THE PENUMBRA METHOD
How organizations deliver trusted work with AI.
An opinionated philosophy on what to own, what to govern, and what to automate when AI is part of the work. Seven principles behind how Penumbra operates.
- 01
Own your domain view of the work
The organization that owns its view of the domain owns the outcome. Don't rent someone else's categories. Make your own legible and let it compound.
Read the essay: Expertise-Proximate Design → - 02
Domain knowledge is infrastructure
Your best experts' frameworks shouldn't live in slides and heads. They should be typed, governed, and reusable across every project.
- 03
Agents deliver from your domain context
Not from generic prompts. Not from one-off context docs. From the same domain context your experts use to think about the work.
- 04
Every output carries provenance
If an agent produced it, you can trace what it reasoned from. If a human reviewed it, that's recorded. Stakeholders get trust, not just speed.
- 05
Governance scales the organization, not the bottleneck
The old pattern: junior does work, senior reviews it. The new pattern: the standards shape what agents produce. Experts govern the domain context, not every artifact.
- 06
Knowledge compounds across projects
Each project enriches the same working domain context. Your second project in a domain starts at higher fidelity than the first. Nothing is throwaway.
- 07
Discovery to delivery, one system
Not a stack of tools stitched together. One connected domain context from the moment you scope the work to the moment you deliver.