Is AI Replacing Developers? (The Complicated Truth)
Junior hiring is down. Tailwind cut 75% of engineering. But software demand is exploding. Toil is shrinking, judgment is rising.
> Key Takeaways
- > Layoffs are three things mixed together: real displacement, AI as a cover story for cost cutting, and premature bets.
- > Software demand is exploding faster than AI can compress labor. Token spend is up 5x year over year.
- > Toil is shrinking across every role. Judgment is the bottleneck, and that's where the work is moving.
- > The Prussian schoolhouse model of memorize-then-execute was already cracking. AI just made the cracks impossible to ignore.
- > Stop trying to know everything. Build judgment by shipping real work and watching what breaks.
> Linked Resources
> Transcript
▶ Show full transcript
Something feels different right now for developers. The job market's much tighter than it has ever been. Layoffs keep coming and, you know, the headlines are saying AI is coming for whatever jobs are left. Some of this is kind of real. Uh, junior developer hiring is declining. Harvard and LinkedIn studies confirm that. Tailwind Labs cutting 75% of their engineering staff.
Front-end and mobile roles are shrinking. Developers are kind of surprised that these AI tools can reproduce what took them decades to learn. And it's an honest reaction. A lot of developers are feeling that. Your job isn't disappearing. It's changing. And most people are preparing for some strange version of what's coming. We're going to look at why the layoff narrative is much messier than it looks and why cracks are showing in companies that cut way too deep.
why demand for software is actually exploding and the real shift and what that means for your career. Some of this displacement is real. When AI chatbots replace documentation visits, the people writing those docs face real pressure. When code generation handles simple CRUD work, the roles that were simple CRUD developers get compressed. And that's happening.
But a lot of what you're seeing in the headlines is not actually AI replacement. It's three things mixed together. First, real displacement. Yes, some tasks actually are getting automated. This is the honest part. Second is AI as a cover story. Companies are using AI transformation to justify layoffs that are really about cost cutting, bad quarters, or restructuring.
It's a convenient narrative for the board deck. Your CEO gets to sound visionary while cutting headcount for the same reasons CEOs have always cut headcount. Third is premature bets. Companies are cutting staff in anticipation of AI capabilities that don't exist yet. They read the press releases, they watch the demos, and they bet the org chart on the future that hasn't arrived yet is something they can have.
And some of these bets are going to look very expensive in 18 months. So when you see headlines that say company X replaces developers with AI, the real question is which of those three stories is it? Because the answer changes everything in how you understand it. And the cracks are already showing. GitHub has essentially zero nines of reliability.
Anthropic has had significant outages. Cloudflare AWS has a major incident. These are the foundations of modern software infrastructure. Now, I'm not saying AI directly caused every one of those, but this is what happens when cybernetic loops stop converging. When you move fast, cut humans from the loop, skip verification, the system can't selfcorrect anymore.
The feedback loops that keep things stable start breaking. If you've watched previous videos in this series, you'll know we talk a lot about loops. The same thing that makes agentic coding work, feedback loops with human judgment. That's what keeps infrastructure reliable, too. When you remove the judgment from the loop, the loop breaks. We're watching that happen in real time.
So, that's the scary part. Here's the other side. There's a concept from economics called Jevans paradox where steam engines got more efficient. People expected coal consumption to drop. Instead, it exploded because the efficiency made new applications economical. More trains, more factories, more ships. Demand expanded faster than the efficiency could compress it.
The same thing is happening with code right now. The cost per token has collapsed. But look at the total numbers. Microsoft processed over 100 trillion tokens in Q1 2025, five times more than 2024. Google increased token production by 50 times. Enterprise AI is spending up 300%. The cost is going down and the total spending is going up. Organizations are discovering they want far more software than they ever thought was worth building.
Internally, tools, automation, custom integrations, things that never cleared the costbenefit bar are now getting built. And here's the thing, despite all of this, experienced developer roles are stable or growing. The ADP payroll data covering 25 million workers alone show entry-level positions taking a hit, but not senior roles. The work is shifting towards architecture, judgment, complex problem solving.
So are developers being replaced? Some roles are getting compressed, but the total surface area of software being built is expanding faster than the compression. Does this mean everything's fine? No. It's complicated. Here's what makes this harder to see clearly. Everyone in the industry is racing to declare other people's jobs replaceable. Developers say they don't need product managers anymore.
Product managers say they don't need developers anymore. executives say they don't need anyone anymore. And if you saw my last video, we talked about how judgment is the real bottleneck when coding becomes easy. Well, that applies here, too. Every one of these roles contains judgment that the other roles don't see. The developer who just knows the system needs a cache layer.
The PM who knows users will hate a certain workflow even if it's technically correct. the support engineer who sees patterns the dashboard can't capture. The question isn't about does AI replace developers or is AI replacing PMs. The question is does the AI shrink the toil in each role and push the judgment bar higher. Toil is shrinking across every role.
The repetitive mechanical parts of every job are getting compressed. But as the toil goes down, the judgment requirements go up. The people who win are the ones who make the right calls, not the ones who crank out the most tasks. And this connects to something bigger. The way we train people is built for a world that's disappearing. Think about how education works.
You batch humans by year of manufacture. You force them to memorize facts for two decades. and then you just yeet them into a job where they use those facts for four more decades. Then they retire. This is a Prussian schoolhouse model. Build a big body of knowledge and execute on it for a long time. And this made sense in a world where knowledge was scarce and stable.
when it took years to learn what a doctor or an engineer or a lawyer needs to know and that knowledge stayed relevant for a career. But now that model which was already cracking before AI, you know, knowledge is not scarce anymore. It's abundant and it's not stable. What you learned 5 years ago in a fastmoving field might already be outdated. AI didn't break this model.
It's just made all the cracks impossible to ignore. If the knowledge part of knowledge work is being commoditized, what replaces the old learning model? Accelerated learning loops. Instead of memorizing a body of knowledge, you learn how to learn fast. And how you can use AI itself to do this. Self-directed learning with AI tools to build domain knowledge faster is going to be a core mode of working.
not replacing the learning but compressing the time it takes to get there. The people who will thrive are not the ones with the biggest knowledge base. They're the ones who can build judgment skills faster than others. And that's a completely different skill from memorization of facts. So what do you actually do with all of this? First, stop trying to know everything and start trying to make better calls.
Developers using AI tools who thought they were faster were measurably slower because they were using the tools to crank out more code not to get better at deciding what code to build. Second, learn the loops, not the tools. The specific products will change. Claw, cursor, copilot, whatever's coming next, but the feedback loops, verification, orchestration, those ideas will all outlast these things.
And if you've been following the series, that's what we've been building towards. Third, work on real problems with real stakes. Judgment does not come from tutorials. You make decisions. You see what happens. This is how it forms. Build something. Ship it. Watch what breaks. Learn from the breakage. That loop is how judgment forms. And one more thing, the way organizations are structured is going to change too.
And that's a whole other video we'll get into. But the short version is if your whole job fits into a narrow box, that box is shrinking. This is what this channel is for. Not for hype, not for fear, just practical techniques for working in this new reality. How to build the loops, develop judgment, and stay ahead of this thing while it's moving fast.
If your team is trying to figure this out together, I run a hands-on workshop for exactly this link down in the description. If you haven't watched the rest of the series, start with the jagged performance frontier video or the judgment bottleneck video on what skills actually matter. Now, links right here.