Research and reports on how teams actually adopt AI — measured on real workloads, published openly.
Aug 2026
Retrieval benchmarks
Blind-judged answer quality on real workplace questions over 220K-document corpora.
Jul 2026
Model leaderboard
How frontier models perform on grounded, team-context tasks — updated as they ship.
Jul 2026
The borrow rate
What share of AI tasks get solved with an existing agent instead of a new prompt.
How grounding changes week-one adoption curves.
What it actually costs to self-host at 100, 1K, and 10K users.
Mapping who builds and who borrows inside 100 team libraries.