Anyone can build a demo
Running it is the job

Senior technical leadership on a part-time or project basis: architecture, roadmap, delivery ownership, team and vendor selection, and AI strategy. The role stays technical, not purely advisory. In practice that means code reviews, architecture decisions, and production debugging alongside the team, without the cost of a full-time hire.

A demo agent and a production agent are different problems. In a demo it answers once, watched by a human. In production it acts on its own, repeatedly, against changing data. The failure modes are looping, hallucinated actions, runaway cost, and silent regressions. Reliability is the work that makes that safe: evals you can trust, guardrails, bounded autonomy, monitoring, and a clear rollback path. It is the part most teams skip.

Both. Hosted APIs are often the fastest start. But when data has to stay in your infrastructure, when per-token cost does not scale, or when you want to avoid vendor lock-in, I build on open models like Llama, Mistral, or Qwen, self-hosted on infrastructure you control. I run open models in production today, so this is not theoretical.

The engagement includes code. I work inside your repository, not from the outside, and I stay on to operate what we ship. Depending on the engagement, that spans implementation, architecture reviews, production debugging, and ongoing reliability.

Not every problem needs AI, and not every AI needs an agent. I start with the use case: what problem are you solving, and does AI change the outcome enough to justify running it? Most AI failures are framing failures. The technology is rarely the hard part. The Readiness & Production Audit answers this before any build work begins.

It is part of the Readiness & Production Audit. I assess whether a proposed AI system falls under the Act's scope, which risk category it lands in, and what compliance obligations that creates. It also covers data protection and governance, scaled to your size and intended use. For live systems, the Reliability Retainer re-assesses this as the rules and your usage change.

Unsure whether AI is the right move? Start with the Readiness & Production Audit: fixed scope, a use-case shortlist, an honest build/buy/skip call, and a cost model. Already know what to build? A Build Sprint ships one system to production. Already shipped? The Reliability Retainer keeps it running.

I am based in Berlin and have lived here since 2007. I work with startups and growing companies regardless of location. Email me at contact@larsullrich.de. I read every message personally and usually reply within a day.