Agent traffic now dwarfs human traffic in production. microsoft's first production-scale copilot…
agent traffic now dwarfs human traffic in production. microsoft's first production-scale copilot study sampled 3.2m users, 13m sessions and 95 trillion tokens, finding 87% of llm calls are agent-initiated and cache hit rates fall 26 points at agent depths.
the cache was built for chat, not for agents.
Context
The arXiv preprint 2608.00101, submitted 30 July 2026 by authors from Microsoft Azure Research and UIUC, samples GitHub Copilot traces from June 2026: 3.2M users, 13M sessions, 761M LLM calls and 95T tokens. It reports about 87 percent of calls agent-initiated against 13 percent user-initiated, and KV cache hit rates averaging 90 percent within a turn and falling to 55 percent across turn boundaries. A figure caption says same-model boundaries cause a 26 percent average drop and model-switch boundaries near-complete invalidation.
These are GitHub Copilot coding-agent traces, not all Copilot or all agent traffic. The 87 percent is a share of LLM calls inside those traces, not traffic versus human web traffic. The 26 in the paper is the average drop across same-model boundaries, with relative or points unclear in the caption, so fall 26 points at agent depths is not supported as stated; the 90 to 55 percent is a 35 point gap. The cache was built for chat, not for agents is the author's take.
Related work
- Earlier note on the same study ↗Same paper.
- Second earlier note on the same study ↗Same paper.
Watch next
- The peer-reviewed version and other vendors' serving traces.
Sources
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 20 September 2026 at 08:02 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →