The AI Workforce is Here: The Reality of Autonomous Code If you haven't seen the recent update from Uber's CTO, Praveen Neppalli Na...
The AI Workforce is Here: The Reality of Autonomous Code
If you haven't seen the recent update from Uber's CTO, Praveen Neppalli Naga, it's worth taking a look. He shared some numbers about their internal AI agent that are honestly pretty wild to think about. Right now, this internal tool is writing about 1,800 code changes a week completely on its own. There is absolutely no human authoring involved in those specific changes. That comes out to roughly 8% of their entire codebase being generated automatically, up from basically zero just a short time ago.
The Shift from Creation to Auditing
The thing is, this completely changes what an engineer actually does all day. When the AI is the one spitting out the syntax, the human's job naturally shifts away from active creation.
Just pure auditing.
Look at it this way, reading someone else's code is usually a lot harder than writing your own from scratch. When you write it, you know exactly what the logic is supposed to do because you just built it in your head. When an AI drops a massive, dense pull request into your lap, you have to read every single line to make sure it doesn't break something else in the background. You are basically transitioning from being a traditional programmer to being a full-time editor.
It is exhausting.
You have to maintain an incredibly high level of focus just to verify that the machine actually understood the assignment. The AI can generate a thousand lines of code in a few seconds, but a human still has to sit there and trace every variable, database call, and dependency. If you approve a bad pull request because you were skimming, that bug goes straight to production. So, the bottleneck hasn't disappeared at all. It just moved further down the pipeline to the code review stage.
The New Economics of Development
At ATX Soft, we're already seeing this shift. Our devs are spending way more time reviewing AI-assisted pull requests than they used to. It used to be that you'd spend most of your afternoon writing a new feature, and maybe a few minutes reviewing a teammate's work before logging off. Now, you might spend hours reviewing complex logic that nobody on your team actually typed out on a keyboard. We still have to build the core architecture and figure out the actual business logic, but the raw typing part is shrinking fast.
Honestly, coding isn't dead. It is just changing into something else entirely. You don't pay people to type boilerplate anymore, you pay them to make sure the automated boilerplate actually works in the real world. The economics are different now.
Reference
- Uber (1,800 weekly AI code changes): Uber Engineering Blog
- Microsoft (20–30% AI-generated code): Microsoft Investor Relations
- Amazon (90-day AI safety reset): AWS Architecture & Engineering Blog
