AI coding agents made a huge leap forward since December, completing complex projects with minimal oversight, meaning “programming is becoming unrecognizable”
@karpathyAndrej Karpathy
Context & Ripple Effects
This marks a step beyond earlier code-generation milestones such as DeepMind’s AlphaCode effort, with the claimed change now centered on agents carrying complex projects through with little human intervention.
The arc also exposes the constraint on adoption: subsequent coverage describes companies struggling to review and secure rapidly expanding volumes of AI-generated code. The capability shift therefore matters as much for software assurance and workflow design as for code production.
First-order effects
Engineering teams can delegate larger, multi-step implementation tasks to coding agents, shifting developers’ immediate work toward specifying goals, reviewing outputs, and handling exceptions.
Organizations adopting these agents face a more urgent need to define where minimal oversight is acceptable and where human approval remains required.
Second-order effects
Code review, testing, and security functions become bottlenecks when agents increase the amount and scope of code produced; tooling that validates agent output gains importance.
AI coding vendors will be judged less on isolated code suggestions and more on whether their agents can operate reliably inside existing development workflows.
Third-order effects
If agent autonomy continues to improve, software teams may be organized around supervising automated implementation rather than manually producing every component, changing the economics of routine development work.
The durable differentiator may shift toward governance, integration, and verification systems that make autonomous coding usable in production—not simply model access.
The trend: This is one data point in the industrialization of software development, where embedded agents move from assisting individual programmers to executing bounded workflows.
I experienced a very similar transition in December. However, for higher-complexity tasks (ML-related), we are still not there yet. Two days ago I had GPT-5.2-PRO-ET and DeepThink argue for hours, converge, be happy, yet they missed a very obvious math issue. Still a huge unlock
This is counterintuitive for some, which is why there's a paradox named after it. But if you lower the cost of something that was previously supply constrained, demand for that thing goes up. Software engineering is just one of the easiest examples to contemplate. The process
4w ago I was a Claude Code skeptic. I'm not a coder. None of the use cases were relevant. I managed teams & projects, drowning in email & overdue reminders. So I tried creating tools that would help me and... holy crap. Now I'm sharing the tools I built: https://claudeblattman.co…
Very little about software engineering has changed over past last three months. A great deal has changed about coding, not unlike when we saw the rise of high order programming languages and compilers, the difference today being that the number of developers is far larger and
I'm seeing Chart 1 below being shared widely, suggesting the possibility that software programming jobs are booming as vibe coding gets cheaper. Hey, look, maybe. But zoom out a bit. Chart 2 is the 4 year history of software jobs. Huge boom, big bust, teeny bounce back. [image]
claude just cleaned my repos, rewrote all my scripts and ran them itself good thing is i didn't do anything wrong and all the results still hold, it just improved inference incredible, best 100 euros I've ever spent
I never tried to build anything with AI coding before this weekend and now I've bought three domains and am working on like 7 small projects that I've been dreaming about for years (!!) And they work! I'm speechless honestly And I'm a stupid peon with seriously limited
Proof below that anyone with ideas can be a builder now. @Replit has empowered me and millions like me to put our ideas to work. This is just the beginning. So much more in store 4 all of us builders soon! [image]
In my views ... To really benefit from AI generated code, you need to understand code better than the AI does. If you don't you might copy whatever it gives you but you won't grow as a developer. You'll end up producing messy work and not improving your skills. The good news
My carpenter told me he's now nailing a thousand nails an hour with his new nail gun. He still hasn't built the in-law unit but at least now he's using a lot of nails.
An experienced programmer told me he's now using AI to generate a thousand lines of code an hour. When I posted a similar stat 6 months ago, I got about a 50-50 mix of indignant disbelief and “Yeah, me too.” I'm curious if the split will be different this time.
Software engineering changed more in the last 3 months than the preceeding 30 years. Everything about running a software company needs to be rethought from first principles.
WAT “Anthropic is the fastest growing company of all time adding $4.5B run rate in 42 days after the $380B round.” “OpenRouter grew 2.5x in 1.5 months. On track to 1 quadrillion token annual run rate.”
It _is_ business as usual. You know why? We haven't been flooded with amazing new apps, tools or workspaces. Just a wordcel machine spamming code based off all the bad ideas most programmers had before, with all the ambition of a rock sitting at the bottom of a pond. The only
I promise you, if you rethink software eng from first principles, you will NOT arrive at the vibe coding slop generation we're in... unless your principles are really just capitalism and not engineering.
At this point, “agentic engineering” has allowed me to build the best AI harness I could possibly get my hands on. Yes, I vibe coded it. That's right. You don't need to wait around for the features you need for your AI agents. Please don't. You could just build them yourself.
“Hey Claude, can you market, find customers, run PoCs, convince stakeholders, negotiate budgets, go through procurement, and close this PO for me?” Yeah, that doesn't work and that's the vast majority of what it takes to run a company, not having 194 agents banging out thousands
“Programming is becoming unrecognizable. You're not typing computer code into an editor like the way things were since computers were invented, that era is over.” As Steve Jobs said, “Musicians play their instruments. I play the orchestra.” We don't write code anymore. We
Programming will never be the same... 🤯 Karpathy gave an agent one sentence. 30 min later... SSH set up, vLLM running, endpoints live, web UI built. He didn't touch anything. This is the new normal now.
Yup. I feel this in my bones. Classical Software Engineering is over, period. It's very weird, and it's honestly a hard thing to ‘tell people’ about. You can't just tell people, they gotta feel it. I tried to step back and actually write down some thoughts about this all. Maybe
in december I experienced a vivid realization of this and decided to immediately pivot my career because of it it was a distinct feeling, specifically after building with opus 4.6 + codex 5.2, that everything was going to be “different”
I love Karpathy's posts because they're so on point. He's not only a leading expert in his field, but he also manages to capture the zeitgeist with his statements. But this post is particularly impactful. Since December, (agentic) coding has undergone a significant