// Section 10 — Human and Vibe review · 3 MIN READ

[✓] VERIFIED MANUAL ENTRY — This concept has been rewritten from primary sources and is legally cleared for production.

Grilling

A planning technique where the agent Socratically interviews the developer, one decision at a time, to resolve ambiguities before committing to code or specs.

Grilling is a planning technique designed to uncover hidden assumptions before code is written. Instead of having the agent immediately write code or specs from a brief prompt, the agent is instructed to act as a Socratic interviewer: asking you questions about edge cases, data constraints, and business logic, and recommending options for each.

You trigger grilling whenever you:

  • Use the /grill-me slash command in your chat interface.
  • Instruct the model: "Do not write code yet. Interview me about the requirements for this payment routing system."
  • Review a list of questions generated by an agent about database constraints.

Stopping the Silently Guessed Spec

AI agents are built to please: if you ask them to write a design doc, they will fill all decision gaps with their own default guesses. The result looks complete and correct, but incorporates design choices you never actually chose.

Grilling inverts this loop:

  • The Checkpoint: The agent is blocked from writing files until you align on the mental Design Concept.
  • Active Interview: The agent asks clear, one-by-one questions (e.g. "How should we handle users who log in with stale tokens? Option A or Option B?").
  • Durable Output: Only after all answers are logged does the agent compile them into a Spec or code file.
Initial Idea ──► Grilling Session (Q&A Interview) ──► Stabilized Spec ──► Code Implementation

Field Applications & Prompt Templates

1. Developer Prompts (Starting a Grill Turn)

To start a grilling turn, paste this template into the chat before detailing your feature:

  • Template:
    I want to build a new payment subscription model. Do NOT write any code or specifications yet. 
Instead, grill me: ask me Socratic questions one by one about our subscription parameters, refund logic, and edge cases. Propose options for each question.

# AVOID

Do not accept a generated code file if the model guessed the business rules instead of asking you.

  • Avoid: Letting the model write the login class if it guessed the password hashing algorithm.
  • Write: Command the model to pause, interview you about security requirements, and then implement the class.

# USAGE

Developer A: "The agent implemented a shopping cart but it doesn't handle currency changes properly." Developer B: "We rushed straight into code. We should have grilled the model first. Let's restart, have the model interview us about currency conversion rules, and write those rules down before we write the code."

// SEE_ALSO

Design ConceptThe shared mental model of what is being built, held in common between developer and agent, separate from any physical file or code asset.
SpecA high-level handoff artifact (like a PRD or design doc) stored in the environment that defines a project's goals, constraints, and ticket checklist across multiple sessions.
Human-in-the-loopA working pattern where the developer actively monitors, redirects, and collaborates with the agent in real time, catching mistakes before they build up.

// SOCRATIC_VALIDATION

Interactive Concept Quiz

QUESTION 1 OF 3SCORE: 0/3

[EDIT_THIS_TERM_ON_GITHUB ↗]