01

Adopting AI-assisted development

Let a team work with LLM assistance without speed turning into technical debt.

  • Tooling choices and model routing — which model for which task, and why that is a cost question
  • Operating doctrine: discovery-first, review discipline, clear abort criteria for autonomous runs
  • Quality gates that catch LLM-typical mistakes before they reach the main branch
  • Onboarding the team on real tasks, not toy projects
02

Software architecture & reviews

Make decisions that hold in 24 months — not just until the next sprint.

  • Architecture review of existing systems with named trade-offs instead of diplomatic padding
  • The cut between service and pipeline, monolith and module — soundly reasoned
  • Multi-tenant isolation and scaling questions before they turn into a migration project
  • Data model reviews: the wrong database cut is the most expensive mistake of all
03

From prototype to production

The invisible half of the work — the one that decides whether a demo survives operations.

  • Observability from day one: logs, metrics, trace IDs that actually help in an incident
  • Migration discipline: linear, reversible, without table locks on production data
  • Deploy pipelines with real smoke tests — not just a green build image
  • Incident and rollback runbooks, so the emergency is not a guessing game
04

EU AI Act & GDPR by design

Compliance as an architectural property. Built in before an audit demands it expensively.

  • AI labelling and transparency, built into the pipeline instead of bolted on afterwards
  • Human-in-the-loop at the points where the EU AI Act requires it
  • Audit logs in audit quality: who decided what, when, with which AI involvement
  • Documented legal basis, deletion concepts, access rights — as part of the design
Process

How an engagement runs

  1. 01

    Understand

    I listen, read code and docs, and ask the uncomfortable questions. Discovery before recommendation.

  2. 02

    Assess

    Findings with clear priorities: what holds, what wobbles, what costs later. With named trade-offs.

  3. 03

    Build

    Depending on the engagement: build it myself, build it with the team, or enable the team to do it.

  4. 04

    Hand over

    Documentation, runbooks and a team that can keep working. No dependency on me as a person.

Next step

A project that needs to hold?

Whether a concrete project, an architecture question, or the idea of bringing AI sensibly into development — a first conversation costs nothing but half an hour.