MEMO · TO anyone scoping an agentic AI deployment · RE one candidate task, five criteria
Free tool · TRACE framework
Is this task ready for an AI agent?
Evaluate one candidate task against the five TRACE criteria, Traceability, Reversibility, Acceptance criteria, Compliance and Escalation, before any vendor selection or architecture decision begins. Each criterion is a gate rather than a weight, so a strong average does not pass a task with an untested override. No email required, and nothing you enter leaves your browser.
The five criteria
What each one tests, and what goes wrong when it is missing
- T
Traceability of inputs and outputs
Every input consumed and every output produced can be logged, attributed and audited against its source.
When this fails. Any erroneous output becomes unattributable. You cannot tell what caused it, cannot correct the root issue, and cannot demonstrate due diligence to anyone who asks.
- R
Reversibility and risk tier
The consequence of an error determines how much oversight is required before any autonomy is granted.
When this fails. Irreversible errors occur without human review. A single mistaken output in a high-stakes domain can generate costs or harms that no amount of subsequent correction undoes.
- A
Acceptance criteria definition
Success is measurable against something defined before deployment and independent of what the agent itself produces.
When this fails. You cannot distinguish a performing agent from a malfunctioning one. The first indication of systematic error is typically a downstream complaint rather than a monitoring signal.
- C
Compliance and regulatory mapping
The regulatory regimes and internal approval processes that apply are identified before architecture decisions are made.
When this fails. The deployment breaches a requirement nobody identified until after go-live. Remediation then means suspension, redesign, or both, at a point where the design is already committed.
- E
Escalation and override pathway
A human can intervene in, override or halt the agent mid-task, and that pathway has actually been tested.
When this fails. When the agent behaves unexpectedly, no tested mechanism exists to stop it. The time spent building a manual stop while the system is running is the time the damage accumulates in.
Evaluate
Score one task at a time
TRACE evaluates a single candidate task. Running it across a portfolio means running it once per task, which is the point: the answer differs task by task inside the same organisation.
Result
- T Traceability of inputs and outputs ·
- R Reversibility and risk tier ·
- A Acceptance criteria definition ·
- C Compliance and regulatory mapping ·
- E Escalation and override pathway ·
A self-assessment for your own scoping. It is not an assurance opinion, a compliance review, or legal advice. TRACE evaluates whether a task is structurally suitable for autonomous execution; it does not evaluate a vendor, a model, or an implementation.
