title: "16 · The Three Waves" description: "From autocomplete, to chat, to agents. Where you sit on the wave changes what "good" looks like."
The Three Waves
From autocomplete, to chat, to agents. Where you sit on the wave changes what "good" looks like.
Three waves, three mindsets
We've lived through three distinct waves of developer AI in a very short time.
Wave 1, Autocomplete
Inline suggestions. Low stakes. The human is in the driver's seat for every keystroke. "Good" means useful suggestions that don't slow me down.
Wave 2, Chat
Conversational assistants in the IDE and the browser. The human asks, the model answers. "Good" means accurate, grounded, well-cited answers I can act on.
Wave 3, Agents
The model takes multi-step actions on your behalf, using tools, across files, sometimes across systems. The human reviews outcomes, not keystrokes. "Good" now means predictable plans, reviewable diffs, and reversibility.
Why this matters
The governance model that worked for Wave 1, basically, "let people use it", does not work for Wave 3. Agents change:
- The blast radius (multi-file, multi-system).
- The review surface (plans and diffs, not lines).
- The failure modes (silent drift, tool misuse, scope creep).
- The trust contract (you're now approving intent, not output).
If your team is in Wave 3 with Wave 1 governance, you will have a bad week. Build the governance for the wave you're actually on.