AI & LLMs

The deepest channel on Edge. Foundation models, agent architectures, retrieval systems, evals, and the moving line between research and production.

This pillar covers the work that determines what AI products can actually do. New model releases get filed here when they shift capability or cost in a meaningful way, alongside the harder material from the labs and the practitioners who turn it into shipping software. Read it for primary sources rather than recap blogs: lab papers and notes, retrieval benchmarks, agent traces, eval methodology, and the long-form essays that hold up six months later.

Two threads run through everything filed here. The first is what is genuinely new at the model layer: capability cliffs, training recipes, alignment work, the shape of the next deployment cycle. The second is what works in production: which patterns of context engineering and tool use compound across teams, where retrieval beats fine-tuning and where it loses, what the operational tax of running an agentic system actually looks like.

The summaries below are sorted by recency. The pillar refreshes as new entries land.