How would you train an LLM agent for long-context, long-horizon tasks? What environments and rewards would you build?

Build environments that mirror the work: a sandbox with a document store, search and editing tools, and tasks spanning dozens of steps over long files. Warm-start with supervised fine-tuning on expert trajectories, then RL. Prefer verifiable rewards, like the right clauses found or the final document passing checks; use a calibrated judge for the rest, a small step penalty, and a curriculum from short to long tasks. Read transcripts for reward hacking.