AN
Warsaw, PL

Writing

Writing

Notes from building distributed systems, ad tech and my own products: what broke, what held and what I’d do again.

Engineering

  1. Our users moved to Sweden overnightAnalytics said a big share of Paperspell's users suddenly lived in small Swedish towns. They weren't bots and they weren't Swedish: a floating SDK version had started sending events from 127.0.0.1.Product analytics · PostHog · Flutter · Mobile · Debugging
  2. Cookieless isn't anonymousChrome kept third-party cookies, yet ads still recognise you. How Unified ID 2.0 turns the email you log in with into a token, what genuinely improved, and what didn't.Privacy · AdTech · Cookieless · UID2 · Data protection
  3. ARTF: the idea is right, the contract works against itIAB Tech Lab's Agentic Real-Time Framework turns the ad auction into a platform for agents from any vendor. Four places where its contract still hurts, and why MCP belongs on the orchestrator.AdTech · RTB · AI agents · MCP · System design

Experiments

  1. 13 cassettes and a program that listens for meSix hours of family tapes from 1980 to 2000, catalogued on a CPU: songs, talk, dates said out loud, and one song found on five different tapes. What real recordings taught me about thresholds and hallucinations.Python · Whisper · Audio · Open source · Side project
  2. My AI memory, my rules5,362 ChatGPT conversations that Claude knows nothing about. An experiment in a memory that belongs to me rather than to an AI provider: plain Markdown in a private git repository, readable by any assistant. The design, and what a 5% pilot showed about cost and quality.AI · Personal AI · Knowledge management · LLM
  3. The NDA was in a git branchA friendly recruiter, a real-looking healthcare project and an NDA that could only be read with git checkout. Inside the tarball: a post-checkout hook that would have run a stranger's script on my laptop. What the hook does, which git commands trigger it, and how I open such archives now.Security · Git · AI agents · Hiring scams
  4. I ran 104 of my emails through Jev. I'm still not switchingJev, the new model from TypeSafe AI, answers typed questions with calibrated probabilities instead of text. On 104 of my real emails it was four times faster than Gemini Flash-Lite and nearly as accurate, and as a pre-filter it would have skipped 35–42% of LLM calls. Why my email agent still doesn't use it.AI · LLM · Open source · Privacy · Email agent

Life

  1. RUНе учить, а участвовать: зачем мне «Кола»Как программист из Варшавы собрал мужской круг у костра на Висле: откуда взялась идея, чему я учился как фасилитатор, как прошли первые две встречи и почему я в этом круге не учитель, а участник.Кола · Эмиграция · Группы равных · Варшава
  2. AI and the human: a conversation with Anthropic's AI InterviewerAnthropic Interviewer, Claude asking questions for Anthropic's research team, asked me how AI is changing my work and my life. The full conversation, translated from Russian: why running several agents at once feels almost addictive, what I don't want AI to change, and who should share in what it creates.AI · Anthropic · Work · Society
  3. RUМоё интервью с Anthropic об искусственном интеллектеAnthropic Interviewer, Claude, который задаёт вопросы от имени исследовательской команды Anthropic, расспросил меня, как ИИ меняет мою работу и жизнь. Полный текст разговора: почему от агентов так трудно оторваться, что я не хотел бы, чтобы ИИ менял, и кому должны достаться его блага.ИИ · Anthropic · Работа · Общество

RSS feed

Let’s talk

Open to new opportunities.

Consulting, part-time or full-time. If you have an interesting project or just want to say hi, I’d love to hear from you.

Send a message LinkedInGitHubCV