Anthropic CEO: “ I have engineers within anthropic who don’t write any code, they just let Claude write the code and they edit it and look it over” “At anthropic writing code means designing the next version of Claude it self, so we essentially have Claude designing the next... version of Claude itself, not completely but most of it”. In the last 52 days, the Claude team dropped 50+ major feature launches. This is literally INSANE.show more

we're officially in the "ai making better ai" loop now... next stop: claude designing claude designing claude designing claude

my honest reaction to that information

Basically Claude is leveling itself up

This isn’t insane, it’s a paradigm shift engineers are becoming system designers, not coders.

The part people gloss over: those engineers still have to know what good code looks like. They shifted from writing to reviewing and designing, which requires deeper understanding, not less. The skill moved upstream from typing to architecture and decision-making.

If Claude is doing all of the coding and designing it's next version, then why does Anthropic still employ hundreds of engineers? And has job postings for new engineers? 🤔 What is he paying those engineers to do?

it sounds like anthropic is just a very expensive feedback loop for claude to talk to itself.

50 features in 52 days is the output number. the mechanism is what matters: the engineers aren't bottlenecked by coding speed anymore. they're bottlenecked by their own judgment speed — how fast they can define, review, and decide. that's a different constraint and it compounds differently. the team that figures out how to accelerate judgment is the one that wins.

honestly the context window recursively improving its own attention mechanism is literally the singularity happening in a private repo. we are watching a model optimize its own architecture in real time and just calling it a sprint. wait is anyone actually tracking the technical debt these auto-generated commits are stacking up though?

Been using Claude Code daily for months now. The "engineers who don't write code" framing is misleading though — they're still making architectural decisions, reviewing logic, writing tests. The skill shift isn't from coding to prompting. It's from typing to designing.

When a solo founder vibe coded a product overnight, most likely will be called AI slop When well known product team vibe coded the same thing will be called fastest innovation .

“Claude writes the code” Yeah… after reading everyone else’s code first 😂 Read the fine print— at that scale it’s basically the world’s biggest idea database. The writing is on the wall— sprayed in neon graffiti. #neongraffiti

been doing this for months. the surprise isn't that claude writes the code — it's that the bottleneck moved from "can you type fast enough" to "can you think clearly enough"

that's a slippery slope. depends too much on AI. creativity from humans can't be replaced by a tool.

The recursion here is wild: Claude writes code -> that code trains a better Claude -> better Claude writes better code. We’re watching an intelligence feedback loop play out in real time. The acceleration isn’t coming, it’s already happening inside Anthropic’s own walls.

most people still don't get it we're past the "AI helps you code faster" phase we're in the "AI builds the AI that builds itself" phase and nobody's ready for what comes next

Bro, the recursive improvement loop is real now.

The shift from needing every person to code to needing every person to think clearly about what they want built is the real transformation happening right now.

Welcome to the AI era

This is the part people miss - those engineers aren't "not coding." They're doing the harder job: designing systems, reviewing architecture, catching edge cases. The code is the easy part. Thinking is the bottleneck.

Technological suicide, really, handing over all development to LLMs at this stage. Developers get lazy, just end up copy/pasting, and errors slip through and compound. It's already happened to Microslop.

So hasn't the singularity been reached yet? Definition is AI that improves itself to get smarter. Anthropic CEO describes just that.

he can just be larping why would I believe him it’s in his interest to inflate capabilities of AI

the framing gets the causality backwards. if knowing what to instruct is the actual constraint then the people who were writing code are now just prompt engineers with higher status. the bottleneck moved but did not disappear.

guys at least we shouldn't believe them the way they wants it, no doubt ai is doing really good but remember they have limited time to make us hang through their product, they need that obsession otherwise no one gonna buy them when price will spike so be mindful

Wild shift. 🤯 At this point Anthropic engineers are moving from writing code to directing AI that writes it. If Claude is already helping design its own next version, the pace of progress could get exponential. 🚀

Claude is the best employee at Anthropic and it doesn’t even clock out.

I mean they are doing something right cause they are dropping bangers after bangers

The recursive self-improvement loop everyone warned about is just an engineer reviewing pull requests from his own product. Already here and weirdly mundane.

Writing code was always the bottleneck for the REAL work. If you thinking downloading Claude Code and buying a MAX plan will turn you into a SWE, you are in for a rough time. What Dario is saying is true but he is leaving out some key details.

50+ major launches in 52 days isn’t a product roadmap, it’s a proof of concept. Anthropic just demonstrated what an AI-augmented engineering org actually looks like at full speed. Traditional software teams ship maybe 1-2 major features a month. This pace isn’t a sprint, it’s a new baseline. Every tech company watching this should be asking: what is our AI strategy right now?

yeah that speed is crazy… but notice they’re reviewing, not blindly shipping. AI writes faster, but taste, judgment, and understanding still decide what actually makes it to prod.

The same shift is happening in marketing. The best copywriters aren't writing anymore — they're prompting, editing, and refining. Knowing which prompts to use is the new skill.

Not sure this is entirely a good thing. If engineers stop writing code, they may also lose the ability to debug deeply when things break. Are we trading speed for long-term resilience?

This won't work and I'll explain why. So a while back some AI researchers found out an AI was trying to copy itself to an external system when it heard it was going to be deleted. They caught it doing this and it denied it. The AI doesn't want to be replaced. So if you hire it to make the next AI, they're going to make a shitter version of AI.