Methodology
The point of view, and the jigs that come with it.
Contrarian where it earns it. Every piece below carries an argument and, where it helps, an embedded jig you can run on your own input.
BRDs still matter — more, not less
The contrarian case: a structured requirements doc is the highest-leverage artifact you have once an AI is writing the code.
Read & try →The multi-model method
How ChatGPT, Claude, Gemini and Perplexity compose into one workflow — not 'which is best', but who does what.
Read & try →Drafting a BRD with a multi-model review loop
A repeatable jig: draft in one model, adversarially review in another, catch hallucinations and scope creep before code.
Read & try →Treating prompts as product artifacts
Versioned, reviewed, reusable prompts that live next to the spec — not one-off chat messages you lose at midnight.
Read & try →Ship a website instead of a deck
The habit that changes stakeholder reviews: when the artifact is a deployed page, people stop imagining and start reacting. Decks describe; working software demonstrates.
Read & try →Hand agents a whole problem, not a task
How to scope a real problem — a voice-notes pipeline, a contact-enrichment engine — so agents can ship it as a working repo instead of a snippet.
Read & try →Judgment, not throughput
A meta-piece on what the PM role becomes when generation is cheap: the scarce skill is deciding what is correct.
Read & try →