Key results
The company
Chainguard
chainguard.devSecure software supply chain platform and zero-CVE container images.
Result highlights
- 2-week continuous session with full context
The story
A developer of secure open source software manages large internal codebases and a constant influx of upstream package updates requiring rigorous engineering review at every step.
Engineering tasks require deep context from multiple repositories, but standard AI tools lost memory during complex workflows spanning brainstorming to production. Developers had to manually curate skills and commands to seed context windows, making long-running tasks impossible to automate effectively.
The team deployed AI agents with a compaction engine that preserves context across multi-week sessions without manual seeding. Developers teach the agent a task once in a collaborative session, enabling the system to subsequently operate independently on similar patterns. The workflow manages git work trees to keep agents isolated while maintaining a continuous thread of work from design to deployment.
Quotes
“Like most real engineering problems, our use cases require a lot of context from various upstream sources and across multiple internal repositories. Specifically at Chainguard, where we juggle both our internal codebases and an ever-growing chunk of the open source ecosystem.”