Sev.

Adrian Sevitz

← Thoughts about…

Open Tabs - 27 October 2026

(things I've spotted of interest)

You Must Draw Your AI Workflow to Control It

Chris Parsons introduces AWML (Agentic Workflow Modelling Language), a way to describe AI workflows in YAML and draw them as diagrams.

Chris Parsons has created AWML. Kind of like C4Model, but for AI workflows. As a visual person the output is rather useful. Not used it in anger yet.

BeijingPalmer on getting AI to sort out their PDFs

BeijingPalmer had AI organise and merge some PDFs in ten minutes, and the replies insisted they should have learnt to code it themselves instead.

Like I posted the other day. The cost of getting little problems solved is dropping to next to zero. There is a lot of toil that can be flattened. For all of the anti-AI this is real utility. The sub-thread is worth reading too. You’ve got a £10k pm dev at your fingertips. Why wouldn’t you get rid of toil.

Introducing TinyLittleOffice.com

Paul Ford and Aboard has created a tiny office suite where the whole document lives in the link, so nothing is saved to a server and there’s no logging in.

Cute. Clever. The entire document is encoded in the link, in the non-saveable bit of the URL. Not entirely sure of its utility but it’s clever.

38 things we’ve learned about building AI products

A set of lessons from four years of building [AI] products at PostHog.

Hadn’t heard of PostHog. Came across this last week. It’s a good list of things to noodle. Then saw them at SEV0 London (👋🏼 incident.io). Good talk by Ben White (look out for the SEV0 video when posted). Although felt their sales pitch at the stand didn’t explain the product to me. It does everything wasn’t a good pitch. Ben’s talk was a good pitch.

Even LinkedIn is getting tired of your AI-written posts

CNN report on LinkedIn stepping up its fight against generic, low-quality AI posts, including a new way to report them.

LI now has a Report as AI Slop button (clicked more than 1 million times in the first two weeks, apparently). They want you to be a bit more human. Although I’m not sure pre-AI LinkedIn was much better. That said, we all can see when something has been AI generated vs AI assisted vs maybe just human. What happens when an LLM can do personal tone better than you can? (Still not sure why, given all my emails and writing, this is so hard to do.)