How Are the Latest AI Models Stressing 37signals?

David Heinemeier Hansson says nothing has stressed the technical team at 37signals like the latest batches of AI models, which are exceptionally good at finding real security issues a clever hacker could exploit.

Used in defensive ways, the models surface a seemingly endless parade of vulnerabilities that still has to be patched, fixed, and sorted. The road is rocky, and teams still have to steer the agents and verify and vet the work. The same skill that finds an exploit also finds the defense, so the aggressive uses of the models are also the defensive ones: spy versus spy, with the same capabilities and the same actions. The end result is vastly more secure systems.

The more concerning part is all the teams that are not stressed. A team right now that is not dealing with a bunch of patches is blind, and adversaries quite probably have access to techniques that can get into the systems.