Notes from the workbench

Thinking out loud. Building with care.

What I’m trying with AI and software, and what I’m learning from the work.

  1. The decisions I put into tools instead of AI prompts

    How Genny and LaiVE apply settled engineering decisions while leaving product behaviour and musical choices open.

  2. AI speed is useful when the result is useful

    Would you trust a reorder list because AI built it quickly? A hypothetical inventory filter shows which decisions make the result useful.

  3. Small tools can make a big difference

    Where should a small tool’s job end? ShowIP and A Minor Idea offer two different answers: a network question and a melody worth keeping.

  4. What I hand over between AI planning and coding sessions

    A worked customer-search brief shows which decisions and reasons need to survive the handoff to a fresh coding session.

  5. Before choosing an AI model, decide what job you’re giving it

    Define the investigation before comparing AI models, including the evidence that makes an answer useful.