This was a weekend experiment to see what is possible with long horizon agents, which is the pattern of using role-based agents on top of some sort of task store.
As someone who has taken on-call rotation and maintenance work at many a startup, this is the pattern that I would have liked to have had.
Fun to build and would certainly do so again next time I'm doing a production-ready build. Would be exciting to add product analytics having the agent do minor feature work too, with it's own A/B testing of the results.
I love what Resend are doing and am a customer. I can also absolutely empathise as our lead engineer did exactly the same thing at a startup I was running a decade ago. It's a horrible situation.
But yeah, both the incident and the report are really tough to read. It would be great if they can do a follow-up with further actions they're taking.
There's a neo-bank called Revolut that allegedly at one point had just two teams: "go fast" and "don't screw it up". I feel like an infrastructure play needs some dedicated hires in camp 2.
This was a weekend experiment to see what is possible with long horizon agents, which is the pattern of using role-based agents on top of some sort of task store.
As someone who has taken on-call rotation and maintenance work at many a startup, this is the pattern that I would have liked to have had.
Fun to build and would certainly do so again next time I'm doing a production-ready build. Would be exciting to add product analytics having the agent do minor feature work too, with it's own A/B testing of the results.