▶
Pipeline
The thirteen-state lifecycle of an issue and the seven agents that move it through them. The pipeline is the integration; the agents are the parts.
See the pipeline → Product
One pipeline. Four surfaces.
Colony orchestrates the lifecycle of an engineering issue — from intake, through analysis and implementation, to a reviewed, merged pull request. Today it covers intake through merge and monitoring. Tomorrow it widens.
What shipping looks like — our own pipeline Colony has been running its own development for months. Commercial pilots are still ahead; the receipts you can verify are public. The full ledger and audit trail are at /built-by-colony/.
2,500+ PRs merged Through Colony’s pipeline on Colony’s own development.
250,000+ Lines of TypeScript Shipped, reviewed, merged through the pipeline.
10,000+ Tests, green Unit, integration, end-to-end.
$0.91 Median cost per PR · live Tokens plus worker time, attributed per merged PR — from the live ledger.
Four pillars
One pipeline, seen from four angles.
The same orchestration layer is the pipeline, the ledger, the governance gate, and the deployment surface. Different audiences see different facets.
$
Observability
Per-issue cost ledger, full audit trail, glass-box state transitions. Every dollar attributed; every decision recorded.
See observability →⌖
Governance
Review discipline, merge controls, conventions enforcement. Your engineering standards survive autonomy.
See governance →⌂
Deployment
Self-host on your infrastructure, run it managed in Colony Cloud, or talk to us about hybrid tenancy. No lock-in either way.
Self-host & cloud → How we’d measure your pilot
Six metrics. Read together.
We don’t publish forward-looking outcome ranges. We tell you the basket we’d measure your pilot on, so you can compute expected results against your own baselines. Commitment ranges live on the pilot page, where they belong.
T
Throughput
Issues closed per week, measured per active repo. Same team size, vs. your pre-Colony baseline.
C
Cycle time
Median time from issue filed to PR merged, on Colony-handled work. Reported per repo, weekly.
$
Cost per issue
Tokens plus worker time, attributed per issue. Visible in the same ledger your engineers see — not a separate finance dashboard.
E
Escape rate
Defects that originated in Colony work, divided by Colony PR count. Read against your pre-Colony defect baseline.
A
Agent attribution
Share of issues handled end-to-end without human steering. The number that says how much of the pipeline is actually autonomous.
R
Review cycles
Number of inspect-rework loops before a PR clears. Trends toward one as the conventions file stabilizes.
The most common mistake is optimizing for one metric in isolation. Throughput up and escape rate up means broken code shipping faster. Read the basket.
What changes for your team
Four human roles. The titles you have already.
Colony expands your team. The four roles below describe how work flows once the pipeline is running — activity-based, drawn from the titles you already have. A person can fill more than one; on small teams, one person fills all four.
Briefs the pipeline.
Files structured issues — context, requirements, acceptance criteria, test criteria, file references. Reviews Colony’s plan before development starts. Validates the merged result.
Common fit PMs, BAs, product designers, sometimes engineers.
Holds the quality gate.
Reads pull requests for architectural fit, correctness, and pattern alignment — the judgment calls automation can’t make. Calibrates trust in Colony’s output by reading its automated review alongside their own.
Common fit Developers, tech leads, architects.
Maintains the colony.
Owns the conventions file. Allocates worker capacity. Intervenes when work stalls. Tunes the system in response to what review reveals — the most common signal is a PR rejected for the same reason twice.
Common fit Senior engineers, architects, platform teams.
Charters the rollout.
Decides which repositories Colony is responsible for. Sets risk parameters — cost caps, automerge thresholds, human-review-required labels. Reads the metric basket and makes the expansion call.
Common fit Engineering managers, directors, VPs of engineering.
Most adoption work in the first ninety days is concentrated in two places: Authors learning structured issue writing (three to four weeks; the LLM-assisted drafting workflow shortens it), and Operators tuning the conventions file from real review feedback. The other two roles read the dashboards.
Want to see how this would land in your team? Talk to us.
Pilot scope back in two business days. Fixed-fee, time-limited, yours to walk away from.