Vibe Coder and Developer

Senior Product Engineer (Contract, 20–40 hrs/wk, remote) MARTHA is building the first AI-native agency for paid social. Today it's a working product running real ad spend for real clients: creative generation, optimization recommendations, campaign launch into Meta, and client reporting. The direction is an autonomous system where agents run the accounts and a human supervises and approves. We're a very small team shipping fast. We're looking for one engineer who can own whole features end to end — and who can be trusted not to break a product that has money moving through it. What you'd actually be doing You own vertical slices, not a layer. A typical piece of work touches a React page, a FastAPI endpoint, a Celery worker, and a Terraform module — and you're expected to carry it across all four, then verify it in a real environment. First 90 days, the work is concentrated on one theme: making MARTHA scale from a handful of accounts to many without engineering time per account. In practice that means onboarding that works without us in the loop, per-tenant configuration instead of hardcoded client behavior, and the test and deploy infrastructure that lets us ship daily without breaking live ad accounts. Requirements Must have: Production Python. FastAPI or equivalent async framework, SQLAlchemy, real migration discipline. You've run mypy --strict on something that mattered. Production TypeScript/React. You can work confidently in a large existing component tree you didn't write. AWS and infrastructure-as-code, hands-on. You can read a Terraform module, change it, and deploy it. You do not need a platform engineer to ship. You work with AI coding tools daily and you review their output critically. Speed matters here. So does knowing exactly where a model will produce something that looks finished and isn't. Third-party API integration experience — OAuth flows, token refresh, rate limits, async jobs, partial failure, webhooks. Ad platform APIs are unforgiving and we live on three of them. You verify your own work before calling it done. See below — this is the single biggest thing we're screening for. 4+ hours of daily overlap with US Pacific, and you respond within the day. Strongly preferred: Meta Marketing API experience specifically (campaigns, ad sets, creatives, Insights, Ad Library). Shipping LLM features to production: managing non-determinism, evaluating output quality, controlling cost and latency. Multi-tenant SaaS — data isolation, per-tenant config, credential storage. Paid social / performance marketing literacy. You don't need to be a media buyer, but knowing what ROAS, CPA and a break-even threshold mean means we spend our time on engineering instead of vocabulary. What actually gets you hired Most candidates will tell us they ship fast with AI. We believe you. Here's the part that decides it. The recurring failure mode in a codebase moving at this speed is not a crash. It's a feature that is fully modelled, fully stored, fully validated, covered by passing tests — and read by nothing. The field exists. The schema is right. The logs are clean. It just doesn't do anything. Nearly every serious bug we've hit in production has been some version of that: a write that never crossed a transaction boundary, a value computed and never rendered, a stub route shadowing the real one, state held in one process while another process reads it. Green tests didn't catch any of them, because the tests stubbed out precisely the seam the bug lived in. So we're hiring for verification discipline as much as coding speed. The engineer we want is the one who, after building the thing, goes and checks the seam: boots from a clean manifest, reads the actual bucket policy, sends one real request over the wire, greps for who consumes the field they just added. Someone who says "tests pass but I haven't verified X yet" instead of "done." If you read that and thought yes, obviously — you're who we want. If you read it and thought it sounds slow, we're not a fit. How we work Small, direct, no ceremony. No standups you don't need. Branch per change, PR, CI runs the test suites, staging before production. Decisions get written down. We keep a running build log so context survives. You'll have real autonomy and real accountability. Nobody is checking your hours; we're looking at what shipped and whether it held up. Money and live ads are involved, so anything that spends budget or publishes an ad goes through human approval. That's a hard rule, not a preference. Terms Contract to start, hourly or per-milestone — your preference. Begins with one paid scoped project (roughly 1–2 weeks) so we both find out whether this works before either of us commits further. Goes well → ongoing retained hours, growing scope, and a real conversation about a longer-term or equity-inclusive arrangement. We're looking for someone to grow with, not to burn through. Tell us your rate. We'd rather pay properly for someone who ships correctly than cheaply for someone we have to check. To apply Send: Two things you built and shipped, ideally where you owned the full slice. Links to code are better than descriptions. A short answer to this (a few paragraphs is plenty — we read every one): Describe a bug that reached production despite passing tests. What was the seam the tests couldn't see, how did you find it, and what did you change so that class of bug couldn't come back? Your AI-assisted workflow — what tools, and how you check the output. Rate, weekly availability, timezone. We reply to everyone who answers question 2 seriously.