The Kojen blog

Notes from the dojo.

How to train AI agents that hold up: practice loops, honest evaluation, red-teaming and shipping. Every claim is sourced.

10 articles

Blank glass pages falling into a glass funnel, with a small obsidian stone with a red seam beneath its spout

Training

Can you just feed an LLM your documents?

Training on raw text teaches style more reliably than facts. What the research says about teaching models new knowledge, and when retrieval beats training.

Nov 5, 20269 min read
A glass scale with glowing red edges, a small obsidian stone on one pan and a large obsidian block on the other

Training

What it really costs to train an LLM, from $48 to $100 million

The arithmetic behind training budgets, estimated costs by model size, what real runs reported, and what the headline numbers leave out. Prices as of November 2026.

Nov 5, 20268 min read
A faceted obsidian computer tower split by a glowing red seam, with a circuit of red light branching from its top

Training

What LLM can you train on a $10,000 computer?

What a $10,000 machine can train from scratch, what it can fine-tune, and whether renting beats buying at November 2026 GPU prices.

Nov 5, 20267 min read
A glass torii gate with a row of glass tiles leading to it, the nearest glowing red

Practice

Practice beats prompting: why AI agents need a dojo

Agents that work once aren't agents you can ship. Benchmarks show how inconsistent they are across repeated runs, and why a disciplined practice loop is the fix.

Nov 4, 20267 min read
A glass brush beside a red stroke on the left, a glowing ingot on an anvil on the right

Training

Prompt tuning or fine-tuning? Pull the cheaper lever first

Instructions, model choice and weights are three levers with very different costs. What the research says about each, and how to decide when fine-tuning is actually worth it.

Oct 28, 20268 min read
A grid of glass cards with a third set apart behind a thin red line of light

Evaluation

Your agent is only as good as its test set

Before you tune a prompt or pick a model, build the thing that tells you whether either worked. A practical guide to success criteria, edge cases, held-out data and formatting.

Oct 21, 20268 min read
A dark metal balance scale holding a glowing red cube and a clear glass sphere

Evaluation

LLM-as-a-judge, honestly: when you can trust an AI grader

AI judges agree with humans about as often as humans agree with each other, but they also have documented biases. What the research shows, and how to set up a judge you can rely on.

Oct 14, 20267 min read
A faceted glass shield deflecting red streaks of light coming from several directions

Safety

Red-team your agent before your users do

Prompt injection tops the OWASP list for LLM apps, and agents with tools raise the stakes. A practical checklist for adversarial testing and stress testing before launch.

Oct 7, 20267 min read
A glass ring forming a loop with red light running along it and a small gateway rising

Shipping

Shipping is where training starts: deploy, collect feedback, retrain

A deployed agent drifts as the world changes. How to monitor it, attach feedback to individual predictions, and turn real failures into the next round of training.

Sep 30, 20267 min read
A ring of glass figures linked by red threads of light to a glowing central orb

Developers

Let your agents run the dojo: training AI over MCP and REST

Training an agent is mostly orchestration, which is exactly what agents are good at. How Kojen exposes its whole training loop through an MCP server, a REST API and the Kojen Agent.

Sep 23, 20266 min read