Agents do the work.
One human owns the call.
This is the model.
Building software got cheap. The work moved to judgment: deciding what is worth building, then knowing whether what shipped actually landed. Three views of the same team make that call visible. How the team runs, how it builds, and whether what it ships is really AI-native. Read together, they are the pit wall: one live surface where a person decides, mid-race, where the next lap of effort goes.
The model on this page, running against a real company.
One company, read from public information only. Fixed code computes every figure, so no number is guessed.
Open ThroughlineWhere the calls get made, what evidence they rest on, and whether each one has a single owner. 27 dimensions across 6 functions: strategy, design, development, intelligence, operations, go-to-market. Five stages, from reacting to compounding.
How clearly the work gets described, how much context the agents get, and what gets checked before anything ships. 34 tasks across 6 stages: Specify, Context, Orchestrate, Validate, Ship, Compound. The loop a human-and-agent team actually runs.
Whether the product is AI-native or AI on the label: how deep the agents go, and whether the product gets better the more it is used. 27 dimensions across 6 attributes. Five stages, from a thin wrapper to a product that compounds.
One view is a snapshot. Three views, read together, tell you what to do next.
A single view misses the gap between the three. Engineering metrics tell you how fast. Product analytics tell you what gets used. Team surveys tell you what people feel. None of them tell you whether the three agree.
A strong team with a thin product is a different situation from a thin team with a thin product. Same product score, different call to make. One view on its own cannot tell them apart.
Two tell you there is a problem.
Three tell you which problem.