Notes on building with AI.
How we build an AI-native company and AI products: agents, context, evaluation and the decisions behind them.
9 October 2026 · 2 min · agents, research
rss feedRunning research as parallel agent lanes
8 October 2026 · 2 min · hiring, agents, responsible AIHow one research question becomes several AI agents working at once, and why giving each of them a narrow lane, a source rule and a fixed answer format is what makes their work usable.
Why our hiring agent never rejects anyone
7 October 2026 · 2 min · context, agents, developer toolsdwij runs the whole hiring process on its own, from the first message to the shortlist. It still never turns anyone down, and a person reviews anything that could count against a candidate. Here is why.
Decisions are the context that matters
Why Relay stores the decisions, constraints and rejected approaches from your work with AI, rather than everything you typed, and why every fact keeps its source.