ai

I make extensive use of AI tooling to augment my work - in creating tools and projects, my writing pieces (both enabling researching and soundboarding my ideas, and generating rough drafts to get me started), and for ideation.

So much of my work and personal thesis around the use of AI rests on scrutiny, verification, and clarity, that it would feel disingenuous not to be transparent around it.

The main tools I've used up until now have been Claude.ai and Claude Code - now I'm branching into more open-weight models. These move and update regularly enough that it feels pointless to list granularly, but Qwen and Mistral feature heavily.

In my writing, nothing is simply AI > dropped straight into the page. Every piece has either been drafted initially by me, or I've provided the arguments, notes or outline, alongside existing context on my preferred writing styles. Then I draft the piece together with AI, and edit and mould the piece until it comes across and makes the point I want it to. Every factual claim I make in a piece comes from me, and is verified by me.

Pangram and other AI detectors will likely flag as mainly AI due to this approach. C'est la vie!

In terms of coding projects - almost all of my recent projects have been driven by my customised, battle-hardened workflows, with tactical-level code generated automatically, tweaks made here and there by me. I generally provide the input and rigorous testing and verification to produce detailed handoff specs that an agent can then produce, with strategic-level guidance, course-correction and inputs throughout.

I also tend to use the above flows to construct the desired infrastucture and setups for my research experiments - I design the experiments themselves, and verify and check the results myself.

So - hopefully this gives a decent overview of my position on the use of AI in my own work. It's something I've grappled with here and there, but in the end, it's the route I've decided works best for me.

Last updated - Oct. 2026

the blog