[01]
Current workflow regime
Name the work as it runs today: owners, inputs, decisions, handoffs, tools, and excluded actions.
aequai ~ / approach
$ aequai approach --method
A method built so the work can be reviewed.
AI rarely fails because models are weak. It fails because access outruns approval, ownership is unclear, and nobody can reconstruct what happened. The AequAI method puts the autonomy ladder, the controls, and the evidence trail in place first.
$ aequai loop --evidence-first
The operating line.
Start with one controlled workflow. Map the evidence. Simulate the AI-assisted process. Add human approval gates. Then decide what should be automated, integrated, or handed to a platform partner.
A service starts by naming the workflow, data boundary, owner, review step, source evidence, and safest next action. AI can help collect, structure, compare, draft, and prepare. Humans remain responsible for approval, interpretation, external use, system authority, and claims.
$ aequai method --regime-transition
From AI output to operating regime
AI adoption is not more prompts. It is a controlled transition in how work is represented, reviewed, approved, logged, and changed.
[02]
Typed evidence map
Turn source material, outputs, claims, and review artifacts into named types that can be checked later.
[03]
Operation boundaries
Define what an agent may read, draft, prepare, change, send, escalate, or never touch.
[04]
Simulation and breaker tests
Run synthetic or sanitized cases against failure modes before a workflow touches real authority.
[05]
Gate verdict
Record whether the workflow should proceed, shrink, park, or stop, with the gate that accepted the work.
[06]
Verified workflow transition
Move only the safe next transition forward, with surviving evidence and a named human owner.
$ aequai ladder --levels L0..L3
How much an AI system is allowed to do, one rung at a time.
We do not push every workflow to the top of the ladder. The right rung depends on error cost, data sensitivity, and how reversible an action is.
L0 assistsuggests · human does everything
L1 draftproduces work · human edits & ships
L2 executeacts on systems · each action waits for a human OK
L3 supervisedruns a bounded workflow · human monitors & can stop it
// approval gate sits before L2. Anything past it needs an owner, a log, and a way to roll back. Adoption levels run L0–L4; the website default is assist, draft, and prepare-with-approval before any real authority.
$ aequai adoption-curve --self-assess
Productivity usually dips before it compounds.
Most teams hit a friction valley first: setup, learning, and rework cost more than they return. Governed workflows are how you cross the dip instead of stalling in it. Place yourself on the curve, then read the next safe step.
// boundary: This result is a directional self-assessment. It is not legal advice, compliance approval, security assurance, public proof, or a production-readiness conclusion.
$ aequai controls --scorecard
Six controls turn "AI is being used" into "AI is being operated".
We assess each one, mark the gaps, and design the missing pieces. These travel with every engagement.
[01]
Ownership
A named human owns each AI workflow and its outcomes. No anonymous automation.
[02]
Approval gates
High-impact actions pause for a human decision before they touch a real system.
[03]
Access scope
Agents get the narrowest credentials that still let them work, staging before production.
[04]
Logging
Actions, inputs, and decisions are recorded so the work can be reconstructed later.
[05]
Rollback
Every execute-level action has a defined way to undo or contain it.
[06]
Evidence
You leave with artifacts a reviewer, internal or external, can actually inspect.
$ aequai engagement --phases 4
Four phases, each with a defined output.
phase 01
Diagnose
Map current AI use, workflows, data boundaries, and the real risk surface. Output: an AI use & risk map and a quick-win shortlist.
phase 02
Architect
Design the department adoption map, agent permissions, approval gates, and governance layer. Output: an operating-model design and responsibility matrix.
phase 03
Pilot
Scope one small, reviewable pilot with success metrics and a definition of done. Output: a pilot brief and the controls it runs under.
phase 04
Operate
Hand over the operating model and the readiness decision where it fits. Output: an implementation-partner brief and evidence pack.
6
core controls assessed every time
4
autonomy levels, gated
0
compliance or legal claims made
1
owner per workflow, always
$ aequai decide --verdict
Every engagement ends with a decision.
No open-ended retainers by default. Each service closes with one honest verdict and the gates that go with it.
proceed
Safe enough for a controlled next step with named gates.
shrink
Useful, but the scope must narrow before pilot work.
park
Potentially useful later, timing, owner, evidence, or advisor is not ready.
kill
Unsafe, unsupported, out of scope, or not aligned with AequAI boundaries.
$ aequai contact --map workflow
See the method on your own workflows.
Bring your messiest AI workflow. We will map it onto the ladder, mark the missing controls, and name the smallest safe next step.