After code becomes cheap

Dario Amodei of Anthropic made a bold prediction: within three to six months AI would write 90% of code, and within twelve months it might write essentially all of it. Sam Altman of OpenAI shares this view: code is gradually becoming just an intermediate link between an idea and a finished product.

The facts suggest we’re heading in that direction: in Q3 2024, Google said that more than a quarter of all new code at the company was generated by AI, then reviewed and accepted by engineers. People who have never written a line are building working applications simply by describing what they want in plain language. What used to take large budgets and months of work can now be launched for a few hundred dollars in just a few hours.

There are skeptics, though. Cognitive scientist Gary Marcus argues that current generative AI can speed up coding without supplying the deep conceptual understanding that genuinely novel systems require. The practical caveat resembles the familiar 80/20 pattern: getting to a working solution is relatively easy; getting it all the way to real production is another story entirely. Current tools still struggle with fully debugging and maintaining complex projects over the long haul.

But the organizational shifts are already underway, and the nature of development itself is changing: designers and managers now turn ideas into code themselves, startups launch without technical founders, and small teams using AI outpace corporations.

Still, senior engineers needn’t panic. The industry isn’t dying; it’s evolving. Routine tasks really will move to AI, but strategy, architecture, maintaining large systems, and understanding what customers actually need will remain in human hands. In my view, the key skill of the coming years is learning to work in partnership with AI systems.

Which AI-powered tools are you already using?
Share your experience — I’m always curious to see how this is reshaping our day-to-day work.