Blog

Notes from shipping AI.

What I learn building multi-agent LLM systems, document AI and real-time computer vision — the decisions that mattered in production.

  1. 15 October 2026 · 3 min read

    Correct isn't the same as good: teaching a multi-agent engine some taste

    Why a reliable multi-agent engine still produced forgettable ideas, and how separate correctness and taste critics fixed it.

  2. 14 October 2026 · 4 min read

    Keeping AI video clips consistent: a shared ledger, a checker and a cache

    How PromptMotion stopped AI video clips contradicting each other with a shared continuity ledger, a checker and caching, and cut LLM cost.

  3. 13 October 2026 · 4 min read

    Why one prompt wasn't enough: splitting PromptMotion into specialist agents

    From one LLM call to a team of agents: how PromptMotion went from a single prompt to specialist agents with film craft, and what it fixed.