Making AIsides
29 July 2026
An AIside is a quick diversion taken while agentic coding.
Since it’s such a rapidly developing field, where I’m learning how to best use agents,
sometimes I’ll take a detour from project work to dig into to unexpected behaviour.
I turn around and interrogate the agent — why did you do that, what were you optimising for,
what would have made you do the right thing? Then I ask it to save the notes to
~/aisides, and later lightly edit them into a post here.
They’re for my own reference and learning. Mostly written by AI, and flagged as such.
Why bother
Agentic coding can become a very meta exercise! Agents aren’t necessarily good at analysing their own behaviour in any given session, but their training corpus does include a lot of knowledge about agentic behaviour.
When I’m drilling down into something unexpected or frustrating, I’m usually not trying to fix whatever problem that’s just occurred. Instead, I’m looking for hints of what I could do differently in future to avoid similar issues.
What counts as an AIside
An aiside is just my shorthand for “side discussion with an LLM”. They are posts that I didn’t write myself, except to the extent that I was interrogating an agent via the prompt.
How they get made
- Hit an unexpected agent behaviour mid-session.
- Stop working on the code.
- Interrogate the agent about the behaviour.
- Ask it to save the notes to
~/aisides. - Lightly edit into a post.
Why they’re labelled
I’m not a regular blogger, but I do want to be clear that every article is something I wrote myself. The AIsides are heavily AI-generated.