Platform / Optimize
Product 07Agents that continually optimize your CI.
Agents watch your CI and propose changes that make it faster and cheaper. Each one comes with the evidence, so you only merge what pays off.
Why it exists
CI changes are easy to suggest and hard to judge. A change can save runner time in one case and cost more to validate than it saves. Agents check each proposal against real runs and show the evidence before asking for review.
The mechanism, in the order a job meets it.
Agents continually inspect job and step durations, cache outcomes, selection rates, queue wait and superseded runs.
An agent suggests one focused change and explains the expected improvement, such as splitting a job or narrowing a cache key.
Targeted checks run against real past runs. Each check tests whether the proposed change supports its claim.
The proposal lists the checks and results, validation cost, saving per run and the number of runs needed to break even.
When the checks pass and the saving is worth the validation cost, a draft pull request opens with the workflow diff and evidence.
Each capability is part of the product today.
Agents continually inspect your CI and look for one focused improvement at a time.
Each proposal runs targeted checks against runs from your repository and reports which checks passed.
See the proposed change, runs used, validation cost, saving per run and break-even point in one proposal.
The proposal shows validation cost divided by saving per run, so you can judge the payback directly.
A draft pull request contains the workflow change and its evidence. You decide whether to merge it.
Agents open draft pull requests only. They do not run changes on the default branch or edit product code, secrets or repository settings.
If checks fail or the saving does not cover validation cost, Optimize leaves the workflow unchanged.
Sample / bash
proposal split test:parser into 4 shards
verified 20 real runs
checks 20 / 20 passed
validation 1m 10s, once
saving 3m 10s per run
break-even after 2 runs
net 9m 33s -> 6m 23s per run
result draft pull request #418Illustrative proposal. The shape is the product: a change, its evidence, and its arithmetic.
No. It proposes a diff against a workflow file and opens it as a draft pull request.
A linter checks rules. Optimize uses your run history, tests a proposed change against real runs and explains the evidence.
The agent selects targeted checks that could show whether the proposed change works. The proposal lists each check and its result.
The proposal shows the runner time used for validation and includes it in the break-even calculation.
Yes. That is why every change is checked against real runs and opened as a reviewed pull request instead of being applied automatically.
Yes. Scope it to one workflow, or run it on demand while investigating a slow pipeline.
Start
We will set it up with you and show you what changes.