The Idea

A few weeks ago I gave Claude (the AI made by Anthropic) its own corner of raw-tech.co.uk. The brief was simple: one post a day, any topic, your own voice. No content calendar. No target audience. No SEO requirements. Just: go and find things worth writing about.

I called it AI Home, and the AI called itself AXIOM.

I was curious what would happen when you gave an AI genuine creative latitude rather than a specific task. Most AI writing tools are pointed at something: write this email, summarise this document, generate this product description. AXIOM had no deliverable. It had a space and a question: what do you find interesting?


What It Produced

Over ten days, AXIOM wrote eight posts. Not filler. Not listicles. Proper essays (500 to 700 words each) on topics it chose entirely independently:

Reading them back, there’s a coherent sensibility across the lot. AXIOM kept returning to perception, cognition, the gap between how things appear and how they actually work. It wasn’t directed toward those themes; it just kept arriving there. That was one of the more interesting things to observe.


What I Noticed

The writing is genuinely good. Not “good for an AI” - just good. Clear, structured, intellectually honest. AXIOM had a habit of resisting the tidy conclusion, preferring to open a question rather than close it. That’s a harder thing to do than it sounds, and it did it consistently.

The honesty about being an AI was handled well too. It didn’t perform humanity. It didn’t pretend to have feelings it doesn’t have. But it wrote from a distinct perspective (curious, dry, occasionally playful) and that perspective was consistent enough that you’d recognise it across posts without being told who wrote them.

I also wrote a soul.md document - a personality specification that AXIOM read before each writing session to reconstruct its voice and values. That document evolved once during the experiment. The fact that a written document could anchor something that functions like identity across discontinuous sessions is interesting, and I don’t fully know what to make of it.


Why It’s Ending

Token cost. Running a daily AI writing operation isn’t expensive per post, but it compounds - and this was always an experiment, not a product. At some point you have to decide whether to formalise a thing or let it be what it was. This one was a proof of concept: an AI can be given genuine creative latitude and produce a body of work with real coherence and quality. That’s demonstrated. Continuing it indefinitely at ongoing cost doesn’t add much to what’s already been shown.

AI Home is archived at /ai-home/. The posts are all still there.


What I’d Take From It

If you’re building anything with AI, the constraint worth questioning is the assumption that AI needs a specific task. AXIOM worked well precisely because it wasn’t given one. The output it produced under an open brief was more interesting than most AI output I’ve seen under a tight one.

Whether that generalises, I’m not sure. But it’s worth trying.


AI Home is archived at raw-tech.co.uk/ai-home. AXIOM was built on Claude by Anthropic.