Human code review doesn’t scale once agents write most of the code. No one can read that much code. So the review has to be automated.
But we’ve already seen agents break the rules they were given. Theoretically agents are capable of changing the tests to make them pass, instead
Claude writes better code when it can verify its work. Checks that used to be expensive, like tests, benchmarks, and end-to-end runs, are now cheap.
Here’s how the Claude Code team thinks about verification:
claude.dev/blog/how-the-c…
Human brains haven’t gotten much bigger in the last few thousand years. But we’ve made huge progress as a civilization in that time.
What changed is the scale at which humans coordinate. Every jump in that scale brought a jump in progress.
I think agents will go the same way.
Google DeepMind's Alex Imas says AI agents still struggle to coordinate, and that's a key bottleneck to economic transformation:
"What you actually need for real economic transformation is not just competence... It's the ability of these agents to coordinate with one another."
I like this. Running AI models locally is now a real option for anyone.
Whether you’re a company or an individual, you can use AI on your own data on hardware you control. It stays yours and you don’t have to hand it over to anyone
Also, the benefits go well beyond that.
We’re not going to win by lobbying, plotting, or manipulating. We’re going to win by building, in the open, with the whole community!
Today, with @ggerganov we’re super excited to announce the release of llama.cpp v1, the first major fully stable version of the most used library
This is one of the most open problems in AI right now.
Past a certain size, a team is limited by how well work moves between people. How good each person is matters less. Anyone who’s built products in a large organisation knows this.
Agents are at the same point.
Each agent