AI Task Delegation Checklist

Not every task is ready to delegate to an AI agent, and not every task should be automated at the same level of autonomy. This checklist helps you assess a task systematically before you automate it.

The AI Productivity Ladder

Before assessing a specific task, understand which level of AI engagement you are considering. Each level creates more value — and requires more governance.

Step 1 — Assess the Task

• □ Is the task clearly defined?
• □ Is it repeated regularly?
• □ Is the desired output known before the task starts?
• □ Would a different person following the same instructions produce the same result?

Step 2 — Assess the Inputs

• □ Are the inputs consistent between repetitions?
• □ Is the information used in this task approved for AI processing?
• □ Is confidential or personal data controlled appropriately?
• □ Does the supplier's data use policy cover this task?

Step 3 — Assess Decision Requirements

• □ Does the task require professional judgement?
• □ Does it affect another person's legal rights or employment?
• □ Is genuine uncertainty or ambiguity common in this task?
• □ Does the task require tacit knowledge not easily documented?

Step 4 — Assess Actions

Will the AI be able to:

• □ Change a business system or database record?
• □ Send messages externally — email, SMS, API calls to third parties?
• □ Spend money or trigger financial transactions?
• □ Delete data or files?
• □ Change permissions or security settings?
• □ Create accounts or user identities?

Mark each capability as: required · optional · excessive · prohibited.

Step 5 — Assess Consequences

• □ What is the worst realistic outcome if the AI produces an error?
• □ Can the outcome be reversed if the error is discovered immediately?
• □ Can it be reversed if discovered 24 hours later?
• □ What is the reputational consequence if the error reaches a client or regulator?
• □ Is there a regulatory or legal consequence?

Step 6 — Define Controls

• □ Is human approval required before consequential actions?
• □ Are permissions scoped to the minimum required for the task?
• □ Will all agent actions be logged in a form the agent cannot modify?
• □ Are spending limits enforced at the payment system, not just in the prompt?
• □ Is there a tested way to stop the agent quickly?
• □ Are the agent's credentials separate from any human account?

Step 7 — Measure Productivity

• □ How long does the task currently take (total, including all steps)?
• □ How long will checking and correcting AI output take?
• □ How will errors be detected and measured?
• □ How will the productivity gain be confirmed after deployment?

Step 8 — Name a Workflow Owner

• □ Who is accountable for what the agent does?
• □ Who will review performance and correct errors?
• □ Who is authorised to change or shut down the agent?
• □ What is the next review date?

Step 9 — Classify the Task

Warning Signs

Be cautious where:

• The process is not clearly documented before automation begins
• The ROI calculation excludes checking time, failure handling and tool costs
• The agent has been given administrator permissions for convenience
• External actions — emails, transactions, API calls — occur without approval
• There is no named owner for the workflow
• There is no tested way to stop the agent quickly
• The task being automated was already unreliable before AI was applied
• The primary objective is headcount reduction rather than process improvement

Operational Heartbeat

Set a review date at deployment. At each review, check: what is the agent doing, who owns it, what tools and permissions it has, what errors occurred, how many human corrections were required, what it cost and what productivity it actually delivered.