I realised last week that I've written my last line of code

I realised last week that I’ve written my last line of code. 😳

I’ve been building apps with AI for the last few years. But for any serious project, I’d still start the traditional way: writing the foundational structure by hand. Last week that changed for good.

Using Claude Sonnet 4.5, I went from concept to architecture to prototype to near-production code for a document intelligence pipeline. All in just a few hours, and all through conversation. When I direct it well, it challenges my assumptions, pushes my design, and builds faster and better than any developer I’ve worked with in 25 years of running engineering teams.

That doesn’t mean I care less about the code. I care more. It’s higher quality now, ruthlessly consistent, and delivered at ridiculous speed.

I’ve also hired my last developer to physically write code. From here on, our job is to use our expertise to tell the models what to build. Writing code has gone the way of long division. It doesn’t matter how good we ever were at it. It’s just not a today skill anymore.

I found that realisation a little confronting, but only briefly. The productivity gains I can now make, for my business and for my clients, are extraordinary. 🙂

I’d love to hear how you’re finding this shift: confronting, exciting or both?

The key to AI-authored code success

The key that makes AI-authored code work for me is the design phase. I have a detailed discussion with the agent about the project, and together we create the commercial project brief, user stories, technical architecture and test plan as simple markdown documents. I care about every detail in those design documents. Because if we get those right, the chances of success for the development phase are very high.

During development, Claude 4.5 is smart enough to alert me when we make iterative decisions that contradict the design. It prompts me to make a call: are we adjusting the design, or getting back on course with the dev? That’s priceless.