A Google developer blog post estimates that pausing a coding agent after two idle minutes and returning its machine to a shared pool, which the author calls "parking," saves 40% on machine costs against one machine per session at 2 prompts an hour, and only 4% at 4 prompts an hour.

Besides tokens, an agent costs the sandbox and machine it holds while waiting for the user, and the 2 prompts an hour is the author's assumption. The author also says prompt frequency is the least certain input in the whole estimate. The post cites a study by Banruo Liu and others of GitHub Copilot agent logs: agents spend about 80% of their time waiting for users, and each turn involves an average of only 396.3 seconds of agent work (Google Developers Blog, 2026-11-16). The author's estimate runs 100 coding agents on GKE, Google's Kubernetes service: one Pod per session needs 34 machines, about US$1,288 a month, while parking needs 15 standby machines plus 1 control machine. Above about 4.5 prompts an hour, parking stops paying off. Another approach, overcommitting CPU without parking, needs only 9 machines for 100 agents; the author's 40% estimate doesn't include it. How we checked →
Source and checks
This item ran in the brief of Nov 17, 2026, alongside 13 other items. How we checked it, the quote checks and what was still missing that day are on that issue's How we checked page.
Cite this item
SecondSource (2026). "A Google developer blog post estimates that pausing a coding agent after two idle minutes and returning its machine to a shared pool, which the author calls "parking," saves 40% on machine costs against one machine per session at 2 prompts an hour, and only 4% at 4 prompts an hour.." https://secondsource.io/en/product/2026-11-17-1/. Accessed access date.