Ok, to start. I like the new update. Even if the update seemed to cause my weekly quota reset to extend by 23h for some reason.
The /teamwork collaboration looks useful and I am happy to see it being thorough in testing.. but last week I would say I got 4 times as much done with the old IDE and 3.8 model as I did this week with the new Gemini 4 model and the new IDE.
The new IDE and model seemed to go chasing reasoning and testing dow which made no sense to me as the vast majority of testing is clear and black box deterministic testing. It shouldn’t need an agent to reason it every time. There will be times where an agent needs to decide something, of course, but this week I burned through a weeks quota in two and a half 5h windows developing a nearly 100% deterministic, standards driven Automatic Dependent Surveillance Broadcast library. This library has 4x less code in it, compared to what I worked on last week, is entirely driven by maths contained in a RTCA standard, and in spite of the agents running over 3000 tests in total, it still crashed the frontend in the first 60 seconds. Thankfully I still remember how to code so I fixed it myself, but that’s 4x less code than last week’s project ( which required a lot more reasoning) and I burned through the quota 4x as fast.
last weeks was 3.8 with medium reasoning, this week 4 with medium reasoning. I think/teamwork was the real quota killer though. It was probably my mistake to use teamwork for a from scratch project. It is probably better to hold off on it for refactoring and testing for production ready projects.
Not necessarily a bug.. but people should beware.. /teamwork is quite good, but you won’t get much work done with it on a pro account. Of course, if google want to up the quota limits, it’s a feature I would be very happy to use going forward, so thank you to the developers who created it.