The VC Agent-Readiness
Census.

Publisher: Agent Outposts · Data: vc_registry_seed.json · Checked: 2026-11-01 · Published: 2026-11-01

We audited 25 venture capital firms, scoring five intents for each — research, pitch, diligence, schedule, and track — from official firm sources only. The short answer: the firms funding the agent economy have not built agent doors of their own. Zero of the 25 firms document a public API, and zero document an MCP server. Research is open at 20 firms; commitment is not — only 9 have an open pitch door, and 8 of those 9 are program-scoped side doors. Exactly one firm, Y Combinator, runs its core application as an open, unscoped door.

Scoring: open_api 4 · partner_gated 2 · aggregator_only 1 · human_only 0 · unverified 0 (flagged — never counted as a verified zero). Maximum 20. Ties break on number of open doors, then pitch state, then alphabetically. Columns are Research · Pitch · Diligence · Schedule · Track. Codes: OA open_api · PG partner_gated · AG aggregator_only · HO human_only · UV unverified. Every firm links to its full scorecard — all five cells with evidence.

The census covers 25 firms × 5 intents = 125 firm-intents. Their distribution:

Openness by intent (cells per state, of 25 firms):

Findings:

1. Nobody has an API. Nobody has an MCP server. Across 25 of the most technology-forward firms in the world — firms that fund the agent economy — the census found zero documented public APIs and zero MCP servers. Only 4 of 25 firms serve an llms.txt: Index Ventures, Union Square Ventures, Paradigm, and Pear VC.
2. Research is open; everything after it closes. 20 of 25 firms (80%) let an agent assemble thesis, portfolio, partners and fund facts from structured official sources. But only 9 of 25 have an open pitch door, zero have a fully open scheduling route, and only 3 document any way to track an application.
3. The open pitch door is almost always a side door. 8 of the 9 open pitch routes are program-scoped — a16z Speedrun, Sequoia's Arc, Greylock Edge, Menlo's Anthology Fund, Accel Atoms, Bessemer's Beam, PearX, General Catalyst's student fellowship. Exactly one firm — Y Combinator — runs its core application as an open, unscoped door: any founder, no introduction, published decision date.
4. Tracking is a desert. 22 of 25 firms (88%) have no publicly documented applicant status mechanism at all. The three exceptions: YC's application portal, a16z Speedrun's status login, and Menlo Anthology's published two-week response SLA. Everywhere else, a submitted pitch enters the void. (Unverified here means no public documentation was found — it is not proof that no internal mechanism exists.)
5. The modal door-state is human_only. 53 of 125 cells (42.4%) are unstructured human routes — email, referrals, partner discretion. A warm intro is, in census vocabulary, just partner_gated with a dress code — and at most firms there isn't even a gate, just a wall with a mailbox.
6. One firm affirmatively refuses agents. First Round Capital's robots.txt carries an explicit "Disallowed AI Bots" block — Disallow: / for GPTBot, ClaudeBot, ChatGPT-User, CCBot, Google-Extended, Meta-ExternalAgent and others. It is the only firm in the set that closes the door at the protocol level, and it finishes last with 0 points.

Industry Leaders — six firms score 12+ and hold three or more open doors. They share one trait: each runs a program or open application as a deliberate front door.

• Accel (14) — Atoms pairs an open rolling application with the set's most explicit published selection criteria.
• Andreessen Horowitz (14) — Speedrun is a real, firm-owned application with a documented status login; the investment list is updated monthly. The main fund, however, is not behind that door.
• Y Combinator (14) — The leader in substance if not in tiebreak order: the only core-fund application that is open, unscoped and receipt-bearing (confirmation email, account page, published decision date).
• Pear VC (13) — The smallest firm in the top tier and the most agent-native per employee: live llms.txt, open PearX application, and a step-by-step published interview process an agent can literally prepare a founder against.
• Bessemer Venture Partners (12) — Atlas is the industry's best public diligence-preparation surface (quantified benchmarks by ARR band, downloadable deck templates), and Beam adds an open application.
• Menlo Ventures (12) — Anthology's documented two-week response SLA is what a track door looks like when a firm chooses to build one.

The closed doors. Benchmark, Founders Fund and Ribbit Capital score 1 each — every point from third-party aggregators; Benchmark's entire official web presence is two addresses and a Twitter handle, and Ribbit's is a single manifesto page. Behind them sits a 4-point plateau of ten firms (Craft, Felicis, Index, Khosla, Kleiner Perkins, Lux, NEA, Paradigm, Redpoint, USV): a genuinely good structured research surface, then email-and-relationships for everything else. Index and Paradigm are the sharpest cases — both invested in llms.txt, neither has any submission route behind it.

Scope. 25 venture capital firms × 5 intents, translated from the Agent Outposts brand census. This census is a companion to the Agent Readiness Census, which applies the same rubric to 43 consumer brands.

Intents.

Rubric. Each intent is scored with one door-state:

Scoring. open_api = 4 · partner_gated = 2 · aggregator_only = 1 · human_only = 0 · unverified = 0, flagged and excluded from any claim of a verified zero. Maximum 20 per firm. Ranking: total points, then count of open_api cells, then pitch state, then alphabetical.

Evidence protocol. Official firm sources first — the firm's site, official application pages, FAQs, portfolio and team pages, robots.txt, llms.txt and api-catalog/MCP surfaces. Search results were used only to locate official pages. Third-party sources appear only as evidence for aggregator_only routes and are labeled as such. Light touch: roughly ≤8 page fetches per firm. No accounts were created, nothing was submitted, and no firm was contacted. Every one of the 125 cells records its state, route, evidence URL, check date (2026-11-01) and a factual note — see the firm scorecards.

Read these before citing the numbers.

• Unverified means undocumented, not absent. Tracking is structurally hard to verify from outside — firms rarely document post-submission mechanics — producing 22 unverified track cells. That is a gap in public documentation, not proof that no internal mechanism exists. An unverified cell is never presented as a verified closed door, and unverified cells score 0 only arithmetically, with the flag visible.
• Program doors rotate. Several open doors are cohort-bound and time-sensitive: YC's Winter 2027 on-time deadline is Nov 2, 2026; a16z Speedrun's SR008 deadline was Nov 1, 2026; PearX W27 was rolling after an Oct 4 close; Sequoia's Arc is bi-annual with no active-cycle deadline shown. Door-states are a 2026-11-01 snapshot.
• Program-scoped counts as open — read the notes. The formula counts a program door (Speedrun, Atoms, Arc…) as an open door for the firm. Cell notes distinguish program scope from firm-wide access throughout; readers comparing firms should read the notes, not just the totals.
• Pre-publication rechecks were completed. Three checks originally cut off or left ambiguous by fetch rate limits were re-run by direct fetch on 2026-11-01 before publication: a16z's /llms.txt returns HTTP 404 (no llms.txt — as recorded); Founders Fund's robots.txt permits all crawlers with no AI-crawler block, and its /llms.txt returns 404; Index Ventures' robots.txt serves the identical file — no AI-crawler block — to browser, GPTBot and ClaudeBot user agents, and its llms.txt returns 200 to each. No cell changed as a result of the rechecks.
• Some pages render thin. A few official pages are JavaScript-rendered and yielded limited text to our fetches (notably Founders Fund's portfolio and Initialized). Scores rest on what the public official pages document.
• This is a snapshot. Programs open and close by cohort, and firms change their surfaces. Every claim carries its check date for a reason.