Full-Stack Engineer (AI-Native) — AgentX

Who we are

AgentX is a platform for building and running production AI agents — and a team that ships real systems on top of it for clients who need more than a demo.

The platform is the infrastructure: orchestration, memory, tools, evals, deployment — the layer that makes agentic systems actually work in production, not just in a notebook. On top of it, we build end-to-end solutions for our clients. One of the systems running on AgentX today is KsięgoBoty — a multi-agent product used daily by hundreds of Polish accounting firms. Paying customers, production load, real consequences when something breaks.

We've also done to ourselves what we build for others: restructured how we work around AI tools, agent-first architectures, and a delivery model that doesn't need a Jira layer between thinking and shipping. This isn't theory — it's how we operate every day.

Who we're looking for

A senior engineer who owns outcomes, not tasks.

This isn't a role for someone who's shipped "an LLM demo at a hackathon." We need someone who has already put agentic systems into production — with real users, real edge cases, real 2am Slack messages when something goes sideways.

If you need a PM to translate the problem, a Jira board with three rounds of grooming, and acceptance criteria to tell you what "done" means — this isn't for you, and that's fine.

If instead you read a Slack message like "accountants are losing 2 hours a day reconciling this" and your brain immediately jumps to "okay, what does the user actually need, what's the simplest thing that works, and how do we ship it" — keep reading.

You'll own problems end-to-end. Frontend, backend, the agent layer, the deploy, the follow-up when a customer hits something weird. You decide what to build, you build it, you ship it, you watch what happens.

And you get that this actually matters. On the other end of every feature there's an accountant closing books at 9pm, a firm owner trying to keep their business running. It's not just code.

What "AI-native" actually means here

Not "I've used ChatGPT." We mean:

• Claude Code / Cursor / Codex is your IDE, not a plugin you switch on when you feel like it. You know when "accept all" is safe and when it absolutely isn't.

• You have strong opinions on agent orchestration — when to reach for LangGraph/CrewAI/Agno vs. when to write 50 lines of plain Python that work better.

• You understand production reality of LLM systems — guardrails, evals, observability, retry logic, cost control, what to do when a model is non-deterministic in a way that breaks your contract with the user.

• You can tell when a model is the wrong tool and reach for code instead. For you, AI is a tool — not the point.

What you'll actually own

• Agentic systems in production — multi-step workflows, RAG pipelines, agent orchestration. From architecture decision to deployment to the 3am on-call when something drifts.

• The AgentX platform itself — the primitives, the orchestration layer, the evals, the deployment story. You're building the thing that makes every client engagement faster and more reliable than the last.

• Full-stack client solutions on top of AgentX — TypeScript/React on the frontend, Python on the backend, integrated with LLM providers (Anthropic, OpenAI, others as they come).

• Infrastructure that holds up — Postgres, AWS, vector stores, ingestion pipelines, the unsexy layer that makes everything else possible.

• Quality standards for AI systems — you help us define what "good" looks like. Evals, guardrails, failure modes. Because "mostly right but wrong in scary ways" is not a feature.

Our stack

TypeScript, React, Node.js, Python, PostgreSQL, AWS, Anthropic + OpenAI APIs, agentic frameworks. You don't need every box ticked — you need to be the kind of specialist who picks up unfamiliar pieces quickly and confidently.

What we expect (must-haves)

• 5+ years shipping production software, at least 1-2 of them with real LLM systems in production — not side projects, not demos

• Strong full-stack background — you're comfortable owning from frontend to infra, even if your center of gravity is one side

• AI-native coding workflow — Claude Code, Cursor, Codex, whatever — as your daily driver

• Hands-on with LLM APIs, RAG, prompt engineering, and at least one agentic framework (or you've built your own orchestration and have opinions on why)

• Strong written English — you'll communicate with clients, partners, and async teammates

• Ability to run a project from discovery to delivery with minimal supervision — we'll align on outcomes, you'll figure out how

Nice-to-have

• Experience building B2B SaaS for non-technical users (SMBs, professional services)

• Voice agents, real-time systems (LiveKit or similar)

• Accounting / fintech / regulated domain exposure

How we work

Fast pace, but Monday to Friday. Weekends are yours. We don't do crunch, we don't do fake urgency, we don't do meetings that could've been a message. We'd rather ship steadily for years than sprint for six months and lose you.

Direct line to the founder. Direct line to clients. No process layer between you and the work. Small team, real decisions, real ownership.

The honest pitch

• Real ownership of real product areas — not a seat on a feature factory

• Frontier work — you're building both the platform that makes production agents work and the client systems running on top of it

• Compensation that reflects seniority — 14 000 – 25 000 PLN netto + VAT (B2B), with room above for someone exceptional

• Fully paid AI stack — Claude Max, Cursor, whatever you need to work at full speed

• Remote-first, flexible hours, Poland-based

• No filler meetings, no Jira bloat — we measure shipped systems in production, not hours logged

• Push back on the founder when he's wrong. He will be, regularly.

How the process works

1. You send us stuff recruitment@agentx.so

• Your CV — yes, we read them

• One thing you built end-to-end you're proud of — and what you'd change about it today

• One AI product (yours or someone else's) you think is genuinely good, and why

• Your take: a client says our agent is "mostly right but wrong in ways that scare them." What do you do first?

2. We reply within 2-3 days — either "let's talk" or "not this time, here's why." No black hole.

3. We meet — 30-45 min call with the founder. We talk about your work, you ask us hard questions. If the chemistry isn't there, no hard feelings.

4. Problem-solving session (top 2-3 candidates) — we give you a real problem we're working on. You come back with how you'd approach it: your thinking, your trade-offs, your questions. We're not watching over your shoulder. We want to see how you think.

5. Decision within 2-3 days. If yes — you start when you're ready. Tomorrow, next week, after notice period. Your call.