Eno Reyes

They want to just send off an agent and have it, they give it a loose description of the task and it comes back and it's done.

It works.

It passes every test.

A senior reviewer looks at it and goes, nice work.

That looks great, right?

There's these other sort of things that people want.

They want incident response to be automatic.

They want documentation to be generated on the fly.

And I think that what people sort of underestimate is that

In order to get there, the current quality level that you'll see is totally a function of your agent readiness.

There's some great research out of Stanford that actually looked at hundreds of global 2,000 businesses that are adopting AI.

And the question was basically, how can we predict whether or not this company was accelerating or decelerating because of AI, right?

Which it turns out, there are plenty of companies that are actually decelerating their productivity because of AI, which is a little counterintuitive.

It is.

So this is what's interesting is they looked at adoption.

what percentage of their users are actually using AI.

They looked at token usage.

What, like, are they using lots of tokens, right?

Everyone can adopt AI, but not use it.

And maybe that's what causes it, right?