A workshop, not a product. One idea at a time, built to demonstrate and prove the idea, and then put in front of real people, with the findings written about, including the parts that didn't work.
Most of what gets built with frontier models is a demo or a pitch: a thing that exists to raise money or to go viral. 8epochs is neither.
It's a place to take a single question seriously enough to build the entire answer: a real, usable thing, shipped to real people, kept alive long enough to learn something. The projects are small on purpose. Depth is the point; breadth is a distraction.
Some of what's here intends to be directly useful. Some is closer to art. What they share is that the AI is load-bearing: it's doing work a person and non-AI software can't do at that scale or in that shape.
The writing-up is not a metaphor: what each experiment revealed (the parts that worked, the parts that fell over, what it cost) gets written up on Julian's blog at julianreed.net.
In machine learning, an epoch is one complete pass through the training data. Models don't learn in a single look; they get better over successive passes, each one correcting the last.
That's the working method here too. Build it, use it, learn what's wrong, go again. The first versions are there to be validated, 'warts and all', and improved on the next pass.
An epoch is also an era, and the quiet premise of this site is that we're at the start of a significant one. As for the 8: stand it upright and it's a loop with no end. Loops provide the framework for AI 'actors' to work as an autonomous system that iterates until the goal has been achieved. The loop that resembles the letter 8 is the mark in your browser tab, raised to the power e.
8epochs is independent and built by one person, Julian Reed. It isn't a company, doesn't have a roadmap, and isn't affiliated with Anthropic or any model provider. It's simply the place the experiments live.