kaedax
← work

An embedded-insurance API and partner console for a gig-economy insurer, taking quote-to-bind from minutes to seconds.

ANCHOR underwrites short-duration insurance for gig workers — ride-hail drivers, delivery couriers, last-mile logistics. They needed an embeddable quote-to-bind API that platforms could integrate in an afternoon, plus a partner console for the platforms themselves.

stack → Node + Fastify Postgres + RLS OpenAPI Redis Streams Datadog OAuth2 + JWT IRDAI-aware controls
PLATFORM A gig-worker app PLATFORM B gig-worker app PLATFORM C gig-worker app api.anchor.com 18 ENDPOINTS POST /quote p95 84ms POST /bind p95 112ms GET /policies/:id p95 21ms POST /policies/:id/extend p95 92ms POST /webhooks/event p95 — requested quoted bound active ended DROP-IN-COMPATIBLE WITH 2 / 3 INCUMBENT CONTRACTS · 224 CONTRACT TESTS ▷ CASE · ANCHOR · BLUEPRINT INSURANCE · EMBEDDED
SCHEMATIC An abstract view of the ANCHOR engagement — not a literal product screenshot. Built to communicate engineering shape, not surface design.

outcomes

01

12 min

Partner onboarding · signup → quote

02

1.4s

Median quote-to-bind

03

1k rps

Sustained load on quote engine

04

T+712h

First pilot live

[ §01 ] the cycle

How 720 hours
actually ran.

  1. Day 01 — 04

    API contract + partner workflows

    scope.agent reverse-engineered the API contracts of three competing embedded insurers, then drafted ANCHOR's contract to be drop-in-replaceable from any of them. Partner console workflows mapped from interviews with three pilot platforms.

    openapi.yaml partner journeys auth model
  2. Day 05 — 19

    API + console build

    build.agent shipped the quote engine, the bind workflow, the policy lifecycle, and the partner console in fifteen days. qa.agent generated 220+ contract tests from the OpenAPI spec — the API was conformant before it had clients.

    49 PRs contract tests × 224 load tested to 1k rps
  3. Day 20 — 26

    Partner SDKs + sandbox

    TypeScript and Python SDKs generated from OpenAPI, a public sandbox environment with realistic mock data, partner onboarding flow that takes a new platform from signup to first quote in 12 minutes.

    ts sdk py sdk sandbox.anchor.example
  4. Day 27 — 30

    Pilot platform integrations

    Three pilot platforms integrated, smoke-tested in production with shadow traffic, formal go-live for the largest of the three on day 30. Two engineers on call from kaedax for the 60 days after.

    3 platforms live shadow traffic clean 60d on-call

[ §02 ] agent log · selected

What the loop
looked like.

cycle-log · anchor
archived
T+120h [ OK ] scope.agent openapi spec frozen · 18 endpoints · drop-in for 2/3 incumbent contracts
T+240h [ >> ] build.agent quote engine shipped · p95 quote latency 84ms at 1k rps
T+360h [ OK ] qa.agent 224/224 contract tests pass · postman + dredd both green
T+480h [ >> ] build.agent partner console shipped · platform self-serve onboarding live
T+600h [ OK ] monitor.agent shadow traffic from pilot 1 · 0 protocol violations in 48h
T+720h [ OK ] deploy.agent pilot 1 cutover · quote-to-bind median 1.4s end to end

[ §03 ] notes from the cycle

Embedded insurance is plumbing. The shape of the work isn’t a beautiful frontend — it’s an API that other engineering teams will read the docs of, integrate against in an afternoon, and then forget exists for a year. We built ANCHOR’s plumbing.

The “drop-in-replaceable” constraint

ANCHOR’s pilot platforms had previously evaluated two incumbent embedded insurers. The fastest path to a pilot was to make ANCHOR’s API drop-in-replaceable from at least one of them — same shape, same error model, same auth pattern. scope.agent reverse-engineered both incumbents’ public docs and drafted ANCHOR’s contract against the union of their patterns. Two of three pilot platforms changed one base URL to integrate.

What makes API engineering different in an agent loop

Contract-first development gets easier, not harder, with agents in the loop. The OpenAPI spec is the source of truth; qa.agent generates tests from it; build.agent implements against it; the SDK generator produces clients from it. The contract is the artifact everyone — humans and agents — derives work from.

The eval was straightforward: did the SDK round-trip every endpoint, did the contract tests hold, did the load test sustain 1k requests per second without protocol drift. All three were table stakes; the agents got us to them in five days instead of five weeks.

What we wouldn’t take on

Claims adjudication. ANCHOR’s claims workflow involves human assessors, regulator-defined SLAs, and integrations into the underlying insurance carriers’ systems — that’s a year of work, not 720 hours. We were explicit about that on day one. The pilot ships quote-to-bind; claims live with the carrier.

from the founder

"The thing I didn't expect was the contract tests. We had a conformant API before we had a customer — that changed how confidently we sold the pilot."

— VP of Engineering · ANCHOR