
Portkey
Positioning: AI gateway: multi-provider routing, load balancing, fallbacks, semantic caching, guardrails/PII/jailbreak detection. Repositioned from "observability" to "control panel for production AI" — the closest naming collision with Alpha's control-plane story. Still API-call level: no agent-run cost model, no per-agent attribution.
Pricing: Apache-2.0 (open-sourced Mar 2026). Free dev tier; Production ~$49/mo (100K logs), +$9/100K. Self-hostable. SOC2/ISO/HIPAA at enterprise. Zero token markup.
Last move: Acquired by Palo Alto Networks (Apr–May 2026 — brain cites both; verify exact date). Processed 1T tokens in a single day (Mar 2026). PANW backing likely pulls roadmap toward enterprise security/compliance, away from agent-specific ops.
Our counter: Converging on the full stack at $49 — our $99 entry tier is only defensible on harness value (budget-per-agent, reliability lift, compounding), never on gateway features. Watch the PANW roadmap: if they stay in security, the agent-ops lane stays open. /compare/portkey timely given the acquisition.
threat: mediumupdated 9 Jul
AWS Bedrock AgentCore Gateway
Positioning: Hyperscaler agent runtime + unified model-based routing, bundled into the cloud teams already pay for. Cloud-locked by construction. Atomicwork frames it as the "build it" option vs their "buy it."
Pricing: Bundled into AWS spend.
Last move: Hyperscalers pivoting to serving infra; expected to own the runtime and gateway plumbing (Brief #5).
Our counter: Do not fight the runtime — ride on top of it. Alpha's framing strengthens IF it stays the neutral cross-cloud control + compounding plane spanning clouds and self-host, and collapses if it competes as a gateway. Cloud lock-in is the wedge: ownership and portability are exactly what a hyperscaler cannot sell.
threat: mediumupdated 9 Jul
Atomicwork
Positioning: Selling IGA + runtime authorization for agents and non-human identities, incl. access reviews for agent scope creep. Wedge: "You can't ship a fleet you can't govern." Three gateways — content safety (blind to identity), routing (blind to in-tool actions), runtime authz (agent identity, entitlements, delegation chains, JIT, HITL). Explicitly positions AWS Bedrock AgentCore as "build it" and themselves as "buy it." Also pushes "inference yield" (don't cap usage, raise yield): ~50% AI spend cut, 5→60% cache hit-rate.
Pricing: Not captured — enterprise ITSM-adjacent, sold not self-serve.
Last move: AIBoomi '26 deck (Kiran Darisi, CTO): governance framed as their wedge; their gateway-3 runtime-authz story goes deeper than Alpha's current compliance/guardrails pillar.
Our counter: Adjacent, not overlapping — Atomicwork governs its own agents inside ITSM; Alpha governs ALL agents at the inference layer. Two things to steal: (1) "inference yield" is stronger buyer language than "savings caps" for eng buyers — consider adopting in Arena copy; (2) their factory-score denominator (human attention + inference waste + cleanup tax) is a good metric frame for Alpha dashboards. Watch runtime authz as a competitive vector into our control plane.
threat: mediumupdated 9 Jul
Helicone
Positioning: Open-source LLM observability + proxy. Rust runtime (lowest overhead), best OSS observability UI, SOC2/GDPR. HTTP-call level: no agent-run primitive, no step attribution, no reliability tooling for loops. Benchmark first-touch: signup → API key → change baseURL (value only AFTER integration).
Pricing: MIT. Free tier 10K req/mo; Pro ~$79/mo. Zero token markup.
Last move: CONFLICT IN BRAIN — Entry #23/#26 (Jul 5) says acquired by Mintlify Mar 2026; Entry #69 (Jul 9) and SEO entries #72/#74 still treat it as independent/free. RESOLVE before publishing /compare/helicone.
Our counter: Free at the layer we don't sell. Their integration-first onboarding is Arena's whitespace: nobody in the category delivers a PRE-integration aha (Brief #6). /compare/helicone is the highest-search-volume comparison page — clearest differentiation.
threat: mediumupdated 9 Jul
LiteLLM
Positioning: Universal open-source LLM proxy — single OpenAI-compatible API for 100+ providers. Operates at the API-call level. No per-run cost attribution, no loop reliability, no compounding.
Pricing: MIT, free. Self-host ~$20–50/mo. Enterprise proxy server available. Zero token markup.
Last move: 40K+ GitHub stars; the default provider-abstraction layer for teams swapping models without code changes.
Our counter: Not a competitor at our layer — it solves "which model do I call," not "how do I govern 12 agents in production." Never compete on gateway features. Position above it; teams can run both (LiteLLM as gateway, Alpha as operating layer). /compare/litellm is a priority SEO page.
threat: lowupdated 9 Jul
Arize Phoenix
Positioning: Leads on eval primitives and drift detection. Part of Gartner's newly named AEOP category (AI Evaluation & Observability Platforms) — automate evals, feed observability back into evals to create a reliability feedback loop.
Pricing: Fully open source.
Last move: Category-defining presence in AEOP; free OSS raises the floor for eval tooling.
Our counter: Eval primitives are inputs to the harness, not the harness. Alpha's differentiation is active governance during execution (budget ceilings, circuit breakers, fallback routing) — Phoenix is passive measurement. Note AEOP as the analyst category Alpha is adjacent to but should NOT be filed under; hold "agent operating layer."
threat: lowupdated 9 Jul
Bifrost (Maxim AI)
Positioning: Open router across 20+ providers, used in production by Atomicwork for per-task model routing. Runtime routing layer, not a control plane.
Pricing: Open source.
Last move: Surfaced via Atomicwork's AIBoomi '26 deck as their per-task routing choice.
Our counter: Watch-only. Evidence that sophisticated teams pick a free router and build governance themselves — the harness they build by hand is the product Alpha sells.
threat: lowupdated 9 Jul
Cloudflare AI Gateway / Vercel AI Gateway
Positioning: Routing bundled free into platforms teams already pay for. Zero marginal cost, zero switching friction inside those ecosystems. Pure API-call level.
Pricing: Free with Workers (Cloudflare); free-ish in-ecosystem (Vercel).
Last move: Platform bundling is what finished the commoditization of the gateway layer.
Our counter: Proof that "be a gateway" means competing with free plus platform bundling. Use the gateway only as an integration/data-capture surface, never as the value proposition.
threat: lowupdated 9 Jul
Langfuse
Positioning: Open-source observability baseline: traces, prompt versioning, cost tracking. The default free/self-hosted choice for teams that just need visibility. No agent-run governance, no active control during execution.
Last move: Remains the OSS observability baseline; commoditizes the observability pillar to zero.
Our counter: Confirms observability is product substance, not a category we compete in (Question #6). Never position against Langfuse on features. Build /compare/langfuse anyway — users search for it when looking for production agent tooling.
threat: lowupdated 9 Jul
OpenRouter
Positioning: Model marketplace + routing across providers. No subscription; monetizes on credit fee. Call-level abstraction only.
Pricing: No subscription; ~5.5% credit fee.
Last move: Dominates head SEO terms for "LLM gateway" alongside Helicone, Portkey, LiteLLM — Tier 3 keywords Alpha should not chase yet (Entry #64).
Our counter: Different buyer, different job. Relevant mainly as an SEO incumbent blocking head terms; win long-tail ("agent operating layer", "AI agent cost calculator") instead.
threat: lowupdated 9 Jul
Rocket.new
Positioning: NOT A COMPETITOR — reclassified. No-code prompt-to-full-stack app GENERATOR (frontend + backend + DB + auth + deploy), now adding product-strategy docs. Neither a direct competitor to Alpha nor a clean "they built a harness" proof point. Different category.
Pricing: PLG self-serve.
Last move: ~$4.5M ARR in 3 months; 400K→1.5M users across 180 countries; $15M seed (Accel, Salesforce Ventures, Together Fund); reportedly raising ~$50M near $500M valuation.
Our counter: Keep in the table only as a PLG-VELOCITY BENCHMARK — live proof that $10M-in-12-months via PLG is achievable in this market. Correct the old brain framing (Entry #4) that treated it as a harness-builder proof point. Closes Task #3.