Andrew Ng wrote a letter last week on what he calls "loop engineering." It named something I had been feeling but had not put into words.
He describes three loops for building 0-to-1 products. Each runs at a different speed.
The agentic coding loop runs in minutes. The agent writes code, tests it, and iterates until it meets the spec. I no longer sit in the middle of this loop. It runs on its own.
The developer feedback loop runs in hours. This is where I look at what the agent built and steer it. What features matter. Where the UI is off. What the flow should feel like.
The external feedback loop runs in days. Friends, alpha testers, production data. This is the slowest signal, and it reshapes the vision that drives everything upstream.
What struck me is the cadence itself. Minutes, hours, days. Three loops nested inside each other, each feeding the next.
For most of my career the loops were not separated this cleanly. Writing, testing, reviewing, and shipping all blurred together at roughly human speed. The agent changed that. It collapsed the innermost loop to minutes. And once the fast loop got that fast, the slower loops became visible as distinct things.
That is the part worth sitting with. The value did not just move up a level. The whole structure came into focus.
Read Andrew's full letter here: The Batch, Issue 359