Chainguard engineer keeps Factory Droid sessions running for weeks without losing context
Chainguard's engineering team works across large internal codebases and a constant flow of upstream open-source package updates and requests. Staff Engineer Josh Wolf tried memory MCPs and other memory tools without success, then found he could run the same Droid session for two weeks straight without losing context awareness or quality, which he credits to Factory's compaction approach.
- Agent or tool
- Factory Droid
- Topic
- Software Development
- Reported on
- Case study (Factory)
- Added
- 2026-11-13
Why it matters
Coding agents are moving from autocomplete to doing whole tasks: reading a codebase, making a plan and changing many files. This case is a concrete example of where that works in practice. It is one of 221 sightings filed under Software Development on Agent Sightings (number 38 by interest), and one of 3 involving Factory Droid.
How you could try it
Here is a reusable prompt, written by us in our own words, that captures the task. Paste it into Factory Droid or any similar agent and adjust the details to your situation.
I want a coding agent to take on work like this across our repositories, show its plan, and wait for review before anything merges. The tool in the source case was Factory Droid.
Tip: Work on a fresh branch, ask for a plan before edits, and review the diff and test results before merging.
Source
Read the original first-hand account on Case study (Factory): https://factory.com/case-studies/chainguard. Our summary is a paraphrase; details are in the source.
Featured in these guides
Built an AI agent? Feature your agent on Agent Sightings, free submission or a featured spot.
Browse more like this
Related sightings
- Blackstone runs Factory's coding agents inside its own data centre with a choice of models
- Groq equips its engineers with Factory's model-agnostic Droid coding agents
- A contractor's personal Claude Code agent allegedly ran inside a secure Pentagon system for two months
- Claude Code's 3-word status line reportedly burns 15 to 20% of tokens