I hire, onboard and lead AI agents.
The model is rarely the hard part. The hard part is deciding what an agent should do, what it must never do, and keeping it running when nobody is watching.
How a project runs
04 steps- Pick the process. Which processes are worth handing to an agent — and which are not. Frequency, cost of a mistake, and whether the data is there decide it.
- Knowledge, tools, limits. Each agent gets its own instructions and knowledge files, only the tools it needs, and limits enforced in code — read-only by default.
- Connect it to the company. Slack, project tools, meetings, ad platforms, CRMs: the agent works where the team already works, and a human approves where a mistake costs trust.
- Run it 24/7. Watchdog every minute, automatic rebuilds, restart-loop protection, queues so no task is lost — and blind tests so every change is measured.
Principles
settingsAgents propose, people confirm. Each decision becomes data, so clear cases can be automated later.
Permissions are set in code, per client — read-only by default.
Blind tests instead of gut feeling: the judge never sees the source. 56 documented test cycles.
Watchdog, automatic rebuilds, restart-loop protection, and knowledge kept in files — nothing is lost after a restart.
Background
since 2006I have worked remotely for US companies since 2006 — first in digital marketing and web development for aesthetic surgery clinics, then for a US marketing agency for home-service businesses, where I have led AI automation since 2024.
I have used generative AI every day since late 2022. Before agents, I built an AI content engine (2023), a white-label platform with 30+ AI assistants and a RAG chatbot platform for clinics (2025). Today I run a fleet of 38 Claude Code agents.
Confidential work runs locally on my own GPU; cloud models only where the data allows it.
- Based inBerlin — remote or on-site
- LanguagesGerman (C2), Spanish (native), English (professional), Italian (fluent)
- EducationM.A. Technology & Innovation Management · Degree in Systems Engineering
- AvailabilityProjects and permanent roles
Tools I use every day
stackFrequently asked questions
faqWhat does an AI agent engineer do?
I decide which processes are worth handing to an AI agent, give each agent knowledge, tools and clear limits, connect it to the systems the team already uses, and keep it running 24/7 with monitoring and self-healing.
Do you run AI agents in production?
Yes. I run a fleet of 38 Claude Code agents on two servers. 18 of them are client-expert agents for a US marketing agency for home-service businesses, in daily use.
How do you stop agents from making costly mistakes?
Limits are enforced in code, not in the prompt: agents are read-only by default, and a person approves wherever a mistake costs money or trust — for example every task proposed from a client meeting.
How do you measure whether AI output is good?
With blind tests: a judge that never sees the source retells what it understood, and positive controls test the judge. In my own evaluation harness this took comprehension from 3.3 to 7.8 out of 10 over 56 documented cycles.
Can you work with confidential data?
Yes. Confidential work runs locally on my own GPU with no outbound connection; cloud models only where the data allows it. Example: 266 customer calls analysed without the audio leaving the machine.
Which tools and models do you use?
Claude Code and the Anthropic API every day, plus the OpenAI and Gemini APIs, Node.js, Python, TypeScript, SQLite and PostgreSQL, tmux, pm2, Docker, Playwright, and local models with ollama and faster-whisper.
Are you available, and where?
Yes — for freelance projects and permanent roles. I am based in Berlin and work remotely or on-site, in German, English or Spanish. Write to [email protected].