Harper is an AI-native commercial insurance company in San Francisco. We're not bolting AI onto insurance — we're rebuilding the entire business as software, on a simple bet: turning expert human judgment into compute is one of the largest transitions left to make, and a trillion-dollar industry still run 90% by hand is the place to prove it. We've grown ~100x in the last year and we move at that speed — on-site, in person, long days, very high standards. Almost no one joins Harper for insurance; they join to build the company that replaces how it works.
The role
Harper isn't an AI tool sold to brokers. We are the broker — we do the work end-to-end and sell the outcome: the right coverage, fast, at the right price, with the right service. Owning both sides is the moat, and the companies that win this transition won't just have great AI; they'll have figured out how to organize themselves around it, so that knowledge gets encoded into systems agents and operators can query. That's the question this role sits inside. You'll own a module of the business end-to-end — the customer experience, the operator workflows, and the AI agents underneath — and run it with a forward-deployed engineer and the operators who live in it: sales, service, underwriting, ops. The founding PM team has gone wide across every module; now we go deep. How a daycare buys insurance versus a trucking company. What "urgency" means for a tow yard with expiring dealer plates versus a GL renewal. Your job is to encode that nuance into the systems until they do the work as well as a human in most places and better than a human in many. Own the module, move the metric, then go own the next.
What you'll do
Own the KPIs. Conversion, handle time, accuracy, autonomous-resolution rate, retention — whatever the leverage point is for your surface. You set the targets, instrument them, move them. If the metric isn't moving, that's your problem.
Encode the nuance. Translate what makes your module's customers different into rules, prompts, agents, and data structures.
Own the eval regime. Probabilistic systems are only valuable when people trust them: regressions on every change, evals mapped to real outcomes (not vibes), backtests against historical applications, call-by-call review where it matters. You'll be paranoid about silent regressions in a way most PMs aren't.
Build the data flywheel. Work hand-in-glove with data labeling and validation to build the golden datasets your module's models need. You define what "right" looks like.
Own the cross-modal experience. Your module spans web, voice, and human. You decide where each modality wins, where they hand off, how the on-ramps feel.
Live with operators. Sit with sales, service, underwriting. Watch the work. Find what's broken before they tell you.
Talk to customers every day. Literally — not "5 calls last quarter."
Prototype with AI. Claude Code, Cursor, Lovable. Walk into the meeting with a working prototype, not a deck.
Hyper-prioritize. Out of 50 asks, find the 3 that move the KPI and ignore the rest with conviction.
What we're looking for
1–3 years in product, or an early-career operator, engineer, or AI researcher who's been doing the work without the title.
Demonstrated end-to-end ownership of a product or system — KPIs, roadmap, execution — and a track record of going deep on a domain and encoding what you learned into a system.
You get what an AI services company is: we're not selling software, we're doing the work and selling the outcome, which means you ship behavior into a probabilistic system real operators and customers have to trust.
You're obsessed with evals — you'd rather ship a worse model with a great eval harness than the reverse — and you think in KPIs ("we cut handle time 40%," not "we shipped the feature").
You can build: Cursor, Claude Code, Lovable. You can argue AI tradeoffs (agents, LLMs, context engineering, data pipelines, evals) with the engineer who writes the code, even though you don't write it.
You go deep before you go wide, and you want to own a thing, not coordinate one. If "PM" sounds like meetings, this isn't it.
Bonus: AI/ML products, voice AI, agent frameworks, or workflow automation; eval/prompt/context engineering; insurance, fintech, or regulated-industry experience; prior startup experience.
The reality
On-site in San Francisco, Monday–Friday, roughly 5 AM–8 PM. The hours are long and the learning curve is steep — we hire early-career PMs on purpose and hand them the kind of surface area and reps that take a decade somewhere else. The people who thrive here wouldn't have it any other way. If that's a cost you're glad to pay, this is one of the few PM seats where you own a real piece of the business from week one.
Logistics
Compensation: $125,000–$170,000 base + performance bonus + equity.
Location: San Francisco, in-office. Based here or willing to relocate.
Benefits: Uber commuter benefits; breakfast, lunch, and dinner provided; snacks and coffee stocked; free gym membership; health, dental, and vision.
Process: 20-min lead screen → technical conversation (walk us through analysis you've done) → on-site.
To apply: Commercial insurance has never been rebuilt, and the gap between what frontier models make possible and what's actually shipped in this industry is the largest it will ever be. If you want to own a module that touches real customers from week one — send your resume and tell us about something you built that moved a number.