Platform · Agents
An agent earns the right to act by proving it understands the work.
Before any agent acts, it apprentices: it predicts each step of a process and is scored against reality, with its confidence and its errors visible. Only after graduating, with a human signing the certificate, does it receive scoped, revocable authority. Never standing credentials. Never action on unverified facts.
How it works
Agent Apprenticeship, step by step.
Observe
The agent studies the documented process, the governed map, not the employee's screen.
Predict
It predicts each step before it happens and is scored against what actually happens.
Graduate
When its predictions hold, a human owner signs off. Authority is scoped to specific actions, time-bound, and revocable in one click.
Act, with a leash
Every action is logged against the claims that authorized it. Unverified fact, no action.
Evidence tiers
Where each claim stands.
Verified
Agents act only on verified claims, T1 declarations never authorize action.
Documented
The apprenticeship record and graduation certificate are governed artifacts.
Stated limits
What it will not do.
Will not watch screens or log keystrokes, apprenticeship runs on the documented process, not surveillance.
Will not receive standing credentials, authority is scoped, time-bound, revocable.
Will not act on unverified facts.
Will not make employment decisions of any kind.
Features
0 available · 1 in preview · 3 on the roadmap.
| Feature | Status |
|---|---|
| Apprenticeship scoring loop | Design-partner preview |
| Graduation certificates with human sign-off | Roadmap · Q2 2027 |
| Scoped, revocable authority grants | Roadmap · Q2 2027 |
| Action ledger tied to authorizing claims | Roadmap · Q3 2027 |
Questions we actually get
Asked by real skeptics.
Why apprenticeship instead of just configuring an agent?
When is this available?
Related
See this pillar inside a real diagnostic.
The two-week engagement uses every pillar above, and hands you the evidence-linked output.