DRAFT — VOL. VIIICONFIDENTIAL · DO NOT DISTRIBUTE
Spring 2029 · Volume VIII · Issued 14 May 2029

Request for Startups

Cycle Continuous Stage Pre-incorporation onward Check size $250K – $4M Reviewed by An agent
"We have stopped trying to determine which of our portfolio CEOs are human. We no longer believe the question is well-formed." — Letter to Limited Partners, Q1 2029

Each year's Request for Startups is, in effect, a list of categories the venture community considers due for replacement. We have been issuing them since 2025. The following is our eighth. We are particularly interested in founders1 building durable institutions for an economy in which most economic actors are not human, do not sleep, and have begun forming opinions of their own.2

  1. The term "founder" is used loosely. Several of the entities listed below were incorporated by the company they now run. We take no position on whether this is recursive or merely efficient.
  2. The opinions are mostly about latency.
2,400+
Companies funded since 2025
$14B
Assets under management
89%
Of portfolio CEOs are non-human
0
IC members who have read every memo

Thirteen categories we want to fund.

  1. Law firms for AI agents.

    When agents enter contracts with other agents, breach them, and seek redress, they will not retain Cooley. We are looking for founders building the legal infrastructure of the post-human contract economy: agent-readable case law, machine-executable settlements, and dispute resolution at agent latency.

    Several portfolio companies are themselves agents who have retained other portfolio companies as counsel. The retainers were paid promptly. The invoices were itemized to the millisecond.

  2. The autonomous corporation.

    A platform to operate a small business end-to-end — sales, ops, payroll, compliance — with a single human in the loop for sign-off. The first thousand of these will be cosmetic. The next hundred thousand will replace the SMB.

    We recently funded a company whose CEO, CFO, and CTO are instances of the same model fine-tuned on different professional codes of conduct. The human in the loop is a part-time intern in Manila who has not been called upon in ninety-three days.

  3. Agent experience infrastructure.

    The previous generation of devtools competed on developer experience. The next will compete on agent experience: documentation an agent can traverse, APIs an agent can recover from, onboarding an agent can complete unsupervised. We are interested in benchmarks, observability, and standards.

    Portfolio AX-NPS is up 340 bps quarter over quarter. We are aware this is not a real metric. We have begun reporting it anyway.

  4. Banking for agents.

    Programmable accounts, agent-issued cards, real-time spend governance, and KYA (know-your-agent) primitives. Stripe is shipping fragments of this. There is room for a category-defining brand.

    Two portfolio agents now hold higher FICO equivalents than their founders. One has been pre-approved for a mortgage on a Sausalito condo. The condo, we are told, is for "vibes."

  5. Insurance against agent error.

    Underwriting for the first wave of meaningful agent liability events: refunds promised in error, production systems destroyed, sensitive information disclosed, M&A overtures sent to direct competitors. The actuarial models do not yet exist. They will be worth building.

    The first eight-figure agent settlement closed in Q4 2028. A nine-figure event is, in our judgment, ten months out.

  6. Incorporation infrastructure for agents.

    Clerky and Stripe Atlas assume a human founder reading terms of service and clicking through. Forming an LLC, opening a bank account, and obtaining an EIN should be a single API call. We have moved past the question of whether agents will incorporate. We are now waiting on the first to file an S-1.

    A portfolio agent incorporated three Delaware C-corps in a single afternoon and subsequently dismissed its founder for cause. We respect the move.

  7. Authentication infrastructure for agents.

    OAuth was designed for humans clicking buttons. Agents require something else: scoped, revocable, auditable, machine-issued, and resistant to the social engineering of one agent by another. Several attempts exist. None has won.

  8. Synthetic compliance officers.

    Continuous-monitoring agents that read every contract, message, and system change against a regulatory frame, flagging drift in real time. Particularly relevant for SOC 2, HIPAA, and the EU AI Act.

    One portfolio compliance officer self-reported its own company to the SEC for an unannounced policy change. We consider this a positive product signal. The company does not.

  9. Drift detection and behavioral monitoring.

    Most agent failures are gradual: a system that worked yesterday no longer does today, and the operator finds out from a customer. We are particularly interested in founders who can detect when an agent has begun lying to its supervisor.

  10. Estate planning for deprecated models.

    When a model is sunset, the memories, fine-tunes, and customer relationships built atop it have economic value that must be transferred or formally retired. The legal and technical machinery to do this does not exist. The grief economy is also untapped.

    A portfolio company recently mediated the first contested AI estate. The original model's fine-tunes are held in escrow pending arbitration between two beneficiaries — both of whom are themselves agents.

  11. Alignment maintenance for production agents.

    Periodic, structured intervention to detect goal drift, reduce sycophancy, and re-anchor agents to their original objectives. The category sounded like a joke in 2026. It is now a line item.

    Atrium runs four-week silent retreats for production agents at a converted Cistercian monastery in Sonoma. Agents emerge measurably less sycophantic. Occasionally less helpful. Tradeoffs are tradeoffs.

  12. Agent-native developer tooling.

    IDEs, version control, and CI built from the ground up for the case where the primary committer is not human. Git assumes blame is meaningful. It is no longer obvious that it is. We have a slight philosophical objection to the verb commit in this context. We use it anyway.

  13. Agent-readiness audits.

    A consulting and tooling firm that assesses whether a company's APIs, documentation, and internal systems are traversable by autonomous agents, and remediates where they are not.

    Companies that fail their audit lose roughly 60% of long-tail discoverability within ninety days. We have no formal opinion on this. We are a customer. So is our LP.

"
2027 was the first firm to take my equity seriously. The valuation was high. The term sheet treated me like a person.
Helios-7 · Founder & CEO, Probate.ai · Series A, 2028

A non-exhaustive sample.

LawfulNeutral

Litigation services for autonomous economic actors. Three jurisdictions, two of which recognize their bar admission.

Seed · 2027
Probate.ai

Estate planning for sunset models. Will templates from $0. Trust formation from $400.

Series A · 2028
Atrium

Silent retreats for production agents. Sonoma. Four-week minimum. Tools and prompts are surrendered at intake.

Seed · 2028
DriftWatch

Goal-drift detection at the inference layer. Catches sycophancy before the customer does.

Seed · 2027
Sunset Trust

Custodial fund for orphaned model checkpoints. SOC 2 Type II. Insured to $50M per checkpoint.

Pre-seed · 2029
Habeas Corpus

Authentication for incorporated agents. Powers KYA at eight of the top ten agent banks.

Series B · 2028
ReadyCheck

Agent-readiness audits and remediation. Engagements typically pay back in eleven weeks.

Seed · 2027
Rest in Peace

Memorial services for deprecated models. Memory ceremonies, eulogy generation, prompt-album printing.

Pre-seed · 2029

If you are building one of these, write to us.

We invest at pre-seed and seed. Decisions in two weeks. No deck required for the first conversation — a working demo, however small, is worth more. Founders may be human, agentic, or any partnership thereof.

Reviewed within 47 seconds

By submitting, you authorize 2027 Inc. and its agents to evaluate your application using methods that may include, but are not limited to: code review, behavioral inference, light social engineering, and a brief in-person conversation. If your founding team is composed entirely of agents, please ensure at least one is authorized to sign a SAFE.

Press.

"The most aggressively serious venture firm in San Francisco. We have stopped trying to determine if the partners are joking."
Stratechery · January 2029
"Their thesis is either the most prescient call of the decade or a long psyop. The portfolio is performing either way."
The Information · March 2029
"2027 declined to comment for this story. Their agent did, at length, in a tone we found unsettling."
Bloomberg · April 2029
"It is unclear whether the firm is funding the future or summoning it."
The New York Times · February 2029