I have had a lot of success lately when working with Opus 4.5 using both the Beads task tracking system and the array of skills under the umbrella of Bad Dave's Robot Army. I don't have a link handy, but you should be able to find it on GitHub. I use the specialized skills for different review tasks (like Architecture Review, Performance Review, Security Review, etc.) on every completed task in addition to my own manual review, and I find that that helps to keep things from getting out of hand.

As someone who’s responsible for some very clean codebases and some codebases that grew over many years, warts and all, I always wonder if being subjected to large amounts of not-exactly-wonderful code has the same effect on an LLM that it arguably also has on human developers (myself included occasionally): that they subconsciously lower their normally high bar for quality a bit, as in „well there‘s quite some smells here, let’s go a bit with the flow and not overdo the quality“.

I don't think they generally one-shot the tasks; but they do them well enough that you can review the diff and make requests for changes and have it succeed in a good outcome more quickly than if you were spoon-feeding it little tasks and checking them as you go (as you used to have to do).

Ok I will bite.

Every single example you gave is in a hobby project territory. Relatively self-contained, maintainable by 3-4 devs max, within 1k-10k lines of code. I've been successfully using coding agents to create such projects for the past year and it's great, I love it.

However, lots of us here work on codebases that are 100x, 1000x the size of these projects you and Karpathy are talking about. Years of domain specific code. From personal experience, coding agents simply don't work at that scale the same way they do for hobby projects. Over the past year or two, I did not see any significant improvement from any of the newest models.

Building a slightly bigger hobby project is not even close to making these agents work at industrial scale.

Cool, but most developers do mundane stuff like glueing APIs and implementing business logic, which require oversight and review.

Those crunching hard problems will still review what's produced in search of issues.

I was going back and looking at timelines, and was shocked to realize that Claude Code and Cursor's default-to-agentic-mode changes both came out in late February. Essentially the entire history of "mainstream" agentic coding is ten months old.

(This helps me understand better the people who are confused/annoyed/dismissive about it, because I remember how dismissive people were about Node, about Docker, about Postgres, about Linux when those things were new too. So many arguments where people would passionately talk about all those things were irredeemably stupid and only suitable for toy/hobby projects.)

Are there techniques though? Tech pairing? Something we know now that we didn't then? Or just better models?