Internal Mobility Playbook for an AI-Changed Role

The decision this guide should improve

A useful guide changes a real decision. For this topic, the decision is whether and how to move toward higher-value work inside the organization before a role boundary hardens. Name who owns that call, the options they can choose, the deadline, and what would make them change course.

Keep observations separate from assumptions. An official source can explain a standard or risk, but it cannot prove what happened in your particular product, portfolio, job, home, or mini-app. Direct evidence needs its own date, subject, method, and limitation.

Use these current primary sources

The workflow below is grounded in ILO: Generative AI and Jobs, 2025 update, O*NET OnLine, U.S. Bureau of Labor Statistics Occupational Outlook Handbook, OECD AI and work. Open the source that supports the exact claim you need, confirm its scope and update date, and record where it does not apply. A smaller claim-to-source map is more useful than a decorative bibliography.

Five-step workflow

1. 1. Map tasks the organization is automating or expanding. Write the evidence, owner, date, and next decision beside the step.
2. 2. Find adjacent teams with recurring unmet work. Write the evidence, owner, date, and next decision beside the step.
3. 3. Document transferable evidence and gaps. Write the evidence, owner, date, and next decision beside the step.
4. 4. Propose a bounded cross-team project. Write the evidence, owner, date, and next decision beside the step.
5. 5. Agree on success and a transition decision date. Write the evidence, owner, date, and next decision beside the step.

Do the steps in order the first time. If a safety, privacy, financial, health, legal, or authorization boundary appears, pause the workflow. More activity is not progress when the prerequisite is missing.

Worked example

An operations specialist sees routine reporting automated while exception analysis grows. They propose a six-week anomaly-review project with the analytics team, create a documented playbook, and use the result to support an internal move.

The useful pattern is the visible chain from context to evidence to decision. Another person should be able to understand what was observed, what remained uncertain, and why the next action was proportionate.

Decision record

Common failure patterns

• Starting with a tool. Start with the user outcome and choose the lightest tool that produces credible evidence.
• Treating exposure as destiny. A risk, score, or capability describes a condition; it does not predict every outcome.
• Moving the threshold afterward. Set success and stop conditions before seeing the result.
• Hiding exceptions. Preserve manual corrections, failures, disagreements, and missing data.
• Reporting activity as impact. A click, test run, generated answer, or submitted form is not automatically the user outcome.

Questions for a second reviewer

1. Which statement is most likely to be wrong?
2. Which evidence is direct, and which is inferred?
3. Does every requested permission or data field support the stated job?
4. Who bears the cost of a false positive or false negative?
5. What would cause us to stop proudly?
6. Can someone reproduce the result from this record?

Printable workbook

1. One-page brief

• Person or team making the decision:
• Current situation and workaround:
• Desired outcome:
• Deadline and consequence of waiting:
• Explicit non-goals:

2. Evidence map

• Claim:
• Primary source and access date:
• Direct observation:
• Limitation:
• What would disprove the claim:

3. Boundary review

• Potential harm and who bears it:
• Preventive control:
• Detection signal:
• Recovery or rollback:
• Independent approval point:

4. Smallest credible test

• Exact subject and environment:
• Baseline:
• One variable:
• Success threshold:
• Stop threshold:
• Owner and date:

5. Decision receipt

• Decision:
• Evidence used:
• Evidence rejected and why:
• Remaining uncertainty:
• Next owner and review date:

Four-week implementation plan

Four scenarios to rehearse

Strong interest, weak commitment

Positive words are not the same as completed behavior. Ask for the smallest real commitment that resembles the desired outcome, and record when it does not happen.

The average improves while one group fails

Review the few user, device, task, risk, household, or market slices that could change safety or trust. Do not let an aggregate hide a concentrated failure.

The workflow succeeds through hidden manual work

Record every correction, review, recovery, and exception. Manual work can be a valuable control, but it changes capacity, cost, privacy, and any claim of automation.

The source is authoritative but has the wrong scope

Authority is not a substitute for fit. Record the population, jurisdiction, product class, reporting period, and exact claim each source supports.

Facilitator prompts

1. What do we know because we observed it?
2. What changed since the source was published?
3. Which user has the least ability to recover from an error?
4. What is the smallest reversible next move?
5. Who can independently challenge the preferred answer?
6. What evidence must exist before scope or spend increases?
7. When will this record become stale?

Sources and review note

• ILO: Generative AI and Jobs, 2025 update
• O*NET OnLine
• U.S. Bureau of Labor Statistics Occupational Outlook Handbook
• OECD AI and work

Source links reviewed 2026-07-13. Follow each publisher for revisions and confirm that the guidance applies to your location and situation.

Related guides

• Your Task-Level AI Exposure Audit
• Build a Skills-Adjacency Map for Your Next Role
• Career Transition Signals to Track Without Doomscrolling

Review appendix

Run a cold review with someone who did not create the first version. Give them the decision, evidence, and boundary without explaining the preferred answer. Ask them to identify ambiguous language, missing stakeholders, unsupported claims, and failure paths. Record the critique beside the revision so the final artifact does not erase how confidence changed.

Treat versioning as part of the evidence. Record the source revision, product or model version, data period, workflow configuration, and publication date. When any of those changes, decide whether the old conclusion still applies. A current-looking page with stale assumptions can be more harmful than an obviously old note.

Plan for recovery before expanding scope. Recovery may mean rolling back a release, revoking a credential, pausing a simulated strategy, asking a qualified professional, redistributing care work, or withdrawing an inaccurate listing. Name the person who can act and confirm that they have the access and information needed.

Measure the user-visible outcome and the cost of achieving it. Include time, review labor, retries, exceptions, fees, support burden, and trust. A result that depends on unrecorded manual rescue is not the same result as one that works repeatedly without intervention.

Publish uncertainty in proportion to consequence. A low-risk reversible experiment can tolerate broader uncertainty than a financial, health, security, privacy, or irreversible decision. When the consequence grows, require stronger evidence, independent review, and a clearer stop condition.

Schedule the next review while context is fresh. Put the owner and date in the record, plus the signals that should trigger an earlier review. If no one owns freshness, readers will eventually mistake an archived answer for a current one.

Build the artifact for handoff, not only for the person doing the work today. Define unfamiliar terms, link raw evidence, state where records live, and explain how to continue if the current owner is unavailable. A durable handoff makes the method useful during an incident, care transition, job change, team change, or product update.

Test the negative path deliberately. Use a missing input, conflicting source, unavailable service, denied permission, failed payment, unexpected market move, changing care need, or inaccessible control. Record whether the workflow fails safely, tells the user what happened, and offers a next step that does not require guessing.

Keep private and sensitive material out of the public artifact. Replace real identifiers with representative examples, minimize what is collected, and preserve protected evidence only in an appropriate access-controlled system. A transparent method does not require exposing the people or credentials involved.

Check incentives as carefully as instructions. Ask who benefits if the answer is accepted, what the tool or publisher is selling, which metric rewards the operator, and whether the recommended action creates a hidden conflict. Document the conflict and add an independent source or reviewer when it could change the decision.

Compare the proposed action with the simplest reasonable alternative, including doing nothing for now. Record the cost, reversibility, likely benefit, and uncertainty of each option. This prevents a sophisticated workflow from winning merely because it is more interesting than a basic fix.

Use thresholds that reflect user harm rather than team convenience. A threshold may cover errors, drawdown, response time, caregiver load, inaccessible tasks, unauthorized actions, or unsupported claims. Explain why the number matters and which person can stop the work when it is crossed.

Preserve dissent. If a reviewer, user, family member, professional, or operator disagrees with the conclusion, record the disagreement, evidence, and unresolved question. Consensus can be useful, but deleting a credible objection creates false confidence and weakens the next review.

Close the loop with the person affected by the decision. Confirm whether the result was understandable, usable, and consistent with their stated goal. Record downstream problems and unintended work. The final measure is not the elegance of the document; it is whether the person can make a safer, clearer next move.

Recheck the artifact on a phone and with basic assistive-technology paths before calling it complete. Tables must reflow, controls need visible labels and focus, links need useful names, and the core action must remain reachable at high zoom. Accessibility failures are workflow failures, not cosmetic defects.

Distinguish absence of evidence from evidence of absence. A test that did not observe a failure may have used the wrong cases, too small a sample, or an environment unlike real use. State coverage explicitly and avoid converting a passing sample into a universal guarantee.

When results are numerical, preserve the denominator and excluded cases. A percentage without the eligible population, time period, missing records, and measurement method cannot support a responsible decision. Show both the count and rate when the size of the sample changes interpretation.

Write the communication plan before the result creates urgency. Name who needs a private update, who needs a public notice, what can be said with confidence, and when the next update will arrive. Clear uncertainty is more trustworthy than premature certainty or silence.

Archive superseded versions with a visible replacement note. Readers following an old link should be able to find the current method and understand what changed. Do not silently rewrite a dated decision in a way that makes the original evidence trail impossible to inspect.

End with one owned next action. It should have a person, date, input, expected proof, and stop condition. If the review produces a dozen unowned recommendations, prioritize the one that most reduces uncertainty or user harm and park the rest visibly.