Andrey Balyasnikov — writing https://balyasnikov.com/writing Notes on products, infrastructure and applied machine learning. en-us Sun, 16 Aug 2026 00:00:00 GMT Eight years of running product remote-first: what actually worked https://balyasnikov.com/writing/eight-years-remote-first https://balyasnikov.com/writing/eight-years-remote-first Sun, 16 Aug 2026 00:00:00 GMT The remote operating system we ran at Zerion, from a one year direction down to a single ticket. Agent-built interfaces drift into slop. Here is how I stop it. https://balyasnikov.com/writing/interfaces-with-coding-agents https://balyasnikov.com/writing/interfaces-with-coding-agents Wed, 12 Aug 2026 00:00:00 GMT The anchors that keep agent-built interfaces consistent: global principles, a versioned spec, a design system, and review at every stage. How I rebuilt product work around coding agents https://balyasnikov.com/writing/product-work-around-coding-agents https://balyasnikov.com/writing/product-work-around-coding-agents Mon, 03 Aug 2026 00:00:00 GMT Company strategy, evidence, product bets, PRDs, delivery and learning in one system that people and agents both read. Every iteration gets a versioned spec https://balyasnikov.com/writing/every-iteration-gets-a-spec https://balyasnikov.com/writing/every-iteration-gets-a-spec Wed, 12 Aug 2026 00:00:00 GMT Each feature gets a versioned spec with its own implementation history, so agents pick up context instead of starting from zero. When the aggregate metric says there is no problem https://balyasnikov.com/writing/aggregate-metric-said-no-problem https://balyasnikov.com/writing/aggregate-metric-said-no-problem Fri, 07 Aug 2026 00:00:00 GMT How segmentation found a 15% failure rate inside a metric that looked like noise, and what kept the fix honest.