Who writes this
I'm Kirill Balakhonov — I build and run AI agents daily: multi-agent pipelines that ship real code, credential boundaries for agents that touch real money, and content systems with real audiences. By day I work in security engineering; this site is a personal project and reflects my own experiments, not my employer's views.
Why this site exists
Most writing about AI agents is either hype or doom. What's missing is the middle: what actually breaks when agents run unattended for weeks, what it costs, and what fixes hold. Every article here comes from a production incident, a real dataset or a reproducible experiment — with receipts.
Some articles end with an offer: an idea for a service that would fix the pain described. If enough people say "I want this and here's what I'd pay", it gets built. If not, it stays an article. That's the whole business model, and it's stated openly.