Things I changed my mind on. Newest first.
2026 · 04 · 12
I assumed hooking Claude + n8n up to handle customer email and refunds would 10x my throughput. After six weeks the failure modes — quiet hallucinated refunds, mis-categorized urgent emails — cost more cleanup than the time saved. AI-driven ops needs a human in the loop for at least a year past launch. I was treating it like deterministic software when it's still probabilistic infrastructure.
2026 · 02 · 03
For ten years I said you can't ship serious products solo. Then I shipped four in six months alone with AI agents handling everything I used to need a junior engineer for. The advice was right for 2015. It's now actively harmful — it pushes solo builders into bad partnerships when they could be shipping.
2025 · 11 · 18
I thought writing in Polish first would build a stronger local moat. It built a smaller audience and made me spend more time on copy. English first, Polish translations on demand, is the right ratio for builders shipping to a global market with Polish ties.
2025 · 09 · 02
I used to say "ship fast, refactor later." I now think the refactor never comes — the version that ships is the version that lives. Spend an extra two days on data structures and you save six months of regret. Speed is a feature; sloppy isn't.