---
title: "Step 3: Make the agent interview you"
step: 3
status: draft
updated: 2026-07-31
description: "Design back-and-forth before code: plan mode, open questions one at a time, forks named and picked by you, approval before edits."
---

For anything non-trivial, the cheapest quality gate happens before the
first edit: a design conversation where the **agent interviews you**,
names the genuine forks, and gets a plan approved. Skipping it doesn't
skip the decisions. It just moves them into the diff, where they're
expensive to find.

**You've met this before.** RFCs and design reviews: decisions on
paper, argued before they're implemented, because a paragraph is
cheaper to change than a system. The agent twist is that the document
becomes a conversation, and the reviewer being interviewed is you.

> Where this comes from: reviewing months of session logs, the pattern
> in the human's interventions was consistent: they were surgical
> corrections *at forks* the agent had driven through without asking.
> The failure isn't wrong answers; it's unnamed questions
> ([AF-15](/agent-workflow-failure-list/)). Putting the fork-naming
> *before* the work is the structural fix.

## What good looks like

- **Plan mode first.** Claude Code has a literal plan mode (the agent
  reads and asks but can't edit). Elsewhere: "don't touch files until I
  approve a plan" does the same job.
- **Open questions, one at a time.** Interview-style beats a wall of
  multiple-choice the agent grades itself. You'll say things the agent
  wouldn't have offered as options.
- **Forks named with a recommendation.** The agent states each genuine
  fork, the plausible readings, and which it would pick and why, then
  waits. Especially for trigger conditions and state transitions, where
  an inferred condition is the classic silent wrong turn
  ([AF-14](/agent-workflow-failure-list/)).
- **The plan is the approval unit.** You approve a plan, not a vibe.
  Scope it to one feature; a plan covering everything approves nothing.

## Try it

<div data-example="unnamed-fork"><a href="/examples/unnamed-fork/">Interactive example: the unnamed fork →</a></div>

## Do it by hand

Next non-trivial task, open with:

- *"Enter plan mode. Interview me about this, one open question at a
  time, until you can state the goal, the non-goals, and the forks.
  Then propose the plan and wait."*

And when an agent presents you options mid-work, notice it: that's the
pattern working. Answer the fork; don't wave it through.

## Or paste this into Claude

```text
Before writing any code for this task, enter plan mode (if you have no
plan mode: make no edits yet). Interview me (one open question at a
time, no multiple-choice) until you can state: the goal in one
sentence, the non-goals, the constraint most likely to bite, and the
two or three genuine forks with your recommendation for each. Where a
trigger or gating condition is involved, restate the exact condition
and the plausible alternative readings, and make me pick. Then present
the plan and wait for approval before touching anything.
```

## Watch out

- **The interview that answers itself**: multiple-choice questions
  where the agent picks a default if you hesitate are forks driven
  through with extra steps. Insist on open questions.
- **Plan-mode theater**: a plan restating the request isn't a plan.
  The tell of a real one is that it contains a decision you had to
  make.