The problem

Administrative teams make hundreds of small decisions every day - approvals, classifications, routing, follow-ups. Not complex individually, but collectively they eat entire workdays. No tool solves this well, because the hard part isn't automation - it's judgment.

We're building an AI agent that doesn't just follow scripts, but learns from how a team works, prepares decisions, executes autonomously where appropriate, and gets meaningfully better over time through structured feedback. A virtual colleague that prepares and takes over repetitive tasks.

The product exists as a solid UX prototype and architectural direction. Now we're looking for the person to shape and build it.

The role

You'll work directly with the Product Lead. No middlemen, no ticket ping-pong. You get context, you make decisions, and you see the impact of your work within days.

The interesting part of this role isn't frontend polish — it's the stuff underneath. How does the agent decide when it's confident enough to act on its own vs. when to ask a human? How do you turn messy user corrections into structured learning signals without overfitting to one person's preferences? How do you evaluate whether the system is actually getting better — and catch it when it's not?

You'll spend your time building a full-stack app with the underlying agent orchestration, knowledge logic, feedback loops, and evaluation pipelines.

Tasks

What you'll build

• A new product from the ground up using a modern AI stack
• Autonomous AI agents that identify, plan, and execute tasks
• Agent steering logic — deciding what to do, when to act, when to escalate
• Evaluation and validation frameworks to measure whether the agent is actually getting smarter
• Human-in-the-loop workflows where the AI prepares and the human decides
• APIs, database logic, and deployment pipelines

Your first weeks

In your first two weeks, you'll dig into the concept, challenge assumptions, discuss the best architecture and start building the first vertical slice of the agent pipeline. By week six, you'll have shaped core architectural decisions and shipped something real users can interact with.

Requirements

Must-have

• You've built a system where AI agents autonomously plan and execute tasks
• Experience with React and Next.js (App Router, Server Components)
• Self-driven work style — you find solutions before someone tells you there's a problem

Highly valuable

• Understanding of feedback loops, evaluation pipelines (e.g. Arize Phoenix or Langfuse), or RLHF concepts
• Experience with LLM APIs (OpenAI, Anthropic) and streaming architectures
• Awareness of data privacy and EU hosting requirements

Nice to have

• Docker
• Infrastructure experience

Not required

• CS degree
• ML / model training
• 10 years of experience with a 3-year-old framework

Language

Lucid Labs is a German team - a lot of our internal communication is in German. For this role, though, you'll work directly with the Product and Tech Lead, and that collaboration (your Slack channel, meetings, code, PRs) happens in English. German is a plus, but not a requirement.

Benefits

Why Lucid Labs

• Your own product — no agency-hopping, no legacy code
• Clear roadmap — first milestone in 6–8 weeks, then we iterate with pilot customers and build based on real user feedback
• Remote-first — async by default, minimal meetings, output over hours
• Flexible — freelance or permanent, starting at 20h/week
• Growth — the project is growing, and your role can grow with it
• Tooling — Cursor Pro, Claude and OpenAI credits

About us

Lucid Labs is an AI-first studio founded by serial entrepreneur Marek Janetzke. We build product and consulting solutions with measurable impact for German startups, mid-market companies, and enterprises.

Our culture: Ownership over micromanagement. We'd rather make a smart bet in two weeks, learn from it, and course-correct - than spend three months polishing something nobody validated. We value people who think along, not just work along.

Working model

• Scope: 20+ hours/week
• Start: As soon as possible
• Duration: Long-term
• Location: Berlin or remote (EU)
• Syncs: 1–2x per week, otherwise async
• Model: Freelance or part-time — both work

**
Sounds interesting? Apply now - we'd love to hear from you!**