Friday notes. A few things crossed my desk this week worth a comment, plus what went live here.

Headless is not the brave choice anymore

The 2026 numbers are in and they are blunt. Depending on whose survey you read, roughly two thirds of enterprise organisations are now running a headless approach, and a similar share of new stacks are going MACH-based. CloudTweaks has a decent rundown if you want the figures. I have been making this argument since it was the contrarian position, so I should be pleased. Mostly I am. But there is a catch worth saying.

When something becomes the default, the failure mode changes. A few years ago the risk was missing out. Now the risk is cargo-culting. Plenty of teams are going headless because the deck told them to, not because they have a reason. Decoupling a ten-page brochure site buys you complexity and a slower build, not speed. The architecture is a tool, not a badge. If a well-cached traditional site would serve your customer faster, that is the right answer and I will tell you so.

The agentic commerce noise

The other theme everywhere this week was AI moving past product recommendations into what everyone is now calling agentic commerce. Bots that browse, compare and buy on a shopper's behalf. The interesting, unglamorous detail is why composable keeps winning here: an API-first backend is simply easier to plug an AI agent into than a monolith. That is the real reason, not the marketing.

My caution is the same one I give about any AI project. Most businesses asking about agentic commerce have not sorted their product data yet. An agent sitting on top of messy data does not fix the mess. It just makes confident mistakes faster, and at checkout that costs real money.

What went live here

I published a piece this week, 'I don't write code the way I used to', on how AI changed the job rather than the headcount. It connects to all of the above. The hard part was never the typing. It was deciding what to build, and that has only got harder. More next Friday.