Senior TypeScript Engineer (AI agents)
⚲ Poznań
23 520 - 30 240 PLN netto (B2B)
Wymagania
- AI
- Mac OS
- Windows
- TypeScript
- Mobile Development
Opis stanowiska
Why this role
Our current product, an AI chief-of-staff for small-business owners, has paying customers on Mac and Windows.
Now we're building its big brother: an always-on AI team that runs on our servers. A chief of staff and specialist agents work the owner's inbox, calendar and follow-ups around the clock, hand work to each other, and only interrupt the owner for decisions and approvals. The MVP works end to end. Next it goes to real customers.
We are hiring one senior engineer to own its core. Small team, no layers in between: you get the hard problems and a real say in how we solve them.
What you'll do
- Take the platform from a working MVP to real customers: hosted agent runtime, an isolated sandbox per customer, wake on work, sleep when idle.
- Own the parts that must never be wrong: a job queue with leases, approvals that execute exactly the action the owner saw, exactly once, and clean recovery when a process dies mid-task.
- Connect the real world: Gmail and Google Calendar over OAuth first, more integrations after that.
- Keep the agents safe. They read untrusted email all day, so prompt injection is a design input here, not an afterthought: tool scoping, sandboxing, no secret within a model's reach.
- Work AI-native. Most code here is written with Claude Code and Codex, so you brief agents well, review hard and keep diffs small.
Stack
TypeScript on Node.js, SQLite, React, OpenAI Codex as the agent loop, LLM APIs, Railway. Very few dependencies, on purpose. Desktop (Electron) and mobile clients come next.
You
- Senior TypeScript and Node.js. You've built backends that hold real users' state and stay correct under failure: retries, idempotency, crashes mid-write.
- You've shipped a product to real users, ideally paying ones, and kept it running in production.
- You build with AI coding agents every day and can show how you keep them honest.
- You think like an attacker when a system touches other people's data. Isolation and security are part of your definition of done.
- Ownership. You'll often get a problem, not a spec, and you care whether customers stay.
- You aim for the smallest diff that works. Edge cases have to earn their code here, and deleting code counts as a win.
Nice to have
LLM agents and tool calling in production, prompt-injection defence, per-tenant sandboxes (Firecracker, Fly Machines, E2B or similar), Google OAuth and its restricted-scope verification, multi-tenant SaaS, Electron.
What you get
- 140–180 PLN net/h on B2B, full-time.
- The most interesting part of the product, owned by you.
- A direct line to the founder. What you ship reaches paying customers within days, not quarters.
- A short, standard B2B contract at an hourly rate (possible equity), with a 14-day trial period for both sides.
- Your own setup: pick the tools you like working with.
Process
1. A short questionnaire (about 15 minutes).
2. A 30-minute call with our CTO.
3. A build task.
4. A call to walk through your recording and your decisions.
5. An offer.
Our current product, an AI chief-of-staff for small-business owners, has paying customers on Mac and Windows.
Now we're building its big brother: an always-on AI team that runs on our servers. A chief of staff and specialist agents work the owner's inbox, calendar and follow-ups around the clock, hand work to each other, and only interrupt the owner for decisions and approvals. The MVP works end to end. Next it goes to real customers.
We are hiring one senior engineer to own its core. Small team, no layers in between: you get the hard problems and a real say in how we solve them.
What you'll do
- Take the platform from a working MVP to real customers: hosted agent runtime, an isolated sandbox per customer, wake on work, sleep when idle.
- Own the parts that must never be wrong: a job queue with leases, approvals that execute exactly the action the owner saw, exactly once, and clean recovery when a process dies mid-task.
- Connect the real world: Gmail and Google Calendar over OAuth first, more integrations after that.
- Keep the agents safe. They read untrusted email all day, so prompt injection is a design input here, not an afterthought: tool scoping, sandboxing, no secret within a model's reach.
- Work AI-native. Most code here is written with Claude Code and Codex, so you brief agents well, review hard and keep diffs small.
Stack
TypeScript on Node.js, SQLite, React, OpenAI Codex as the agent loop, LLM APIs, Railway. Very few dependencies, on purpose. Desktop (Electron) and mobile clients come next.
You
- Senior TypeScript and Node.js. You've built backends that hold real users' state and stay correct under failure: retries, idempotency, crashes mid-write.
- You've shipped a product to real users, ideally paying ones, and kept it running in production.
- You build with AI coding agents every day and can show how you keep them honest.
- You think like an attacker when a system touches other people's data. Isolation and security are part of your definition of done.
- Ownership. You'll often get a problem, not a spec, and you care whether customers stay.
- You aim for the smallest diff that works. Edge cases have to earn their code here, and deleting code counts as a win.
Nice to have
LLM agents and tool calling in production, prompt-injection defence, per-tenant sandboxes (Firecracker, Fly Machines, E2B or similar), Google OAuth and its restricted-scope verification, multi-tenant SaaS, Electron.
What you get
- 140–180 PLN net/h on B2B, full-time.
- The most interesting part of the product, owned by you.
- A direct line to the founder. What you ship reaches paying customers within days, not quarters.
- A short, standard B2B contract at an hourly rate (possible equity), with a 14-day trial period for both sides.
- Your own setup: pick the tools you like working with.
Process
1. A short questionnaire (about 15 minutes).
2. A 30-minute call with our CTO.
3. A build task.
4. A call to walk through your recording and your decisions.
5. An offer.
🔍 Dekoder Ogłoszenia
🔴
Small team, no layers in between: you get the hard problems and a real say in how we solve them
Może oznaczać dużą odpowiedzialność i presję przy ograniczonym wsparciu.
🟡
always-on AI team that runs on our servers
Sugeruje pracę nad systemem o wysokiej dostępności, co może wiązać się z dyżurami.
🔴
Next it goes to real customers
Projekt na wczesnym etapie; możliwa zmienność wymagań i chaos.