Naive Hires 4 Founding Engineers. Founders Read Every CV.

Why Naive Is Hiring Now

Naive, a ten-person Y Combinator-backed startup that closed a Series A on August 6, 2026, has opened four founding-engineer roles, the first deliberate expansion of its technical core since the pivot that won it YC admission.

The Mountain View company, backed by Y Combinator's Spring 2025 cohort, is building an autonomous company runtime: infrastructure that lets AI agents incorporate businesses, open bank accounts, spin up cloud resources, and operate them end to end. Over the past six months, annual run-rate revenue has grown 10x to the low double-digit millions, TechCrunch found, and more than 30,000 developers have signed up since launch, TechCrunch's figures put. That traction forced a hiring plan the founders hadn't expected to execute this early.

Sean Dorje and Dennis Zax founded the company in 2025 under the name Relixir, a generative-engine-optimisation platform. They pivoted after realizing the bigger opportunity wasn't optimizing content for AI search — it was giving AI the tools to execute real business operations. The Y Combinator acceptance followed the pivot, and the Series A, announced August 6, 2026, TechCrunch reported was led by Nexus Venture Partners with participation from Y Combinator, Zetta Venture Partners, Liquid 2 Ventures, 468 Capital, and angels including Gokul Rajaram, Apollo.io co-founder Tim Zheng, and former HubSpot COO JD Sherman.

Dorje told TechCrunch the proceeds are earmarked for four infrastructure projects: virtualized sandboxes for agents, a model router that sends queries to the most efficient model while preserving reasoned context, a memory layer that stores and surfaces business context, and a governance-orchestration layer for dividing work among agents. A fifth effort (a serverless runtime that runs agents in lightweight JavaScript environments instead of full virtual machines) aims to cut the inference-cost line that Dorje says is now the largest expense for anyone operating agent fleets at scale.

"Part of running an autonomous company and running agents, like that's your biggest cost line now, and so the highest growing demand right now, I would say is for inference and serverless agents," Dorje said.

That cost pressure is pulling enterprise interest, though Dorje declined to name specific companies. The same infrastructure that helps a solo founder launch an AI automation agency or a faceless TikTok channel (both early customer archetypes) also appeals to established businesses trying to run thousands of agents without burning through compute budgets. Naive's API already powers deployments at more than 500 companies, including Airwallex and Hackerrank.

The product surface has expanded accordingly. Beyond the core incorporation-and-infrastructure API, Naive now ships templates for AI SEO agencies, full-stack SaaS apps, recruiting workflows, accounting agents, customer-support agents, and a mobile emulator that lets agents operate smartphone apps on virtual devices. Internal products (Mobilerun, Naive Agent VISA, and Vetta) sit alongside the public platform.

With ten full-time employees and a roadmap that spans systems research, distributed runtime engineering, and reinforcement-learning-driven agent optimization, the four Mountain View roles (Founding Growth Engineer, Founding Product Engineer, Founding Member of Technical Staff (RL), and Founding Member of Technical Staff (Agents)) represent the first deliberate expansion of the technical core.

The Four Open Roles and Their Core Requirements

Naive's four open positions reflect a deliberate split: two deep technical tracks focused on the agent runtime itself, one product-facing engineering role, and one growth-oriented engineer who sits at the intersection of product and distribution. All four carry the "Founding" prefix, signaling equity stakes of 2–4% and base salaries within that range, with a 3+ year experience floor across the board.

Founding Member of Technical Staff — Agents

This role owns the core agent runtime: the orchestration layer that turns a natural-language business description into a fleet of AI employees that write code, close deals, run campaigns, and serve customers. The job demands fluency in LLM orchestration frameworks and production-grade prompt engineering that survives edge cases at scale. Candidates need to show they've built RAG pipelines backed by vector databases, designed multi-agent workflows with explicit evaluation harnesses, and shipped those systems to real users. Python is the baseline; the team expects clean architecture, rigorous testing, and the ability to optimize latency and cost without hand-holding. A GitHub trail of substantial agent projects carries more weight than a PhD in model training.

RL Track

The reinforcement learning track is narrower but deeper. Naive uses RL to tune agent behavior post-deployment: reward shaping for sales agents that negotiate, support agents that resolve tickets, coding agents that refactor without breaking tests. The role requires hands-on experience with policy gradient methods, offline RL, or preference learning (RLHF/DPO) applied to LLM agents. You should be comfortable designing reward models from sparse business signals, debugging credit assignment across multi-step tool use, and running large-scale experiments on GPU clusters. Publications help, but the bar is a working system that improved a production metric (conversion, resolution rate, code quality) and the receipts to prove it.

Founding Product Engineer

This is the bridge between the agent runtime and the customer's business logic. The product engineer builds the interfaces (dashboards, APIs, onboarding flows) that let a non-technical founder describe "we sell B2B SaaS to dentists" and get a deployed agent suite that qualifies leads, demos the product, and onboards paying customers. The role demands full-stack competence, strong opinions on developer experience, and the ability to ship fast without accumulating technical debt. You'll work directly with the 500+ deployed companies, so empathy for operators who've never written a prompt matters as much as your implementation skills. Prior experience building developer tools or no-code platforms is a strong signal.

Founding Growth Engineer

Growth at Naive doesn't mean SEO or ad buying. It means building the self-serve activation loop: the onboarding wizard that connects a customer's CRM, email, calendar, and codebase in one click; the instrumentation that tells the team which agent skills drive retention; the automated evals that catch regressions before a customer notices. The stack overlaps heavily with the product engineer, but the success metric is time-to-first-value for a new signup. You need to have owned a growth funnel for a technical product, run statistically sound A/B tests on onboarding flows, and built internal tooling that lets non-engineers run experiments. Familiarity with product-led growth motions at developer-tool companies is the closest proxy.

Across all four roles, the unspoken requirement is evidence of shipping AI systems that operate autonomously in production — not demos, not notebooks, not research prototypes. The company's own description makes the scope clear: "Naïve is an autonomous company runtime. You describe what your business does, and Naïve deploys AI employees that actually execute." That execution mandate filters out candidates whose experience stops at prompt tuning or single-model fine-tuning.

How Resumes Get Screened: General Best Practices

Applicant tracking systems are filing cabinets with search bars. They parse every resume into plain text, store the result in a searchable database, and rank candidates against the job description so a recruiter can work through hundreds of applications in a single sitting. The system makes no hiring decisions. Every ATS used in production lets the recruiter open and review the original file, and no credible research supports the claim that 75 percent of resumes are never seen by a human; that figure originated in a 2012 sales pitch from a resume-optimization vendor.

Roughly 78 percent of employers run an ATS, and a 2026 Jobscan analysis detected one at 97.4 percent of Fortune 500 companies. In a survey of recruiters, 82.3 percent said they use an ATS and 37.4 percent called it central to their process. A high-demand role can draw 400 to more than 2,000 applicants within days; a recruiter reads a fraction of them. Sixty-four percent of organizations now layer AI or automation on top of the ATS to narrow the pile before a person looks at it, and 88 percent of employers believe they lose qualified candidates whose resumes are not ATS-friendly.

The parser reads plain text. Tables, columns, graphics, text boxes, and special formatting often break extraction even when a human recruiter would love the layout. Creative headings like "My Journey" can send experience into the wrong database field or nowhere at all. Standard headings (Experience, Education, Skills) keep data in the right buckets. A .docx file or a text-based PDF is safest; if you cannot select the text in your PDF, neither can the parser. Contact information must be highlight-able so the system can pull name, email, and phone automatically.

"An ATS is a database and workflow tool, not a rejection robot. It parses your resume, stores it in a searchable format, and ranks or filters candidates so recruiters can do so." — Novoresume

Keyword matching drives ranking. The system scans for terms lifted directly from the job description. If the posting says "data visualization" and your resume says "charts and dashboards," you score low on a skill you actually have. Mirroring the posting's language is the single highest-leverage edit. One resume matched to one role beats the same file sent fifty times; people who send the same resume to every job are three times more likely to end their search with zero interview invitations than those who tailor each application. Job seekers who keep four or more tailored versions average 4.2 interview invitations per search; those who send one version everywhere average 2.0. Ninety-three point six percent of people who customize every application land at least one interview.

Common mistakes that tank parsing or ranking include complex formatting, missing relevant keywords, inconsistent job titles, unclear section headings, and unsupported file formats. Listing responsibilities instead of measurable results remains the most cited resume mistake. Recruiters decide whether to keep reading in twenty seconds or less, 74 percent skim that fast once the resume reaches a human. Forty-two point six percent of job seekers used AI the last time they updated their resume, and more than a quarter submitted it without editing a word. Sixty-eight point two percent spend an hour or less on the update. AI-written resumes are everywhere, and recruiters notice.

After a rebuild, one candidate went from a 5 percent HR screen-to-interview rate to 75 percent. In a user success study of 915 job seekers, 46 percent of those who rebuilt their resume received an offer within one month and 79 percent within three months. Average .doc resumes flagged with significant ATS gaps saw a 33 percent rejection rate; resumes built on a clean foundation hit 97 percent approval.

Many companies now add an LLM layer that reads and summarizes fit in plain language. This layer rewards context, not repetition. Job descriptions leak priorities: roles titled "Founding Member of Technical Staff - Agents" and the RL role signal a stack built on agentic reinforcement learning. Candidates who list "PyTorch" and "RL" pass the first filter. Candidates who specify "GRPO," "PPO with KL penalty," "GAE estimators," or "multi-node H100 rollout pipelines" signal deeper relevance.

For product engineering roles, the technical gate shifts toward on-device model integration patterns. Apple's Foundation Models framework now exposes on-device foundation models through Swift-native sessions with guided generation, tool calling, and streaming built in. The framework supports constrained sampling that defines what the model can emit, reducing invalid structures. Experience with guided generation against custom Swift data structures (not raw string parsing) and LanguageModelSession state management across multi-turn interactions carries weight. So does Core ML optimization: INT4 and INT8 weight quantization, palettization, sparse representations, and activation quantization tuned for Neural Engine on A17 Pro and M4 silicon.

Growth engineering roles demand a different keyword set. Startup ATS tools hunt for "zero-to-one," "scrappy," and "cross-functional," cultural keywords that signal you understand startup chaos. But the LLM layer looks for quantified outcomes: "Grew MRR from $5K to $120K in 8 months," "Built data pipeline processing 2M events/day," "Shipped MVP in 6 weeks, acquired 500 beta users in first month," "Reduced customer onboarding time by 60%." Vague categories ("web development," "design tools," "payment systems") fail. Specific tools pass: Next.js, Tailwind CSS, Prisma, Stripe API, Vercel, Figma.

Project-based filtering outweighs pedigree. Side hustles and freelance work often carry more weight than traditional corporate roles. Hiring networks increasingly aggregate signals from GitHub, arXiv, Google Scholar, and LinkedIn into continuously updated profiles because recruiters need to see how a candidate contributes, who they've worked with, and where their impact is visible. A candidate with an open-source agent framework contribution, a technical blog post on RL reward shaping, or a conference presentation on on-device model compression clears the technical gate faster than a candidate with a FAANG title and no public artifacts.

The reverse chronological format remains what hiring managers expect and what ATS systems are programmed to process. But the content within each role must follow a cause-and-effect narrative: problem, action, measured result in one line. "Led a team of eight through a product launch that shipped three weeks ahead of schedule" beats "managed a team." "Chose to delay launch by six weeks, resulting in a 40% drop in customer support tickets" beats both; it reveals judgment, not just responsibility.

Screens match for the language of the posting. If the description says "revenue operations" and you wrote "sales support," you don't exist. If it says "agentic RL" and you wrote "reinforcement learning," you might survive the ATS but stall at the LLM layer. The candidates who advance mirror the company's vocabulary honestly, then back each keyword with a measurable outcome a human can verify in thirty seconds.

Interview Rounds: Industry-Standard Structure

The interview pipeline at a YC-backed AI startup follows a pattern that has hardened across the industry: structured, time-bounded, and ruthlessly focused on job-relevant signal. Stakteck's 2025 technical interview guide outlines a five-stage model that most high-signal companies now approximate (recruiter screen, technical assessment, live technical interview, system design, and hiring manager conversation) with the entire loop targeting two weeks from first contact to offer. Every additional week of delay increases candidate dropout by 10 to 15 percent.

The recruiter screen runs 30 minutes and is not a formality. It serves three functions: qualify the candidate's experience, salary expectations, and availability against the role; sell the opportunity to top talent who are evaluating the company as much as the company evaluates them; and set expectations for the process, timeline, and what each stage covers. Candidates who know what to expect perform better and feel more positive about the experience. Duke's career hub notes that for full-time technical roles, candidates should expect at least two or three technical interviews beyond any initial assessments, with senior roles demanding more.

Stage two is the technical assessment: 60 to 90 minutes, either a live coding session or a take-home assignment. The Stakteck guide recommends limiting this to two or three problems of varying difficulty, focusing on practical problem-solving rather than obscure algorithms, allowing the candidate's preferred language, and capping time at 90 minutes. Evaluation covers not just whether the solution passes test cases but code quality, edge-case handling, and approach. For take-homes, the guidance is clear: scope to three to five hours of work, provide clear requirements and a two-to-three-day deadline, allow normal tools and references, and evaluate test coverage, documentation, architecture decisions, and edge cases. Written feedback should follow regardless of outcome.

The live technical interview (stage three) is a 60-minute conversation between engineers, not a repeat of the coding test. This is where depth gets assessed: data structures and algorithms in the context of real problems the team faces, object-oriented design, and the ability to articulate trade-offs. Harvard Career Services notes that technical questions for roles like software engineer, data scientist, or product manager can take the form of coding challenges, brain teasers, or product case scenarios.

System design arrives at stage four for mid-level to senior roles: 45 to 60 minutes of whiteboard architecture discussion. This evaluates architectural thinking, trade-off analysis, and the ability to design systems that work at scale. The Stakteck guide emphasizes that if the role involves building REST APIs, the test should cover API design, not binary tree algorithms from a textbook.

The final stage is a 45-minute hiring manager conversation. This is not a technical evaluation, the technical bar should be cleared before this point. It evaluates cultural fit, career alignment, and team dynamics. Every interviewer at every stage must evaluate against a pre-defined rubric, not gut feeling. A simple five-point scale runs from "does not meet the bar" to "exceptional, top 5 percent of candidates at this level." Companies that track pass-through rates at each stage, dropout rates, offer acceptance rates, and new-hire performance at six months can correlate interview scores with on-the-job outcomes and calibrate interviewers over time.

The Duke guide emphasizes that every company is unique, there is no universal pattern, even among tech companies. But the structural constants remain: structured stages, same criteria for every candidate, time-bounded process, and evaluation tied to actual job requirements. Candidates who reach the onsite at a top-tier tech company have already cleared a funnel where only 5 out of 100 applicants typically make preliminary screening, and only 1 in 5 of those receives an offer.

Where to Apply and What to Prepare Next

The four founding-engineer roles are listed on the Y Combinator jobs page at ycombinator.com/companies/naive/jobs. Each listing links to an application on workatastartup.com that accepts a CV and a short note. The instruction is explicit: "Send your CV with a short note explaining why this role makes sense for you." That note is not optional fluff; at a company with no recruiting function, founders read every submission themselves, and they notice when a CV is recycled. Mirror the job posting's language. Lead with their problem (autonomous company infrastructure, agent reliability, RL training loops, product velocity, growth loops) not your ambition.

Naive's own careers page at bynaive.com/pages/work-with-us lists different roles (Production Coordinator, Customer Experience, Brand Operations Lead) and invites speculative applications at [email protected]. For the four founding-engineer roles, the YC jobs page is the verified route.

If none of the four posted roles matches your background, the YC page invites a speculative note. At fewer than 50 people, Naive has no recruiting team; the inbox is the system. A generic CV sent to [email protected] will sit unread. A tailored note that references a specific technical challenge from the job description gets a reply. The median B2B application at this stage gets no response ever. One follow-up at day five roughly doubles reply rates. Mark your calendar the day you apply; set a reminder for five business days later. Send a short, polite nudge referencing the original note and any new signal (a shipped feature, a relevant paper, a conversation with a current employee). Do not follow up more than once unless they respond.

After you apply, prepare for a multi-stage process. Industry data for <50-person B2B companies suggests a three-stage loop (intro call, technical deep dive, final round) typically compressing into 7–10 days. Stage one is a 30-minute intro call with a founder or hiring manager testing culture fit and role expectations. Stage two is a 60-minute technical deep dive with a technical founder or lead, focused on past projects and problem-solving approach. Stage three is a 45-minute final round with the founding team covering team fit and offer discussion. There is typically no take-home assignment. The company screens for relevant experience and ability to operate autonomously; at this size they care about self-sufficiency over textbook knowledge. Have two or three concrete project narratives ready: what you built, why the constraints were hard, what you would do differently. Be prepared to whiteboard a system design on the spot. Know the company's product: agents that incorporate businesses, provision infrastructure, and do so.

A final caution: the job market in 2025 is saturated with fake postings. The Federal Trade Commission recorded roughly 105,000 job scams in 2023, a fivefold increase over five years, with losses exceeding $450 million. Scammers clone careers pages and spoof recruiter emails. Verify the domain (ycombinator.com or bynaive.com, not a lookalike) before you upload any document. A legitimate employer never asks for bank details, payment for equipment, or background-check fees before an offer letter. If a "recruiter" pushes you to a Google Form or a personal Gmail address, it is not Naive. The only legitimate endpoints for the founding-engineer roles are the YC jobs page and workatastartup.com.

The same infrastructure that lets a solo founder spin up an AI agency today is what Naive's founding engineers will harden for the enterprises arriving tomorrow. The screen is mechanical, predictable, and beatable — but only for builders who have already shipped what the job description asks for.

Working in AI? Zero G Talent tracks the openings: see every open Naive role, browse AI jobs, the companies hiring, and the people building the field.