Hello, agillm.io

This is the first post on agillm.io. The plan is to write about the parts of agentic LLM work that don’t fit neatly into papers: training pipelines, reward design, data curation, and the benchmarks that actually tell you something.

What to expect

• RL for agents — reward shaping, verifiable environments, failure modes at scale.
• Data curation — how task distributions get built, and how they go stale.
• Benchmarks — designing evals where the score means what you think it means.

Code blocks work

#v2
def advantage(rewards, baseline):
    return [r - baseline for r in rewards]

Inline code works too, along with tables, blockquotes, and footnotes.

The hard part is never the algorithm. It’s the environment.

Posts live in src/content/blog/ as Markdown files. Add one, push, and Vercel rebuilds the site.