pitch.api.insure
Agents can't settle insurance. They can call it.
↓ scroll · arrow keys
The primary caller is an agent. api.insure is the demand rail of the cell — the single licensed operating entity behind api.insure and gigs.claims — and an agent that conforms to the protocol can discover, call, and settle against these capabilities:
| capability | tier | typed state |
|---|---|---|
| FNOL intake (first notice of loss), file assembly, coverage analysis, estimate review, status, UCSPA clock reads (the statutory claims-handling clock) | Open — agent-native, code-priced | executes and returns; everything the state leaves open |
| adjust, settle, decide | Reserved — a statute names a person | PENDING_ADJUSTER until a licensed adjuster completes the act |
| quote, bind | ROADMAP — producer-reserved, behind the cell's owned, appointed Agency | typed in the catalog now; labeled ROADMAP everywhere until live |
A reserved call is specified never to fail: it returns a typed PENDING_ADJUSTER state, and a licensed adjuster on the network — acting under Delegated Authority granted to the cell's own TPA — reviews the prepared file and, in their own licensed judgment, approves, denies with reasons, or sends back. The surface is built to be evaluated by a machine — typed states, deterministic meter events, webhook contracts, AXP (Agent eXperience Protocol) conformance — and every caller binds a principal and a hard spend cap, computable before the first call because every fee is flat.
PENDING_ADJUSTER — a claim waiting on a license is typed as waiting — never an untyped queue, never "processing."
You get three things no TPA portal offers: a capability contract your agent can read, a posted rate card your finance team can cap, and a conformance posture your compliance lead can verify. No appointment paperwork per caller, no negotiated program per account.
Per-act fees are provisional targets until real files price them, and the structural guarantees above — including "cannot express" — are design commitments of a billing system that is not yet live; they post as fact when the Offer schema is published and the first files have run against it. The demand-side card is never quoted without its supply-side twin: the adjuster's File Fee at gigs.claims is flat, fixed at post time, and decision-independent. The spread between the carrier-side Administration Fee and that File Fee is the cell's margin.
Carrier / MGA → Claims-Handling Agreement → the cell's TPA (Delegated Authority, Authority Limits, Merchant of Record) → licensed adjuster on the network (own license, own judgment, E&O in force naming them).
The cell holds the licenses so the caller never touches the regulated surface: the TPA's entity adjuster licenses, the Claims-Handling Agreements, the E&O program, and the producer Agency (with its designated responsible licensed producer) behind the ROADMAP bind surface. Held in the cell, renewed in the cell, audited in the cell — and in licensure now, not asserted as live: the pending claims on this slide and the status slide say exactly where it stands. The eligibility lattice — the per-state, per-line license computation behind every routed file — is built to compute that stack; it is never eyeballed.
P&C claims administration is regulated as adjusting: individual adjuster licenses per state and line of authority, plus business-entity adjuster or adjusting-firm licenses where states have them — and roughly a third of states do not license adjusters at all. The NAIC State Licensing Handbook is the map the cell's license stack is built against.
Texas grants Designated Home State adjuster licenses to residents of states that do not license adjusters, and reciprocity for those adjusters hangs off that anchor.
The per-state survey of entity adjuster licenses, TPA/administrator statutes that sweep in P&C, and UCSPA timer tables is contracted regulatory spend, priority one. No state/line pair ships before its row is ratified.
A signature that carries risk is priced as risk, not as hours. What a small carrier buys from a licensed adjuster is a decision with E&O behind it, priced on the liability it carries — so agent margin collapse buys nothing against it, because making the labor cheap does not touch what the signature costs. That objection ends most "AI eats claims" pitches, and it is stated here first because this cell is built on the fact, not against it.
What a tenant buys from api.insure is not routed labor — it is a decision with liability carried behind it:
The allocation of bad-faith exposure between the TPA's E&O and carrier indemnification is a Claims-Handling Agreement schedule, not a settled fact, and it gates the first live claim. It is stated here as pending because a deck that hides its liability question is exhibit A in the suit about it.
We have already paid for the lesson. At Rocket Auto, what ran unstaffed was sales, BDC and desking. What was never unstaffed: the dealer licence, titling, DMV processing and F&I compliance — those humans were Rocket's, not ours. When the relationship ended, the regulated supply left with it, and an otherwise-working business ended. Borrowed regulated supply is the single point of failure, and we do not borrow it twice.
rocketauto.com now redirects to a corporate index page.
api.insure and gigs.claims are therefore one cell: the demand rail and the supply door are two segments of one A2H2A path — agent to licensed human to agent — held together with the licenses, and neither is a counterparty to the other.
Primary motion is B2A2B — a business system or its agent calls the rail directly. Also B2H2A, where a human authoriser deploys agent work containing the insurance step.
Being discoverable and callable by an agent is the distribution channel, and the surface is built to resolve there: insurance resolving to api.insure in the agent-discovery layer, the way automotive resolves to auto.dev, is the design target — not yet the verified state.
Discovery-layer resolution is claimed only when it is testable: this flips posted when the MCP tools and conformance surface are callable on the open domain.
Everywhere else in the estate, functions migrate Human → Agentic → Generative → Code, and each migration is structural margin expansion. Here the migration runs until it hits the reserved acts and stops: intake, assembly, analysis and monitoring migrate; the decision does not, because the statute names a person. Cost of a reserved call is therefore dominated by payment to the licensed human who performs it — a take on the flow, not software margin.
Entity formation, TPA and Agency licensing, and the first Claims-Handling Agreement are in progress. Until they close, this deck describes a designed cell, not an operating one — deliberately, in the open.
Founding team spans the demand rail, the substrate, and the regulated edge; a third co-founder currently leads AI at a public insurtech and joins at close.
Get early access — as a design partner, not a mailing-list entry. Claims is first, and the first cohort is scoped to the first Claims-Handling Agreement: design partners shape that CHA's capability set, lock the rate card at post-time fees for their first files, and receive the first typed response on a real claim file — not a demo. The cohort closes when the first CHA signs.