← Jobs

AI Generalist

We’re hiring an AI Generalist to work directly with the founders on whatever matters most. Own a meaningful problem. Use the best tools available. Bring it through to a result.

Reports to
The founders, directly
Location
Austin, TX or remote
Type
Full-time or part-time
Compensation
Salary and equity
Status
Open to the right person. We hire when the fit is right, not on a calendar.

Why this role exists

Our hardest problems don’t sit inside one department. A records request that stalls is an operations problem, a product problem and a sales problem at the same time. Hire a specialist for each and you get three handoffs and no owner.

So we hire people who can carry a problem across those lines and finish it. Former founders and early employees tend to be brilliant at this: they’ve had to improve whatever was broken, fast, without waiting to be the world’s expert in it.

The work

Mostly, the problems are ours: how we run, sell, price, onboard and measure. A pricing experiment. A revenue process that should run itself. A manual step in onboarding turned into an automated one. The dashboard the founders check every morning. Sometimes it’s a firm’s problem that nobody else owns yet. You won’t just take tickets; you’ll help decide what gets built.

You’ll own outcomes, not a function. You’ll build the queries, dashboards, automations, agents and lightweight tools yourself, rather than coordinating someone else to build them, and leave behind systems that keep running when you move on to the next problem.

  • Understand → build. Talk to the people doing the work. Find the bottleneck. Prototype the fix and test whether it helps.
  • Manual → repeatable. Turn a recurring task into a reliable workflow, with checks, documentation and a clear owner.
  • Idea → evidence. Put a product or commercial bet in front of real firms. Use what happens to decide what’s next.

How we think

  • Leverage is the job. Code and models are leverage that needs nobody’s permission. A person with the right tools should do the work of a team.
  • Specific knowledge compounds. Learn how a case really moves and you’ll see problems no one else can. That can’t be taught in a course; it can be learned here.
  • Judgement over effort. The right problem solved simply beats the wrong problem solved heroically.
  • Compounding takes time. We’d rather build something durable with a few people who stay and get better than something fast with many who don’t.
  • Simple beats clever. If a rule does the job, don’t use a model. If a model does the job, don’t build an agent.

Who this is for

Any background, any path. Engineering, sales, product, operations, research, law, or something with no name yet. You don’t need a particular degree, career path or previous title. You do need evidence that you can understand a difficult problem, learn what you’re missing, and bring something useful into the world, and a strong pull toward AI that you already act on every day. Startup experience is strongly preferred, not required. Former founders and early employees tend to do especially well.

  • You’ve made something meaningfully better and can explain exactly how.
  • You ask the right question before reaching for a solution.
  • You pick up new domains fast. This one is legal and medical.
  • You’re as clear with an engineer as with a managing partner.
  • You know the difference between a demo and a system people rely on.
  • You finish, and you check whether it actually helped.

AI fluency

If you live in this world, most of the following will be familiar. You don’t need all of it, and equivalent tools count. You should recognise most of it, have opinions about some of it, and have shipped something with a few of them. Naming tools isn’t the signal; what you built with them is.

  • Claude
  • Claude Code
  • GPT
  • Gemini
  • Grok
  • Cursor
  • Codex
  • Open-weight models (Llama, Qwen, Mistral)
  • Ollama / vLLM
  • Context windows
  • Tokens and cost
  • Prompt caching
  • Structured outputs
  • Tool use
  • MCP
  • Agents and sub-agents
  • Evals
  • Embeddings
  • RAG
  • Rerankers
  • Vector databases (pgvector, Pinecone)
  • Graph databases (Neo4j)
  • Postgres
  • OCR and document AI
  • SQL
  • Python / JavaScript
  • n8n / Make / Zapier
  • Retool
  • LangGraph
  • APIs and webhooks
  • Retries and idempotency
  • Access controls

What that looks like in practice: you can trace a system from input to output and explain why a model got something wrong (the context, the retrieval or the prompt), then fix it. You know when long context beats retrieval, and when a plain rule beats both. You build an eval before you trust an output. You’ve thought about what happens when an API times out or an event fires twice. You know exactly what data leaves the building, and when a model should run locally instead. You can ship a working tool with Claude Code in an afternoon, and you know what it would take for someone else to run it without you.

Judgement

We handle medical records and legal work. Every fact we produce is cited to its source page, and firms approve every step. Knowing when to verify, when to bring in a specialist, and what never goes into a third-party tool is part of the role, not a footnote to it.

What this isn’t

  • A prompt-engineering role.
  • An administrative catch-all.
  • A job whose value is sitting next to the founders. The access matters because of what you’ll be trusted to own.
  • A demand to be expert in every discipline. You’ll be expected to own something that matters, not everything.

Applying

Send a thoughtful cold introduction to [email protected].

Show us what you’ve built and automated with AI: links, repos, screen recordings or a few paragraphs. Tell us what the AI did, what you decided, how you knew it was good enough, and one thing that failed. Then tell us what you’d want to take on here.