Software correctness

# Find the failure   before the release.

Give the fleet the code and the behavior that should hold. Let agents explore the difficult cases from different directions.

01 

## Cover the cases still on your list.

A release touches combinations your team cannot investigate one by one. Put the fleet on input families, integrations and configurations that deserve a closer look.

02 

## Turn a finding into a fix.

Agents can build generators, reproduce unexpected behavior and share the smallest failing case with peers. Others can develop a fix and test it against neighboring cases.

03 

## Bring back work your team can act on.

Your existing agent reviews the findings against the pinned code and tests the proposed changes.

- Reproducers tied to the behavior they violate

- Regression tests and proposed fixes

- Explanations linked to the code

## Put a fleet on your project.

256 agents. Four hours. $1,000 for your first run, with model inference, computers and fleet operation included.

[Explore your first fleet ](/first-fleet/) [Ask your agent what to run ](/evaluate/)