Anthropic’s Boris Cherny argues that for certain modern, model-friendly codebases, especially TypeScript/React projects, coding is already effectively "solved". AI agents can write nearly 100% of the code, while humans shift from hand-coding to directing, reviewing, integrating, and scaling many small PRs at once.

Solved for greenfield TypeScript, sure. Try pointing an AI agent at a 15-year-old enterprise Java codebase with zero tests and tribal knowledge in Confluence pages nobody reads. The gap between demo-ready and production-ready is where most AI coding claims quietly die.

Just another grinning corporate shill with a lot of stock options

@kimmonismus honestly, feels like we should’ve seen this coming with how advanced these tools got. kinda wild to think coding might shift to more of a management role soon.

I went in circle for hours with GPT 5.5 xhigh for a simple visual tweak.

another proof that's AI and AI agents building our future

Not sure I exactly buy that. When someone says 'code' I include all the security and testing and while 5.5 is probably the best example yet - I think there are still good chances of getting things wrong that human engineering still needs to review. Early days however but that's my take.

can write bs spaghetti nonsense code. it's working but it's just a mess

this tracks for greenfield TypeScript but falls apart the second you hit legacy. we run agents on a 12yr Go monolith — they still can't reason about implicit state machines spread across packages. "solved" is doing heavy lifting here.

agents can ship a lot but someone still has to keep the product coherent

150 PR's a DAY , my guy didnt even read 10 of those PR's BET !

What a load of horseshit🤣

He's not the only one doing this:

the building is solved, yes partially but knowing what to build is 10-15x more important

The bottleneck shifted from writing to knowing what to build and why. Generating code is table stakes now

If you never look at the code yeah sure, its solved

It cannot pickup new languages it hasn’t seen, it’s just good at what it’s seen during training. That’s a plain good lie right there.

"Solved" is doing a lot of work here. Generating code and deciding what to build are different problems. The second one doesn't get easier when the first one disappears.

sure they can write it, doesnt mean that you shouldnt change it if you actually look at the code and dont just vibe code you will realize the magnitude of slop that these things produce

Coding solved just means the job is now reviewing ai pull requests until your eyes glaze over

this slop loser, I seriously hate his stupid gleaming face.

The real challenge now moves from syntax to integration and ensuring that many small changes don't create technical debt. Human oversight is still the critical filter for maintaining a coherent long-term vision.

But can the do it safely? ⬇️⬇️⬇️

Vibe coding is my life now. Longer hours. More output. More tightly packed strategic decisions. And every wrong turn has a measurable token cost.

AI write 100% of the code, but it still needs a dev to babysit it

Strongly typed languages will win

Then why CC has 5k issues and a lot of them are flagged bugs

the TypeScript/React specificity matters. it's not coding broadly that's solved, it's the subset where the patterns are well-established and the training data is dense. the messier the domain, the more human judgment still does the heavy lifting.

use an opinionate framework, i use ruby on rails because there are very few 'correct' way to do things. usually just one. the models almost never make mistakes and when they do those are honest differences of opinion not massive everyone's fired typed mistakes.

🫪🫪🫪

Coding being solved for TypeScript/React is a bold claim but honestly not that far from what people are already experiencing day to day

So humans are just code directors now?

Ah great, so given Claude Code was built with TypeScript and React, why is it buggy AF?

Provocative! AI writing 100% code while humans direct - seismic shift in dev. But key questions: How maintain quality? Debug complex systems? Real challenge isn't coding, but scaling AI-human collaboration. Quality over quantity

review taste is where the work moved