Doneas puts you in a small group of founders on a fifteen-minute call at the same time every day, where you say what you did yesterday and what you are doing today, and if you do not turn up somebody goes and checks on you. Bengeekly tried everything else first, including a standing bet with himself of one KitKat for every day he skipped the coworking space, and hiring an assistant purely so somebody would be on an early call with him. All of it held until he travelled, and then every habit went at once. What finally stuck was other builders on the same call, so he got the best founders he knew into one and it grew from there.

heal.dev goes looking for bugs by running your app, not by reading it. It stands up a production-like sandbox with your external dependencies mocked, then goes exploring for edge cases, and it will not report anything it has not first reproduced in there. What you get back is the failing case with a script that proves it, which you can hand straight to a coding agent and then rerun to check the fix took. Malo's team benchmarked it against the AI code reviewers and found 70% of what Heal caught, no reviewer did, and that stacking two or three reviewers barely helps because they all miss the same class of thing.

Claude Duet is an agent you hand a job to in a chat and then leave alone. It works on its own computer in the cloud using whatever apps you have connected to it, gets on with the thing, and comes back to you when it has finished or when it reaches something only you can decide. Every lab has put one of these out in the last few weeks, Meta's Muse, OpenAI's Dots, Grok Bot, and all of them are competing on how much they will take off your hands. The line Anthropic is selling here is the opposite one, that it is always on and still knows what can wait.