Kinro claims first autonomously sold insurance policy

What Kinro Is Building

Kinro is hiring two founding engineers to wire up what it calls the first autonomous insurance brokerage for small businesses. The Y Combinator Spring 2026 batch company posted the roles in San Francisco as it scales beyond a three-person founding team and pushes toward a product that sells and binds commercial insurance policies without a human broker in the loop.

The startup frames the work as a direct response to a distribution problem in the US small-business market. About 36 million SMBs operate in the US and roughly 77% are underinsured, in part because brokers cannot serve low-premium accounts profitably. Handling a $2,000 policy takes nearly the same time as handling a $20,000 one. Kinro's pitch is to replace that labor with AI sales agents that run around the clock, handling buyer qualification, real-time quoting, coverage recommendations, and policy binding across channels including ChatGPT, Google Gemini, WhatsApp, and embedded partner experiences. The company says it has already bound what it describes as the first autonomously sold insurance policy.

The team behind the vision is unusually concentrated for a regulated-finance startup. Co-founder and CEO Pierre-Alexandre Kamienny previously worked as an AI researcher at Google DeepMind, where he helped adapt Gemini for financial services, and earlier did a PhD at Meta FAIR after a master's at Oxford. Co-founder and CTO Parth Ainampudi led infrastructure and systems work across training and inference at Zoox, with prior stints at Samsung's Advanced Technologies Group and graduate study in robotics at UT Austin. Co-founder and COO Corentin Hugot was the first employee at Sharelock, an insurtech that reached profitability working with major carriers and brokers. YC partner Nicolas Dessaigne is listed as the primary YC contact. The company is based in San Francisco, is seed-stage, and lists Crystal Venture Partners among its funders.

Before joining YC, the founders say they built insurance workflow software for brokers and hit $1 million in annual revenue, then entered the batch with a signed $1 million enterprise sale to the second-largest US auto-insurance comparator. About a third of that amount had already been wired. Within three weeks of launching as a brokerage, Kinro says it secured licenses in 30 states and carrier appointments from 15 insurers, and it has run a pilot with WTW, the third-largest insurance broker on Earth. Other named customers include Tuio, Lovys, Dalma, Assurland, and Quartz. Kinro focuses on admitted markets and lines such as General Liability, Business Owner's Policy, Workers' Compensation, and Commercial Auto, where checking carrier appetite and generating quotes can be fully automated.

The two engineering openings sit squarely on top of that ambition. One is a Founding Software Engineer, Infrastructure, advertised on Kinro's team page as the role for someone comfortable being the only infrastructure engineer at a small company, partnering directly with founders and shipping fast inside a regulated industry. The other is a Founding AI Engineer. A separate Content & Brand Intern posting sits alongside them. According to the YC directory, the company currently lists one engineering role, though the team page describes both founding engineering positions. Kinro has framed these hires as defining how the company builds software "in the age of agents," a hands-on remit that spans product, backend, evaluations, and infrastructure for engineers who treat agentic coding as a force multiplier rather than a buzzword.

That hiring ask reflects a bet that the hardest part of autonomous insurance is not the AI itself but the regulated plumbing around it.

Founding Software Engineer, Infrastructure: Where the Regulatory Weight Sits

The infrastructure seat at Kinro is the half of the founding-engineer pair that holds the company's regulatory weight. Where the ML role designs the conversation, this one builds the rails the conversation runs on: the carrier integrations, the compliance plumbing, and the production environment that has to stay up while real transactions clear.

Responsibilities cluster around three jobs that can't be cleanly separated at this stage. First, building AI product systems end-to-end, from tool use and backend services to evals and production monitoring, which on the infrastructure side means the services, queues, and observability stack that the agent calls into. Second, using agentic coding tools aggressively and responsibly: breaking down large problems, setting guardrails, and reviewing generated work with high standards, an unusual expectation for a backend-heavy role and one that reflects how a six-person team plans to ship at LLM-company speed. Third, improving agent quality, latency, safety, and cost through better architectures, feedback loops, and evaluation harnesses. The infrastructure engineer owns the latency and cost halves and partners with the ML hire on evals.

The stack reasoning is spelled out in the job spec: candidates are expected to reason across application code, APIs, databases, cloud systems, and operational tooling, and to understand tool calling, latency, reliability, security, and cost tradeoffs in agentic systems. There's no public list of languages or cloud providers, so a candidate walking in should expect to defend choices rather than recite a stack, and the screening is built around that posture. Kinro also explicitly flags experience with evals, model routing, prompt injection defenses, sandboxing, or other AI systems safety work as a plus, which means an SRE-style resume without an AI-safety line item will read as adjacent rather than on-target.

The regulatory layer is what separates this from a generic founding-infra role. Insurance distribution in the US and Europe is gated by carrier integrations and compliance checks, and the founders' bet is that "platforms like OpenAI and Google remain general-purpose and are less likely to build the vertical infrastructure required for regulated industries." That bet is the reason the seat exists. Kinro's agents handle qualification, quoting, recommendations, and binding, and the company's own framing claims the agents stay "100% compliant" through a proprietary evaluation platform, which is the system the infrastructure engineer will be hardening, not just deploying.

Experience priors that the spec treats as concrete signals rather than nice-to-haves include high-ownership product environments where engineers talk directly to users and ship quickly, time in fintech, insurance, or other regulated environments, and prior production work on AI products, agent workflows, or LLM systems. Co-founder Parth's background in infrastructure for training and inference at Zoox sets the bar the role is written against: someone who has carried a real inference or training stack and can do it again inside a six-person insurance company. Corentin, who was the first employee at a prior insurance startup and worked closely with insurance leaders, is the closest internal reference for the carrier-side context the infrastructure hire will need to absorb fast.

The interview process the candidate will eventually face is short by design: a 20–30 minute intro, a 45–60 minute technical and product deep dive, 45–60 minutes with founders and team, a one-week paid trial, and references plus a final conversation. The deep-dive slot is where the infrastructure expectations get tested in practice, and the trial is where the agentic-coding posture either becomes real or doesn't.

Founding ML Engineer: The Brain of the Sales Loop

The ML seat will own the brain of the company's flagship product: AI sales agents that handle the entire insurance buying flow, from initial qualification through quoting, coverage recommendations, and binding. The north star is "selling insurance in LLM platforms (ChatGPT, Gemini…)," which means the agents don't just live on Kinro's site. They have to function inside third-party assistants, which raises the bar on latency, reliability, and the kind of evaluation rigor most early-stage teams never build.

The YC listing demands an engineer who has "built AI products, agent workflows, or LLM systems end-to-end in production." Kinro isn't hiring someone to fine-tune weights in a vacuum. The job description spells out the actual scope: "build AI product systems end-to-end, from tool use and backend services to evals and production monitoring," and "improve agent quality, latency, safety, and cost through better architectures, feedback loops, and evaluation harnesses." That last phrase is the one that separates this from a typical ML role. Evals aren't a nice-to-have; they're the product, because Kinro's agents must stay "100% compliant" while operating inside a regulated sales motion.

A third-party scrape of the same listing (swiftcruit.ai, dated 2026-04-08) breaks out the required skill stack: LLM, agentic workflows, tool calling, backend services, evals, production monitoring, cloud systems, databases, APIs. Notice what's missing. There's no mention of training foundation models, no RLHF pipelines, no PyTorch or JAX in the list. The job is about orchestrating and evaluating large models in a high-stakes vertical, not building them from scratch. That fits the founder profile. Pierre-Alexandre "led research improving Gemini for financial services at DeepMind," so the team already knows what frontier-model work looks like and has chosen to spend its engineering capital on the application layer instead.

The ML engineer also has to be a full-stack operator in disguise. Kinro writes: "work across the stack wherever leverage is highest, including product surfaces, infra primitives, and customer deployments," and "turn one-off lessons from production into reusable product and platform capabilities." In a six-person company, there's no one hiding behind a specialist title. The ML hire will debug a tool-calling failure on a Tuesday, ship a new eval harness on a Wednesday, and sit in on a broker call on a Thursday to figure out why the agent's coverage explanation confused a buyer. The listing calls this out as "strong product taste and care whether something feels simple, fast, and reliable for the end user."

The bar for domain knowledge is real but soft. Kinro lists "experience in fintech, insurance, or other regulated environments" as a qualification, not a requirement. The harder requirement is the agentic-systems mindset: "you know when to trust an agent, when to intervene, and how to structure work so the agent succeeds," and "you understand tool calling, latency, reliability, security, and cost tradeoffs in agentic systems." Candidates who've shipped agent products at scale, and who can point to a specific eval or guardrail system they built, will land closer to the top of the resume pile than candidates who've only run prompt experiments.

A one-week paid trial is part of the process, a signal that Kinro expects hires to integrate into the team rhythm before a formal close. The founders want to see how candidates handle real production problems, not whiteboard ones.

Comp at a Glance

The infrastructure posting is listed under Standout (YC P26) as a full-time founding-engineer position based in San Francisco, with visa sponsorship available, and Kinro itself is a YC P26 batch company with a six-person team founded in 2026. The ML seat is on-site in San Francisco, with visa sponsorship available and a minimum of three years of experience. The wider band on the infrastructure side signals how much of the negotiation will hinge on how much systems the candidate has already shipped.

The Pipeline: Three to Five Weeks, Four Rounds

Kinro hasn't published a formal interview playbook, but the shape of its pipeline is easy to read from how comparable YC-backed AI startups and insurtechs hire founding engineers. The expected structure runs roughly three to five weeks across four rounds, moving from a recruiter screen to a hands-on technical exercise to founder sessions that probe autonomy as much as code.

The first round is an initial screening, typically a 30-minute conversation that covers background, motivation, and a high-level read on technical range. It's also where candidates should expect to defend, at least in outline, why they want a founding seat rather than a senior IC slot. The second round is a technical screen, usually 45 to 60 minutes with an engineer, designed to surface software fundamentals: data structures, system design at small scale, and the candidate's ability to talk through their reasoning out loud rather than just producing an answer. Founders consistently say that watching how someone thinks matters more than whether their first solution is optimal. Public coaching material puts it bluntly: "Think out loud… a much better way to respond than saying 'I'm not really sure.'"

Round three is where the process distinguishes itself from a standard big-tech loop. Rather than another whiteboard round, candidates get a practical exercise, either a take-home assignment or a paired coding session running two to four hours. For the infrastructure seat, this tends to look like building a small piece of a backend service or wiring up an API against a mock data source, then walking through tradeoffs. For the ML seat, exercises lean toward prompt-and-pipeline work: wiring an LLM call into a retrieval or tool-use flow, evaluating outputs, and writing the test cases that prove it won't regress. Candidates should expect to defend their choices (language, database, architecture) and to debug their own bugs in front of the interviewer. The same coaching puts it this way: "If your solution has bugs, are you able to walk through your own logic to find them and explain what the code is doing?"

The final round is the behavioral-and-founder alignment stage, where the founders probe the traits that don't show up in code reviews. HyperNest Labs, which has studied founding-engineer hiring extensively, frames these as autonomy, pragmatism, breadth, speed, communication, resilience, and ownership. The questions are pointed: would I trust this person to ship a critical feature while I'm fundraising? Will they still be excited when things break at 2 a.m.? "I need clear requirements" is treated as a red flag, because founding engineers define requirements; they don't receive them. Other red flags include only big-company experience, perfectionism, "that's not my job" framing, and questions that lead with equity. Candidates are coached to organize answers using the S.T.A.R. method (Situation, Task, Actions, Results) and to keep responses grounded in shipped work rather than aspirations.

A few practical signals worth flagging for applicants. Most cycles close within a few weeks, and recruiters at well-run startups typically give a timeline up front; if you haven't heard back within a week of an interview, follow up. Kinro's team operates from an in-office setup, a relevant data point, since founding roles often carry an implicit expectation of overlap with the founders during the early months. And throughout the loop, the thing being optimized isn't a score; it's a yes/no on whether this person can carry a critical surface area of the product without a manager looking over their shoulder.

What the Market Looks Like From the Hiring Side

Kinro's twin founding-engineer search lands inside an AI-insurtech labor market that has tightened sharply even as capital flows back into the sector. Global InsurTech funding rebounded to $5.1 billion in 2025, the first annual increase since 2021, with AI-labeled startups capturing roughly two-thirds of that total and nearly 78% of Q4 investment, according to BCG.

That money is chasing a small, contested talent pool. ML engineer compensation in San Francisco now spans $110k–$145k at entry, $150k–$210k at mid-level, and $215k–$320k for senior practitioners, per January 2026 compensation data. Stripe's own job board shows the ceiling from the demand side: a Machine Learning Engineer role posted in South San Francisco carries a $212,000–$318,000 band, and Zero G Talent reports Stripe added 74 roles in the past seven days alone, illustrating the velocity at which well-funded AI-adjacent employers are pulling candidates out of the market.

The competitive backdrop is shaped by more than capital. BCG's State of InsurTech 2024 report, released June 2025, found that median pre-money valuations in InsurTech had climbed 48% to an all-time high. Investor concentration at the top has hardened: Series A and B activity dropped from 28% to 19% of deals year-over-year, while Series C funding rose 96%. Capital is concentrating in companies that already look like they can scale, raising the bar for seed-stage teams like Kinro to prove technical credibility fast. The number of active InsurTech investors also rose from 788 in 2024 to 852 in 2025, the first increase since 2021, which means more funds are writing checks into a market where AI fluency is now table stakes.

Talent competition is intensifying on three fronts. First, incumbents are buying in rather than building. Liberty Mutual Strategic Ventures launched a $200 million fund in July 2024 focused on AI, cyber, and data analytics; Aviva Investors followed with a £150 million vehicle in December 2024. (Re)insurers completed 162 private technology investments in InsurTech during 2025, the highest annual count on record. Second, carriers are hiring AI engineers directly to avoid dependence on vendors. Prudential's adoption of Google's MedLM in Singapore and Malaysia, which the BCG report says doubled claim automation rates, is one cited example. Third, a wave of AI-native competitors is going after the same workflow layer Kinro targets: Indemn uses AI-powered conversational agents for quoting and underwriting; Linqura generates risk-profile insights for agents; BluePond.AI's Broker CoPilot automates policy checking, quote comparison, and submission ingestion.

For founding engineers weighing Kinro against offers from better-known AI shops, the trade-off is concrete. A late-stage employer like Stripe offers brand, comp, and an existing codebase. A seed-stage AI broker offers scope, equity, and the chance to define the stack. The wider market data suggests that scope is increasingly scarce: BCG's 2024 report counted only seven InsurTech mega-rounds that year, all at Series C or later, meaning early-stage AI-insurtech talent has fewer large-employer alternatives than at any point since 2022.

What Actually Gets You Past the Screen

Kinro's job listing draws a sharp line between candidates who have shipped AI products and those who have only experimented with them. The posting spells out the bar: "You have built AI products, agent workflows, or LLM systems end-to-end in production." Résumés that show prototypes, side projects, or Kaggle notebooks will not substitute for that production record. Hiring managers at a six-person YC P26 startup need engineers who have already carried a system from idea to live traffic and absorbed the operational scars that come with it.

The technical stack that signals competency is narrow and concrete. Kinro calls out LLM agentic workflows, tool calling, backend services, evals, production monitoring, cloud systems, databases, and APIs. Applicants should be able to point at work where they owned at least a meaningful slice of that chain. A candidate who shipped an agent that handled quoting or claims triage at a previous employer, and who can describe the architecture, the failure modes, and the cost numbers behind it, speaks Kinro's language. So does someone who built an evaluation harness that caught regressions before customers did. These are the project experiences the team will recognize on a résumé without having to translate them.

Domain familiarity is a meaningful accelerant. The listing flags "experience in fintech, insurance, or other regulated environments" and Kinro's founders have already lived in those worlds: Pierre-Alexandre led research improving Gemini for financial services at DeepMind, and Corentin was the first employee at an insurance startup. Candidates who have navigated SOC 2 audits, carrier integrations, or state-level insurance compliance bring context the team would otherwise have to teach themselves. Even adjacent regulated exposure (payments, healthcare, lending) reads as relevant because the underlying discipline (audit trails, deterministic answers, strict data handling) translates.

AI safety work is the differentiator most candidates underweight. The listing singles out "experience with evals, model routing, prompt injection defenses, sandboxing, or other AI systems safety work." In a brokerage where every recommendation has to be 100% compliant and where insurers pay roughly $100B a year in commissions, an agent that hallucinates a coverage limit is a lawsuit waiting to happen. Résumés that include a named evals framework, a published postmortem on a prompt injection attempt, or a sandbox architecture move to the top of the stack.

Product taste and ownership are equally weighted. Kinro states it cares most about "people who can build useful products fast, reason clearly about tradeoffs, and raise the quality bar across the stack." That bias shows up in the process: a 45–60 minute systems deep dive where interviewers walk through architecture, tradeoffs, and how the candidate shipped in production, then a one-week paid trial working on real problems. Engineers who have run user interviews, written their own product specs, or shipped UI work alongside backend code signal they can survive the trial without hand-holding. Kinro's "high-ownership product environments where engineers talk directly to users" line is doing real work in the posting; applicants who can name specific customer calls they ran will have a clear edge.

The compensation range ($120,000 to $300,000 plus 0.75% to 2.00% equity) is wide enough that strong signals pull candidates toward the upper band. A founding engineer who can demonstrate production LLM systems, regulated-industry experience, and a working eval discipline is the profile that lands at the top of that range and earns the offer before the trial ends.

Why Closing These Seats Is the Gating Move

Securing both founding engineers reshapes Kinro's roadmap along two tracks that run in parallel: a near-term product milestone tied to carrier and channel expansion, and a longer-horizon bet on becoming the default insurance layer inside AI assistants like ChatGPT and Google Gemini. The hires are not incremental headcount; they are the gating input to the company's stated headcount plan.

Kinro publicly frames its growth as a math problem: "$1M with 3. Next target: $10M with 10. … We will take Kinro to $100M ARR with a team of 20." That ratio means each new engineer carries roughly $5M of incremental ARR responsibility on the way to $10M, and about $6M at the $100M stage. The two founding engineer roles are the first two of the seven remaining spots to hit the $10M milestone. Closing them quickly is the single highest-leverage move on the company's near-term plan, because every week the seats stay vacant compresses the timeline to revenue targets the founders have already put on the record.

On the product side, the infrastructure engineer maps directly to the workflow Kinro has already proven out. The team bound "the first-ever autonomously sold insurance policy," got licensed in 30 states, and was appointed by 15 carriers in three weeks, a footprint that only works because admitted-market appetite and quoting can be "fully automated with no communication with carriers." Scaling from 30 states to a full US footprint, and from 15 carrier appointments to the broader appetite graph, requires an engineer who can harden the integrations that brokers and carriers expose through APIs. The same infrastructure also has to support deployment across ChatGPT, Google Gemini, WhatsApp, and embedded partner experiences.

The ML engineer drives the second track. Kinro's agents handle "the full insurance sales lifecycle with minimal human touchpoints, from initial customer discovery through policy issuance," which positions the ML seat as the owner of the model's behavior at every regulated handoff: routing complex cases to licensed agents while preserving conversation context, drawing a clean boundary between estimate, quote, and purchase, and producing the evidence that regulators now expect before AI touches a real buyer. That evidence work is no longer optional. NAIC's April 1, 2026 implementation map lists 25 adopted AI model bulletin actions including D.C., and its AI Systems Evaluation Tool was being piloted by 12 participating states as of March 2026. A founding ML engineer who can ship the production-readiness artifact, not just the model, is what turns Kinro's pilot coverage into multi-state production coverage.

Fundraising follows product proof. Kinro closed a $15M seed in 2023 led by Andreessen Horowitz, with Jeff Dean, Amjad Masad, and Daniel Gross participating. That round bought the runway to demonstrate an autonomously bound policy and to post the $1M ARR result that put it inside the 7% of YC W26 startups that hit $1M, versus the historical 2–3%. The next round will be priced off what the two founders can ship between now and the next milestone, not off the demo alone. With the infrastructure hire in place, the company can credibly talk about scaling carrier integrations and channel distribution. With the ML hire in place, it can credibly talk about regulatory-grade autonomy in additional states. Either hire alone narrows the story; both together let the company pitch the $10M ARR target as engineering-led rather than sales-led.

The competitive backdrop tightens the timeline. Simply Business launched an insurance app inside ChatGPT on April 23, 2026, with final quotes and policy sales kept on simplybusiness.com and no PII collected inside the ChatGPT interface. Insurify has done similar work. Both incumbents have distribution Kinro does not, which means the window to convert a technical lead into a distribution lead is measured in months, not quarters. Filling the two seats is the precondition for closing that window before it narrows.

Working in frontier tech? Zero G Talent tracks the openings: see every open ASML role, browse frontier tech jobs, openings at Stripe, and the people building the field.