
AI fatigue is real
Running agents instead of writing code felt like 10x productivity. Then I broke, and needed 4-6 weeks to recover.

Running agents instead of writing code felt like 10x productivity. Then I broke, and needed 4-6 weeks to recover.

Command line, thick clients, the web, and now agents. The interface itself is starting to look optional.

Big-bang rewrites used to be a bad idea for good reasons. AI agents just changed the math.
AI can write all the code. It still can’t magically create a great product - the taste has to come from somewhere else.

Unrestricted database access for AI agents is dangerous. Excessive restrictions defeat the purpose. Here’s how to find the middle ground.
My team spent years automating £50 million in promotional income. A three-day Sudoku app got me a standing ovation at dinner.
Under stress, people fall into a handful of default patterns - and pressure is the fastest way to find out who you’re actually working with.

The real cost of AI writing code is starting to show - what the data is telling us about the human/AI balance in software work.
On a busy week, I ship around 10,000 lines of code changes a day. There is no way to review all that line by line — and that’s not what review should be anymore.

Nearly 300,000 lines of code changed in my GitHub over 10 weeks. Most of it is throwaway. That’s exactly the point.