90% of Code will be written by AI

In March 2025, the CEO of Anthropic (Claude) CEO Dario Amodei made a bold prediction that in 3–6 months AI will write about 90% of all code and that in about 12 months AI will write 100% of all code.

So how are we doing, 9 months on?

Wharton professor, Ethan Mollick (https://lnkd.in/e_a5NETf) set the provocation in this LI post (https://lnkd.in/ekFsgTmx)

Although it may have taken slightly more than 6 months, the comments suggest the results are beyond expectations and give a clearer picture of how AI is shaping the world of coding.

Not only that, whilst engineers are best equipped and benefit most from using AI coding tools, they are making coding accessible for non coders too, albeit with limits.

Tim Monzures highlights that leverage as a coder has shifted to:
- Precise prompting (constraints, edge cases, non-goals, interfaces)
- Interrogating assumptions (what did it silently decide?)
- Reading + verifying like your job depends on it (because it does): tests, types, reviews, threat modelling.

Not everyone is using AI for coding though, and whilst it works well for new (green field) development, it may be harder with existing infrastructure.

It is a stark reminder of the importance of clarity, of defining things well, the boundaries, the constraints, the assumptions and testing for the interpretation of others.

Other comments include:

"AI writes 90% of my code, but my needs have ballooned—I'm architecting more complex systems, writing more specs, and debugging more. I write less code but work harder than ever."

"Working with Claude Code is surreal, I increased my speed 3x to 4x!"

"It’s been writing 95% of mine for the past year."

"The interesting shift isn’t “AI writes the code,” it’s that humans move up the stack faster, from syntax to intent, from implementation to judgment."

"This has been the case for me. ~90% is accurate!"

BESCI AI OPINION

"I’ve been at 100% since 8-9 months and development is less and less about coding but more and more about requirements engineering and QA."

The shift from coding to the importance of having clear expectations, boundaries, clarity of what good and bad looks like, feels very similar to the advice we give to leaders leading transformation programs.

The shift from technical, hands on, skills to those that enable others to work and management skills.

The field of Behavioural AI will expand to understanding how the instructions given became the outcomes given - sound familiar?

Instead of codifying predictable patterns in human behaviour, we will predict and diagnose the patterns in agentic AI.