My take: you most likely are not thinking large enough with coding agents. Think much bigger.
Next time you are using a coding agent ask yourself “what is a version of this task that would take years for a team of humans to do?”
For example - instead of saying “Make this kernel faster” Try “Optimize my entire system and come back to me in 7 days”
With recent coding models I have accomplished 50 SWE-year projects in single digit weeks.
Straight talk: If you're a senior+ engineer in the year of our machine lords 2026, and you refuse to use AI at a company that pays for your AI use, I have questions.
It's 3 years the world is talking about this. Learning is part of your job. Why are you still not learning?
@valigo Yes and no. AI can also help process a ton of profiling data, revealing patterns you didnt know existed. It can help write tools for compression. Etc etc. So you cant just prompt "optimize my game" but it can still play a huge role in getting you to your perf goals.
@EpicVogel@Lionheart@TimSweeneyEpic No agents.md at all? What about other guidance docs for coding standards etc? Trying to figure this all out myself and its hard since this is all uncharted territory.
@SebAaltonen@Grummz Is this the modern equivalent to googling something and finding a Stack Overflow post from yourself 10 years ago giving someone else the answer?
@SebAaltonen How long do you think it would have taken you to recreate the clay book tech? 10k seems pricey for code you wrote yourself and could likely rewrite.
@sebify For them I imagine it is the right decision, at least for now. There is probably a deluge of PRs that they have no way to meaningfully review. Easier just to ban them. The software industry will find a way to manage it in time. We have the same struggle on a smaller scale
@andreintg For one off tools I might not even look at the code, and don't bother to commit it. For game code though you have to do a whole review and iteration before you even get to your regular human review processes.