Why Karpathy and Yegge Think Developers Are About to Get Way More Valuable, Not Replaced
You’re scrolling through your tech feed, and there it is — another bold claim: “AI will replace all developers by 2026.” But then Andrej Karpathy (OpenAI co-founder) and Steve Yegge (the Amazon and Google veteran) share their real predictions. And what they say? It turns that idea completely upside down.
Their take? Everyone’s got it backwards.
I’ve been working closely with AI coding tools for months now, and when these two independently came to the same conclusion, I knew I had to look into it further. What they’re saying isn’t just different from all the doom predictions — it’s actually the complete opposite.
Who Are These Guys (And Why You Should Care)
Karpathy isn’t some AI hype man. He’s a founding member of OpenAI, former director of AI at Tesla, and the guy who literally helped build the systems everyone’s arguing about. When he talks about AI’s limitations, he knows them intimately.
Yegge built critical infrastructure at Amazon and Google back when “scale” meant something different. Now he’s at Sourcegraph, working directly with enterprise teams trying to actually implement this stuff in production. Not demos. Real code that has to work.
When both of these guys look at the current AI coding explosion and say “developers aren’t going anywhere,” I listen.
The Core Insight That Changes Everything
Here’s what clicked for me: This isn’t about replacement. It’s about the next abstraction layer.
Think about it. How many developers today write assembly? Almost none. Did high-level languages kill programming jobs? The field exploded instead — because suddenly you could build way more complex stuff way faster.
Both Karpathy and Yegge see this exact pattern happening again. As Yegge puts it: “building enterprise software will always be monumentally difficult, so engineers and AIs will team up to build it together.”
The keyword there is “team up.” Not “AI takes over.” Team up.
What This Actually Looks Like Right Now
Let me show you something concrete. Karpathy coined this term “vibe coding” that perfectly captures what’s happening:
“There’s a new kind of coding I call ‘vibe coding’, where you fully give in to the vibes, embrace exponentials, and forget that the code even exists.”
Sounds terrifying, right? But here’s how it actually works.
For his weekend projects, Karpathy doesn’t touch the keyboard much anymore. He just talks to Cursor Composer: “decrease the padding on the sidebar by half.” The AI makes the change. If it looks right, he moves on. No hunting through CSS files, no debugging margins.
When he hits errors, he literally just copy-pastes them back to the AI with zero comment. “Usually that fixes it.”
Now here’s the key part everyone misses: This approach scales with the stakes.
Throwaway weekend project? Full vibe mode. Let the AI handle everything.
Production enterprise system? You need what Yegge calls “supervised AI” — the AI does heavy lifting, you guide and verify everything.
Same tool. Different levels of human oversight depending on what you’re building.
Yegge’s Technical Progression (This Is Happening Fast)
Yegge’s been tracking this evolution closely. He coined “chat-oriented programming” about a year ago — basically having conversations with AI to write code instead of autocomplete.
But now? “Chat programming is current, but agent programming is already rocketing past that approach with exponentially better results.”
Agent programming means AI that can handle whole workflows by itself while you watch. Think less “help me write this function” and more “build user login for this app, including password reset.”
The speed of this progression is wild. We went from autocomplete to chat to autonomous agents in like 18 months.
The Three Programming Eras (And We’re Entering #3)
Karpathy breaks this down into three clear stages:
Era 1: You write explicit instructions. Want to sort data? Write a sorting algorithm.
Era 2: You show examples, the computer learns patterns. Want to classify images? Train a neural network on thousands of labeled photos.
Era 3: You describe what you want in English. Want user authentication? Say “create a secure login system with password reset functionality.”
Here’s his key insight: “LLMs are a new kind of computer, and you program them in English.”
This isn’t just better tooling. This is programming becoming accessible to anyone who can communicate clearly. Your PM could prototype their ideas. Your designer could build interactive demos without waiting for engineering.
That doesn’t replace developers — it multiplies what everyone can accomplish.
The Problem Nobody Talks About: Jagged Intelligence
But here’s where I get skeptical, and where Karpathy’s honesty is refreshing.
AI has what he calls “jagged intelligence” — it can solve incredibly complex problems while failing spectacularly at simple ones. An AI might nail a complex algorithm, then confidently tell you that 9.11 is bigger than 9.9.
“For now, this is something to be aware of, especially in production settings. Use LLMs for the tasks they are good at but be on a lookout for jagged edges, and keep a human in the loop.”
This is why the “AI will replace all developers” takes are nonsense. The AI is simultaneously brilliant and incredibly stupid in unpredictable ways. Production systems can’t handle that kind of inconsistency without human supervision.
The Economic Reality (This Is Already Happening)
Here’s the part that should get your attention: “Some companies have already laid off 30% of their engineers who wouldn’t adopt AI.”
Not future tense. Already happened.
“Companies with deep pockets can simply pay to play, but others will face difficult decisions — either absorb the costs, fall behind competitors, or reduce headcount to cover the new expenses.”
Translation: If one developer using AI tools can do the work of three developers without them, guess which two are getting cut?
This isn’t theoretical. I’m seeing this in real companies right now. The developers who are using these tools are becoming incredibly valuable. The ones ignoring them are getting left behind
What You Actually Need to Learn (It’s Not What You Think)
The skills that matter aren’t about memorizing new APIs or learning prompt engineering tricks. They’re higher-level:
AI Supervision: Getting good at spotting when AI output is solid vs. garbage. This is surprisingly learnable — you start recognizing patterns in AI mistakes.
Problem Architecture: Breaking complex requirements into pieces AI can handle reliably. This is actually classic engineering skill applied to a new tool.
Quality Verification: Getting fast at spotting the subtle bugs AI introduces. Different from normal debugging because AI mistakes follow different patterns.
Natural Language Precision: Getting better at expressing requirements clearly. If programming becomes more conversational, communication skills become technical skills.
The weird thing is, these are all fundamentally human skills that get amplified by working with AI, not replaced by it.
The Timeline (Ignore the Hype, Watch the Reality)
Karpathy has a reality check for all the “AGI by 2025” breathlessness: “when I see things like, ‘oh, 2025 is the year of agents,’ I get very concerned, and I kind of feel like this is the decade of agents.”
A decade. Not next year.
“Patience, unfortunately, is not a virtue embraced by Wall Street, so expect the AI hype train to continue spewing nonsense as the real practitioners figure out how to shape a new era of computing.”
But here’s what’s happening right now that you can use: Tools like GitHub Copilot and Cursor are already making developers 30–50% faster on routine tasks. Not theoretical productivity gains — measurable improvements you can see today.
The transformation is gradual enough that you have time to adapt, but fast enough that you need to start now.
Why I’m Actually Optimistic About This
Look, I was skeptical. The whole “AI will democratize programming” narrative sounded like typical Silicon Valley BS. But after using these tools for real projects, something clicked.
Every major programming shift has followed the same pattern. Assembly to C. C to Python. Command line to GUI. Each time, people worried about dumbing down the field. Each time, the field got bigger and more creative instead.
The developers who thrived weren’t those who fought the new tools. They were the ones who learned to use them to build things that weren’t possible before.
This feels like the same thing, but bigger. We’re not just getting new syntax — we’re getting a new way to think about problem-solving that’s more about clear communication and less about remembering API details.
As Yegge puts it: “Computer science education will need to evolve, but the fundamentals remain valuable. When assembly language was replaced by higher-level languages, people worried programming skills would deteriorate, but instead, the field expanded and jobs increased.”
The developers winning right now aren’t the ones with the deepest knowledge of React hooks or Kubernetes configs. They’re the ones who can clearly articulate what they want to build and guide AI to build it correctly.
What to Do This Week
Stop reading about this and start using it. Pick one AI coding tool — GitHub Copilot if you want something stable, Cursor if you want the newest stuff.
Start with a throwaway project. Build something stupid and fun. A random quote generator. A simple todo app. Something where it doesn’t matter if the code is perfect.
Don’t try to revolutionize your entire workflow immediately. Just get comfortable with the rhythm of human-AI collaboration on something low-stakes.
The developers already thriving in this new world didn’t wait for the perfect tool or comprehensive tutorials. They started experimenting, made mistakes, and learned by doing.
The Real Future
Here’s what both Karpathy and Yegge understand that the replacement narratives miss: This technology amplifies human intelligence, it doesn’t replace it.
We’re not becoming obsolete. We’re becoming orchestrators. We’re not losing our jobs to AI. We’re learning to dance with AI to solve bigger problems than we could tackle alone.
The future belongs to developers who can think at a higher level, communicate clearly, and know how to direct AI to solve complex problems. If you’ve made it this far as a developer, you already have most of what you need.
You just need to start learning the dance. And honestly? It’s a pretty fun dance once you get the hang of it.
