Six Months Until the End of Coding as We Knew It
The World Economic Forum is known for big statements that rarely materialize. This year, however, Anthropic’s Dario Amodei didn’t come to sell a vision-he came to deliver a timeline for the future of AI coding. And that timeline is close enough to be uncomfortable.
The AI Coding Timeline
In three separate interviews in Davos-with the Wall Street Journal, Bloomberg, and on a panel with DeepMind’s Demis Hassabis-Anthropic’s CEO made a specific prediction: 6–12 months until AI models can do “most, maybe all” of what software engineers do, end-to-end.
That prediction didn’t come out of nowhere. Anthropic already has engineers who don’t write code anymore-Claude writes it, and they edit and organize it. The head of Claude Code said he hasn’t written a single line over the past two months.
Most impressive? Claude Cowork-the new tool for non-developers-was built almost entirely by Claude Code in a week and a half.
The loop has closed: AI that writes code → AI that builds tools to write more code → AI that accelerates the development of the next AI.
Why This Time Is Different
If you follow the AI space, you’ve heard similar predictions dozens of times. Amodei’s argument, however, isn’t based on hype. It’s based on numbers:
- 2023: $0 → ~$100M in revenue
- 2024: ~$100M → ~$1B
- 2025: ~$1B → ~$10B
10x growth, three years in a row. Anthropic is targeting $70B ARR by 2028.
Amodei described something he called “Moore’s Law for Intelligence“-models’ cognitive capability doubles every 4–12 months. While public opinion swings between “it will change everything” and “it’s a bubble,” the evolution of capabilities remains exponential and steady.
The Counterpoint: What DeepMind Says
Demis Hassabis took a more conservative stance. Coding, he said, is easy to automate because it’s verifiable-you can tell immediately whether it works. The natural sciences require experiments, years of validation, and human judgment.
He emphasized a critical gap: today’s models solve problems you give them. They still can’t identify which problems are worth solving.
Both CEOs agreed on one thing, though: they would prefer to have more time. But geopolitical competition doesn’t allow them to take breaks.
The Scenario No One Wants
Amodei warned about something we haven’t seen before: high GDP growth and high unemployment at the same time.
In the past, new technology meant disruption-but also new jobs. The automobile killed the horse industry, but created millions of jobs. The internet eliminated some roles, but gave birth to entire industries.
AI may be different. We could see 5–10% GDP growth and 10% unemployment, simultaneously.
The most worrying scenario? A “zeroth-world country” of 10 million people (mostly in Silicon Valley and other tech hubs) that forms its own economy with 50% growth, while the rest of the world falls behind.
“I’d almost call it dystopian,” Amodei said. “And we need to think about how to prevent it.”
What It Means for Us
6–12 months for AI to write most production code. A generation of junior developers that may not find the traditional entry point into the profession. An economic model we haven’t seen before-wealth and unemployment at the same time.
Amodei is neither optimistic nor pessimistic. He’s a realist. And his realism says something simple: reality will force policy to change. The question is whether it will change fast enough.
For anyone working in software, the message is clear: coding is being fundamentally transformed. Those who are first to understand how to work with AI tools will be the ones who remain.
