Emergent design: embracing iteration in building software
Coding agents are useful for more than writing code fast. Share your intent and they surface corner cases, some you thought of and didn’t say, many you missed, before any of it becomes a bug. Tests then reveal the rest, and some of those findings change the intent itself. With agents the scale of building has exploded, but the iterative flavor has not changed. The best name I’ve found for this is emergent design.
Takeaways
- Autocomplete looks ahead a few tokens. An agent looks ahead across requirements, failure modes, interfaces and future extensions.
- Human judgment gives the direction; the agent brings breadth. Then the baton goes back to the human to update the intent.
- The open question: can agents look two or three iterations ahead, so you skip some implement-test-update cycles? If not, what is the blocker?
Have a system that needs a second opinion?