i still don't get why i would need an agent sandbox and when i would need one. It works if you have unlimited tokens to spend, and you can trust an agent completely that it did what it said (even fable doesn't do this perfectly). Demo videos don't work for backend code, you cant trust an agent to write accurate tests. I just don't get it, maybe one day i';ll come around to it
Hard disagree, an engineering manager is not shipping products, the people they are managing are doing the work and its okay for an engineering manager to not read every line of code, not for people doing actual work.
If you're struggling with the idea of not reviewing every line of code and feeling like you can still deliver a good product, talk to an engineering manager that used to be an individual contributor. They've already been through what you're experiencing.
been working with a raspberry pi pico for a fun gift, and its crazy how many failure modes you need to take into account becuase once it's shipped you would need it back to fix it. We depend so much on OTA updates to our products that we forget sometimes that hardware is tough!
having kimi k3 not being slow is really exciting ngl. i'm having it find and exploit vulnerabilities in my project and its nice that i don't have to deal with constant rejections.
How can you possibly write perfect constraints? Arent you doing double the work at that point? Because a perfect constraint is just the code itself. And if you use ai to generate your constraints, then youve completely lost the plot. Im not anti ai but the fud on this app is unbearable at times. Sorry for a rant.
The whole stop reading the code, stop reviewing the code because of very tight constraints is pure bs. I use ai heavily and i still review the code. It isnโt possible to not review code in a complex codebase. Add cross service boundaries and cross service effects to that, how can u ship with confidence? The goal isnt to just make it work, to goal is to make it work correctly, sustainably. Not to mention you lose the mental model of the thing you are working on, its a slippery slope, that leads straight to slop. Even fable makes mistakes.
Kinda disagree, the math problems themselves donโt matter, its the tools and the ingenuity that we build along the way, the discovery of new novel ways to do things that matter. And these take time to become useful once more people discover and use these tools.