Engineering at WithCoverage
Artificial Intelligence to Protect the Real World
We build systems that handle tedious, repetitive work—so our team can do what AI can’t. You’ll work closely with the people who use what you build daily, and ship software that affects thousands of real businesses with real problems.
- Role
- member of technical staff
- Stack
- aws, vercel, eks, argocd, postgres, iceberg
- AI
- custom-built harnesses, autonomous coding, self-improving loops
- Where
- nyc, in office, five days
Changelog
Recent production changes.
Why this job exists
Engineering is suddenlyvery different.
Companies used to hire people to fuel growth and pay SaaS vendors for software they could use. That's all changed — now companies want to buy outcomes.
Our team builds the platform and tools to enable our risk management experts to scale themselves 2x, 10x, or 100x more than was possible in the past.
We are working to break the scaling laws of services businesses with AI agents. We are a deeply collaborative, in office, hard working and curious team in NYC.
Diagram: AI agents pull insurance documents from a task queue, work them concurrently, and ship finished artifacts without supervision. Interactive: choose a file in the queue to dispatch it to an idle agent, select an agent to see its live log, or select a shipped artifact to preview it.
Diagram: an engineer and a risk expert deploy as a pair into client verticals — defense, hospitality, logistics, energy — build on site, and ship measurable outcomes. Interactive: choose a site to deploy the pair there, or select a deployed site for details.
Diagram: an AI harness loops through run, eval, learn, and update stations. Each lap appends an improvement to the harness changelog, evals flip from fail to pass, and the pass rate climbs from 62 to 94 percent. Interactive: select the next station to drive an iteration by hand, or any station for details.
Diagram: a fleet of AI agents serves a growing wall of client accounts, routing only rare judgment calls to a single risk expert. Accounts per expert climbs from 10 to 100. Interactive: choose an unlit account tile to onboard it, or select the expert or an agent for details.
Real, interesting problems — You'll build AI infrastructure that can read and understand complex documents, ask questions, make decisions, and take actions that have real consequences for our clients' businesses. You'll learn what it takes to build and deploy autonomous agents that manage risk at scale in the real world.
Inside the team
The actual work.
- What are you building?
- An AI-driven claims platform that gives clients full visibility into their claims, losses, and legal processes — turning messy insurance data into something clear.
- Where does your time go?
- Deep dives into a problem, designing the system, shipping it, then iterating fast with the people who use it. One day I'm evaluating the latest AI, the next I'm shipping what the team uses tomorrow.
- What was hard to get right?
- Nuance. Insurance is a language-based industry — every policy is specific to the client and their business, and one small detail flips the answer. Building repeatable systems on that takes real domain depth.
- Why did you join?
- The team, and how embedded we are with our users. I get to shape the product next to the people using it and watch the impact land in real time.
Leadership
Who you'll build with.
An entrepreneur and product leader with a track record of building category-defining companies. He co-founded Opendoor, taking it from inception through IPO, and brings a product-first instinct to everything we build.
A pioneer in AI-native engineering with a background that blends finance and technology. Previously a senior principal engineer and chief architect in charge of transforming a 700-person R&D org to be AI-native.
How we hire
A process builtto respect your time.
Technical Deep Dive
precision and depthWe go deep on the decisions, the tradeoffs, and what you'd do differently. No trivia, no whiteboard puzzles.
Coding Round with AI
how you steer the modelBring your own AI tooling and pair on a real problem. We don't ban the model or treat it as cheating — using it well is the job. We watch how you prompt, review, debug, and steer it, because that's exactly how you'll ship here.
Onsite — 2–3 Interviews
systems and product judgmentA few focused sessions on systems design and product thinking. Architect something real, then reason about what to build, for whom, and why.