Anthropic accounts diverge on loop adoption rate

An Anthropic engineer says 80% of the company's engineers now run self-improving loops. Graphs are becoming the default build pattern within four to six months, per the same thread on X.

That's the kind of number that matters if you're wiring feedback loops into your own workflow instead of treating AI as a one-off assistant. Eighty percent inside the company that makes the models would be a strong signal, if the number holds up.

It might not. A separate Anthropic session put loop adoption at 90%, with more than 30% of code now written by loops instead of by hand. Ninety doesn't match eighty, and 'engineers who use loops' isn't the same claim as '30% of code comes from loops.' Those are two different measurements, not one number said twice.

Neither account says which definition it's using, or whether it's counting the same population of engineers. That's the actual story: the company running the experiment has two unreconciled numbers for a stat its own people keep repeating in public.

Treat both figures as directional, not exact. If Anthropic wants the industry to take loop adoption seriously as a metric, its people need to agree on one number before they post it.