AI-generated missions

Claude Code and Codex can generate missions far faster than a human can read them. A single session can produce dozens of .intent files. That speed is the point, and it is also the risk: volume hides what changed, what is risky, and what is unverified. IntentLang is the layer that keeps AI-generated intent reviewable.

Part of the family of concepts for working with large changes.

The problem with volume

When an agent writes 47 missions, the failure mode is not that any single mission is wrong. It is that:

• No human reads all 47, so risky ones slip through.
• "It compiles" and "the tests pass" say nothing about weakened never rules.
• The intent and the code can diverge silently over a day of edits.

Speed without a review model produces confident, unverified software.

How IntentLang keeps it reviewable

Every mission records its provenance: whether it was AI-authored and whether a human reviewed it. That metadata flows into the tools that make volume tractable:

• The Build Session Digest says what the AI generated and modified this session.
• The Risk Radar counts AI-generated code and unreviewed as risk factors, so AI-authored, unreviewed missions rise to the top.
• The Proof Matrix shows which AI-generated missions are actually verified.
• OpenThunder checks the generated code against the declared intent (drift).

The discipline

• AI proposes, humans approve. A mission carries aiAuthored and humanReviewed flags; intent approve records the human sign-off.
• Verification is not optional. An AI-generated guarantee with no test is a claim, not a proof. The Proof Matrix makes that visible.
• Review by risk, not by order. Use the Risk Radar; do not read 47 missions top to bottom.

Worked example

In the customer portal example, most missions are aiAuthored: true and humanReviewed: false. The Risk Radar surfaces the AI-authored, unreviewed billing missions first, and the MVP Readiness report keeps the product at demo_safe until they are verified. That is the whole point: move fast, and still know what you can trust.