One axis was never enough
Ten problems, one dependency chain
These aren't independent pain points — each gates the ones after it. Tap to expand.
Six organizational stages
Stages describe the organization as a whole — its posture, not any single agent. Movement between them isn't guaranteed to go forward.
Ten independently-scored dimensions
Every dimension can sit at a different stage — sometimes inside the same team, at the same time. None of them can be averaged away.
Score your own organization
Set each dimension to the stage that honestly describes it today. Nothing here is sent anywhere — it's saved only in this browser.
The full 6 × 10 grid
Your self-assessment from above, plotted across every stage and dimension. Hover a cell for the stage's diagnostic text.
What moves you, what stalls you, what pulls you back
Progress is not monotonic. Every documented regression traces to one of three mechanisms below, and all of them cluster at the top of the range — not the bottom.
Triggers
Stalls
Regressions
The capability frontier
What no vendor can currently do, and where the boundary between two very different product categories actually sits.
Semantic understanding
Reads a prompt or response and judges intent against a natural-language policy, regardless of phrasing.
Structural enforcement
Knows this identity, this action, this resource, this context — and can intercept or block it in real time.
Sector, company type & posture
Three lenses for reading the same grid differently. Sector is the weakest predictor of the three — mandate and organizational shape matter more.
Requirements checklist
Adapted from a real enterprise buyer's MoSCoW-prioritized spec for an agent governance platform. Check what your organization already has to see your own readiness split.
Must have
Should have
Could have
From one axis to two structures
What changed from the original single-axis version, and where this model is honest about its own limits.
| Before | Now | Why it changed |
|---|
Where the evidence is thin
Stage placements are most reliable at S0–S3. Held to the same standard the model applies to everyone else: