Krzysztof (Chris) Dudek · Senior Software Engineer · Kraków, Poland
I build with AI agents and prove their work.
Through my own company I work with the business to decide what a production application should do and how, then I build it with coding agents. I write the requirements, run the acceptance scenarios, design the user experience and ship the result. I also set up the environment in which agents work autonomously or close to it.
How far can you let agents go?
As far as my proof reaches. What I read myself depends on the risk.
Before
- 2025 to 2026Betsson Group. Migrated core services of a platform with 100k+ transactions a day to .NET 8. Moved its messaging to Apache Kafka.
- 2024EcoVadis and other clients. Contract .NET work. Removed the MSDTC dependency that held back the move to Kubernetes.
- 2022 to 2024Frontiers. Coordinated delivery for 20+ engineers in 2 teams.
Open source
- Yggdrasil (opens in a new tab)
Say it once. I keep my architecture rules next to the code. Yggdrasil gives the agent the rules for the file it is about to edit. It checks every change in a pre-commit hook and in CI. Open source, MIT.
- Jarl (opens in a new tab)
Every finding becomes an issue. Each issue gets a worker in its own copy of the repository, and nothing closes without evidence. I ran the 6.1.0 and 6.2.0 releases of my tools in it. Open source, MIT.
- ResearcherSkill (opens in a new tab)
It asks me what to optimise and how to measure it. Then it runs experiments on its own branch, keeps what works and reverts what fails. Open source, MIT.