AI isn't replacing jobs. It just found its natural habitat.

I'm Amos, co-founder of Swan AI. I'm building the first autonomous business and documenting every step. $10M ARR per employee. No bloat. No theory. If that's not the game you're playing, reply to unsubscribe.

The CEO of Anthropic just predicted AI will make most jobs obsolete.

He's right about the evidence. Wrong about the conclusion.

And that distinction changes everything.

I just spent 2.5 hours listening to Dario Amodei's podcast with Dwarkesh. One of the most honest, dense conversations I've heard in years. Genuinely worth your time. But there's one idea buried inside it that I think Dario gets fundamentally wrong. And if enough people accept it unchallenged, it steers an entire generation of founders and operators in the wrong direction.

Here's his argument in its strongest form.

At Anthropic, AI is already writing nearly all the code. Soon it won't need humans to review it. Then it won't need humans to plan what to build. Then the whole loop runs end to end without a single human involved.

Software engineering was supposed to be one of the hardest domains to automate. It requires logic, creativity, problem solving, deep domain knowledge. If AI can crack that, the implication feels obvious: no other job is safe.

The evidence is real. But the conclusion it produces is false.

Dario is asking the wrong question. And that's where the extrapolation error begins.

When a species dominates a specific environment, biologists don't ask "will it dominate everywhere?" They ask: what is it about this environment that made dominance possible?

Nobody is asking that about AI.

Dario is looking at AI excelling in a software-native environment and concluding it will dominate all environments. But software is a special case. And understanding why changes everything.

Software is digital. Logical. Built on formal language. And this is the part that matters: verifiable. You can test whether code works. Run it, check it, fix it, repeat. The feedback loop is tight. The system is closed. Success and failure have a binary definition.

This isn't a coincidence. It's the reason.

Software isn't just a domain where AI happens to perform well. It's the one environment on earth almost perfectly designed for how AI operates.

Dario is watching AI dominate in its natural habitat and concluding it will dominate everywhere. But, this is an extrapolation error.

Consider what most human work actually looks like. A creative brief. A sales call that reads the room. A strategic decision made on incomplete information. A negotiation. A piece of writing that lands.

None of these produce a binary output. None of these have a tight feedback loop. The gap between these two categories of work is much larger than the current AI discourse is willing to acknowledge.

So if the prediction overshoots, what's actually happening?

Something more specific. And arguably more significant.

Software itself is evolving.

For decades, software worked one way: humans wrote the code, deployed it, and every user got the same static version of the product. The software was fixed. The user adapted to it.

That era is ending.

We're entering the generative software era.

And it changes three things most builders haven't fully processed yet.

Software is starting to generate itself around what each user needs. The same underlying system produces different experiences for different people, not through configuration, but through autonomous generation. The product isn't fixed anymore. It's fluid.

Systems are beginning to participate in building and rebuilding themselves. This is the part of Dario's argument that's actually right, but it describes a transformation of the tool, not the elimination of the human. The loop changes shape. It doesn't remove the human from it.

The speed of creating, changing, and adapting software is increasing by orders of magnitude. Competitive advantage no longer comes from being able to build. It comes from knowing what to build, and being able to test and adapt faster than anyone else.

AI isn't replacing human work. It's transforming the most important tool humanity ever built. The tool is changing. Not the species.

If you want to learn and build using generative software, I'm keynoting in Boston.

Decision Circuits is a hands-on agentic GTM bootcamp for founders, operators, and growth teams. No code. No theory. You leave with a live growth playbook you built in the room.

I'll be on stage at 3:15 pm. You'll be building agentic GTM workflows by 16:00.

Reserve your seat

Are you using generative software, or just a SaaS with an LLM bolted on top?

Three questions to find out.

1. Is the product static, or does it adjust to your needs?
A SaaS tool gives every user the same experience. Generative software shapes itself around how you actually work. If your tool looks identical for every user, you're not in the generative era yet.

2. Are the capabilities constrained or extensible?
Can you discover use cases the product wasn't explicitly built for? Or are you limited to a fixed menu of features? Generative software expands with you. SaaS with an LLM bolt-on doesn't.

3. Are you configuring it, or improving its context?
Configuration is static. Context is dynamic. If you're spending time on settings, you're using old software. If you're spending time teaching the system, you're in the generative era.

The tool is changing. The species isn't. But only the people who understand the difference will know what to do next.

-Amos

Community Notes

I’m Amos Bar Joseph, co-founder of Swan, the first Autonomous Business OS. At Swan, we’re building what we call the Autonomous Business: a company that scales to $10M ARR per employee with no bloat, no assembly lines, no Cog Culture. Just humans in their zone of genius, amplified by AI agents.

I write The Autonomous Age to share contrarian insights from that journey, on GTM, leadership, and the future of work. If you want to understand how GTM evolves beyond playbooks and assembly lines, this is where the story unfolds. Connect with me on Linkedin or X.