The first bottleneck agents hit inside anthropic was the test runner, not the model. ci jobs grew…
the first bottleneck agents hit inside anthropic was the test runner, not the model. ci jobs grew 25x in six months as agents per engineer rose while code output hit 8x per quarter, and the fix was test impact analysis so each change runs only the tests it can touch.
the constraint moved from writing code to proving it.
Context
Anthropic's post of September 14, 2026 says its CI job volume rose 25x over six months. It says Anthropic engineers on average ship 8x as much code per quarter as they did from 2021 to 2025, that Claude authors 80% of that code, and that the amount of tests grew 10x with only a nominal increase in engineers. Not every test runs on every pull request, which is why the job count is not simply a product of those figures.
The post says the load threatened to overload the test impact analysis service several times. Three quick fixes lasted 70 days, 29 days and less than a day, and the team then redesigned the service. It says CI jobs increase as the average number of agents per engineer rises and that writing code is no longer the bottleneck.
The 25x CI growth over six months, the 8x code figure and the test impact analysis fix match the post. The post's 8x is relative to a 2021 to 2025 average per quarter. The note's 'hit 8x per quarter' compresses that and should be read as 8x the earlier rate.
The note says the first bottleneck agents hit inside Anthropic was the test runner. The post describes CI and the test impact analysis service as what came under strain, but a ranking of 'first' against other bottlenecks was not found in the page read. That is unsupported, not refuted. The post itself says the model was not the bottleneck for coding.
'The constraint moved from writing code to proving it' is the author's phrasing. The post says writing code is no longer the bottleneck, which is consistent with it.
Related work
- Agentic coding is straining CI. Here's how we scaled test impact analysis at Anthropic (Claude blog, September 14, 2026) ↗Primary source for the 25x, 8x and 80% figures, the three fixes and the redesign.
Watch next
- Read how the redesigned service selects tests and how it handles flaky tests. Look for similar scaling reports from other teams running agents.
Sources
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 9 October 2026 at 00:46 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 →