Decide whether AI should proceed, wait or stay human
Are the conditions strong enough to expand AI or agentic capability? Establish what must be repaired, what should remain human and what evidence would justify a larger role.
AISEF-X · A netlinkE decision framework
Inspect the evaluation gatesDefine the boundary before expanding capability.
AISEF-X connects readiness, architecture, authority, governance and implementation around the proposed job.
AISEF-X is a netlinkE decision framework, not a certification; organisational accountability remains with the organisation.
It supports a context-specific operating decision. It is not an external standard, legal-compliance assessment or guarantee of safety. Technical feasibility alone does not establish permission to proceed.
Inspect the conditions around the decision.
Read the gates together. Supporting evidence for one condition does not cancel a weakness in another. These are evidence requirements; no organisation has been evaluated on this page.
For each evidence trail, retain the source of the operating fact, record owner, observation or review date, permitted use and unresolved contradictions. An unavailable record remains unresolved.
-
Gate 2
Information and evidence provenance
Can the inputs be traced, accessed legitimately and trusted for this job?
- Evidence that supports this gate
- Identifiable source records, their owners and review dates, permitted access and documented gaps.
- Evidence that weakens this gate
- Untraceable claims, stale or conflicting records, missing context or access the proposed capability should not have.
- Implication for scope
- Repair the information path. Wait where the required evidence cannot be established.
- Related decision outputs
- Repair conditions first; Wait, do not build
- Connection to the next inspection
- The workflow must handle missing or conflicting information.
-
Gate 3
Workflow and exception handling
What happens when the work leaves its expected path?
- Evidence that supports this gate
- A documented handoff sequence with owners, exception examples and a usable escalation path.
- Evidence that weakens this gate
- A happy-path demonstration that depends on unrecorded manual rescue when an exception occurs.
- Implication for scope
- Repair handoffs and exceptions before automation increases their volume.
- Related decision outputs
- Repair conditions first
- Connection to the next inspection
- Capability must stay within the workflow and its exception boundaries.
-
Gate 4
Capability proportion
What is the smallest useful capability for the consequence at stake?
- Evidence that supports this gate
- A defined operating job, bounded tools and permissions, and an explicit comparison with assisted or human work.
- Evidence that weakens this gate
- Broad autonomy without a defined constraint, unnecessary permissions or a tool-led use case.
- Implication for scope
- Wait when no useful bounded job is established. Retain human decisions where the consequence requires judgement.
- Related decision outputs
- Wait, do not build; Remain human
- Connection to the next inspection
- The chosen scope determines what must be stoppable and recoverable.
-
Gate 5
Recovery and stop conditions
Can the capability be stopped and its effects recovered or reconciled?
- Evidence that supports this gate
- A named intervention owner, tested stop and recovery paths, and explicit conditions for suspension.
- Evidence that weakens this gate
- Unrecoverable side effects, an untested rollback assumption or no owner for reconciliation.
- Implication for scope
- Repair recovery before proceeding; keep the work human when its effects cannot be acceptably controlled.
- Related decision outputs
- Repair conditions first; Remain human
- Connection to the next inspection
- Expansion evidence must include failures, intervention and recovery.
-
Gate 6
Measurement and expansion evidence
What observed result would justify a broader scope?
- Evidence that supports this gate
- A defined outcome, attributable operating records, reviewed exceptions and evidence that controls work in practice.
- Evidence that weakens this gate
- Activity totals without outcomes, a successful demo without operating evidence or expansion justified only by time invested.
- Implication for scope
- Proceed only within the evidenced boundary when the other gates support it. Otherwise hold scope and repair or wait.
- Related decision outputs
- Proceed; Repair conditions first; Wait, do not build
- Connection to the next inspection
- New operating evidence returns to ownership and the full register.
Resolve the next decision, with its conditions attached.
These are possible conclusions for accountable review, not calculated results. Different parts of a job may require different boundaries.
- Proceed
- Introduce only the bounded capability supported by all six gates. Record the owner, permitted scope, stop conditions and evidence needed before expansion.
- Repair conditions first
- Resolve named weaknesses in ownership, information, workflow, recovery or measurement. Assign each repair an owner and evidence to re-inspect before proceeding.
- Wait, do not build
- Hold implementation when the operating job, capability proportion or required evidence remains unresolved. State what new information would make reconsideration useful.
- Remain human
- Retain consequential judgement where authority, uncertainty or recovery limits do not support delegation. Any assistance must remain inside that human decision boundary.
Keep uncertainty in the decision record.
Evidence can be incomplete, outdated or specific to one operating context. A pilot does not establish that a larger capability will behave the same way. A framework conclusion does not replace specialist legal, security or domain validation where required.
Record unresolved conditions, contradictory evidence, the accountable decision owner and the next review trigger. Missing evidence does not count as support. Organisational accountability remains with the organisation.
- Unresolved owner or permission boundary: hold delegation.
- Untested recovery or stop conditions: hold expansion.
- Insufficient outcome evidence: retain the current boundary and re-inspect.
Let operating evidence change the answer.
- Operate within the agreed boundary. Keep human authority and stop conditions available.
- Record evidence. Capture outcomes, exceptions, interventions and recovery with their sources and owners.
- Re-inspect the gates. Compare observations with the conditions behind the earlier conclusion.
- Revise the output. Continue, repair, suspend or return work to people, then operate within the revised boundary.
Observed evidence can reverse or narrow a prior conclusion. Expansion is earned through evidence and recoverability, never assumed from time or investment.