Method
Grill Me
Matt Pocock's grilling pair: grill-me hands off to grilling. Work a design tree, ask the whole frontier each round with recommended answers, until nothing is silently assumed.
What you get
- Structured rounds instead of scattered questions
- Every question ships a recommended answer you can accept with yes
- ZIP includes both grill-me and grilling
Good for
- A plan or architecture is not coded yet and needs assumptions pressure-tested
- A product decision is stuck and the tree needs to be laid out
- You want a "grill me" session to reach shared understanding
Not a fit
- You are ready to implement. Use writing-plans or start coding
- You have a known bug to fix. Use systematic-debugging
- You need landing-page structure. Use landing-page-design
Practical example
Grill a feature plan until the design-tree frontier is empty
Input / task
Add auto-summaries to an internal knowledge base via a hosted LLM API.
Visible deliverable
Round 1 frontier questions with recommended answers (success criteria, granularity, privacy, failure UX) Recompute the frontier from your answers and ask the next round Facts are looked up by the agent, not asked of you When the frontier is empty, summarize shared understanding and wait for confirm before acting
Deliverables
- Per-round questions with recommendations
- Design-tree / decision log
- Shared-understanding summary after confirm
Acceptance checks
- Each round asks only the current frontier
- Questions include recommended answers
- No code changes before shared understanding is confirmed
Note: examples demonstrate inputs, outputs, and acceptance criteria. Unless explicitly stated, they are not claims of real client work or production testing.