Intent-Driven Development (IDD) is an AI-native approach to software engineering that turns implicit human intent into durable context coding agents can use. As agents take on more implementation work, the bottleneck shifts from writing code to making the intended outcome clear enough to build and verify. section 1.0
A successful prompt is not enough if its important decisions disappear with the conversation. Agent context windows are bounded, and compaction may omit a constraint needed later. IDD uses structured human-agent dialogue to refine intent, then preserves it in specifications, tests, or governing constraints that future sessions can retrieve. These artifacts guide execution rather than leaving each agent to infer purpose from existing code. section 1.3 section 1.8
Specify the password reset before generating it
“Let users reset passwords” leaves consequential behavior open. A user story states the goal: an account holder must regain control after suspected compromise. Acceptance criteria make it verifiable: completing a reset invalidates every existing session; a reset link works within 30 minutes and is rejected afterward. A test can create sessions on several devices, complete the reset, and assert that no active sessions remain. The agent can implement against that intent instead of treating whichever behavior it first generated as the requirement. section 1.2
Keep the control surface aligned as the product changes:
- Change the intent artifact when the desired behavior changes, then use it to guide implementation and verification. A Windows log-cleanup script is a poor source for a Linux replacement; the original rule about deleting old logs only when an archive exists travels better. section 1.1
- Make specifications shared team context, not documents passed between roles. For a URL shortener, they can preserve why shortcodes are never reused and URLs remain permanent while directing new agent sessions to relevant code, tests, and guidance. section 4.0
Spec-Driven Development applies IDD by making specifications the primary control surface. Behavior-Driven Development can make those specifications verifiable, while Test-Driven Development can guide implementation details. Autonomous iteration workflows offer a complementary route: governing constraints steer repeated execution and convergence. IDD requires continued investment for fixes and features; for an urgent demo or proof of concept, a less structured approach may be preferable. Its long-term value also depends on stakeholders treating artifacts as shared intent, not paperwork. section 1.2 section 1.6 section 1.8
