AI made your engineers 10x faster. So why is your team shipping the same amount?
Here's the uncomfortable answer: you sped up the one loop that was never the problem.
Three Loops, Three Speeds
Andrew Ng shared a diagram last week that explains it perfectly. Building software with AI runs on three loops, and they move at completely different speeds.
The coding loop takes minutes. The AI writes it, tests it, fixes its own bugs, done.
The developer loop takes hours. A human looks at what got built and asks the questions AI cannot answer. Is this the right thing? Does this flow make sense? Should this even exist?
The external loop takes days. Real users touch it and tell you the truth.
The Loop Everyone Sped Up
Everyone poured their energy into the first loop because that's the one that got fast. New tools, faster agents, better prompts. And it worked — the code flies out now.
But here's what nobody wants to admit: the coding was never the bottleneck. The thinking was.
Your team can now generate features in minutes and still spend three weeks building the wrong ones. Speed on the first loop just means you arrive at your bad decisions faster.
Where the Real Edge Is
The teams pulling ahead right now are not the ones with the fastest coding loop. They're the ones who took all that saved time and poured it into the two human loops — deciding what to build, talking to users, thinking harder about what actually deserves to exist.
Faster code is not your edge anymore. Everyone has that.
What you do with the time it frees up is the only thing left that matters.
Framework by Andrew Ng, The Batch.
Be honest — is your team using AI to make better decisions, or just to make the wrong ones faster?