These questions can help a team examine how its work is changing. A recent piece of work usually makes the discussion more concrete than a general statement about AI readiness. Different answers may be appropriate for different services or kinds of change; there's no score or expectation that every activity becomes autonomous.
Choosing and directing the work
- Can the people and agents involved find the intended outcome and explain the reason for the work?
- Does the agent have enough context to make progress without someone repeatedly filling in missing information?
- Can checks assess the result without a person having to inspect every detail?
- If the agent makes a mistake, will those checks catch it before it affects other work, or does that still depend on a person noticing?
- Are important decisions available beyond the conversation in which they were made?
- When a plan changes, how does that change reach work already in progress?
Choosing where AI can improve delivery helps identify a starting opportunity. Keeping plans connected to the work and making context available address the information participants need as work proceeds.
Connected work
- Can the team distinguish work that can proceed independently from work that depends on another decision or change?
- Is someone responsible for the next decision when an agent finishes or asks for help?
- When work starts in a terminal or another tool, can teammates find its purpose, changes, and check results where they normally review work?
- Can the team see duplicate or overlapping work before it creates conflicting changes?
- Does the rate of starting work fit the capacity to check, review, and deliver it?
- If an attempt stops after making changes, can the next participant find out what happened and recover without repeating those changes?
Coordinating human and agent work across delivery develops ways to address these questions.
Quality and continued ownership
- Do the checks assess the intended behavior independently of the agent's own account of success?
- Can someone trace an action back to the request, the access used, and any approval?
- Are important restrictions enforced by the systems that control the action?
- Could the person or agent making a change also weaken the checks that decide whether to accept it?
- If an agent follows a malicious instruction, what prevents it from reaching unrelated files, credentials, or services?
- Can the team connect what was released to the exact artifact that passed its checks?
- Who can stop the work, withdraw access, and account for tasks already running elsewhere?
- What would cause the team to reduce a workflow's autonomy, and what would need to change before restoring it?
- Do review findings and production experience improve subsequent instructions, evaluations, or checks?
- Who maintains the context, integrations, and capabilities the team depends on?
- Do people have time and support to learn how to assess agent work, challenge it, and recover from mistakes?
Choosing and revisiting workflow autonomy helps decide how much work to delegate. Building feedback into delivery explains how to choose evaluations and checks. Operating the team's AI capabilities carries that work into updates, support, and recovery.
Results and investment
- What improvement would justify the investment, and what has changed compared with the team's starting point?
- Does the comparison include review, correction, unsuccessful attempts, and unfinished work?
- Are time saved, additional capacity, and financial savings distinguished rather than counted as the same benefit?
- Could differences in workload or another process change explain the result?
- Is the organization using any released capacity, and are quality and customer outcomes improving alongside delivery?
- Who can change or stop the approach if its costs outweigh the benefit?
Evaluating business value and the cost of delivery explains how to connect those answers to an investment decision.