🧭 Product leader | 🤖 Agentic dev workflows | 🧪 Practical AI guardrails | 🇵🇱 Warsaw

Building for the part where the demo is over and the repo still has to survive.

I'm working on agentic development workflows for passionate product creators: better context, better routing, stronger review loops, deterministic checks, and fewer excuses.

Peter Steinberger says AI agents are "slot machines for programmers." I think for product builders they are a hell of a drug. The trick is adding enough guardrails before the dopamine turns into production debt.

• 🧭 BMADX — BMAD-first Codex guardrails for product-led and non-technical builders.
• 🧪 pyfallow — deterministic Python repo-graph checks agents can run before they claim the code is done.
• 🧿 Oracle — second-opinion workflows for when coding agents need deeper reasoning and context.
• 🐦‍⬛ KOS2 — a local-first Knowledge Operating System for Obsidian.
• 🦞 Secret OpenClaw experiments — personal agents, memory, routing, scheduled runs, and receipts. The world is not ready for such madness.

• I'm testing agentic coding workflows in real repos, not just writing takes about them.
• I care about the unsexy parts: architecture, tests, review loops, routing, and evidence.
• I connect product judgment with developer tooling and AI workflow design.
• A senior architect once told me I understand architecture better than many very good programmers, which is either a compliment or a warning sign. Either way, I keep receipts.
• I'm allergic to AI hype, but still building close enough to the edge to smell the smoke.

You can't make stochastic models deterministic by wishing harder.

But you can shape the riverbanks:

• context,
• tools,
• tests,
• static analysis,
• second-opinion reviews,
• project memory,
• clear guardrails,
• and boring evidence trails.

That is the work: deterministic outcomes in a nondeterministic world.

Before AI, words were mostly words. After AI, words increasingly become code.

Welcome to 2026.