In brief

Treat the checker’s high-pressure flag as a prompt to inspect one real workflow, not as a forecast about your employment. Record the task’s frequency, exceptions, review burden, and accountability, then check whether an approved tool is used and whether responsibilities or workload have changed. If the wider role remains valuable and workable, investigate redesign first; consider a broader move when local evidence or your practical needs support it.

What should you do first when one task is flagged?

Do not treat a high-pressure task flag as a forecast that you will lose your job. Treat it as a reason to inspect one piece of work. If the wider role still provides useful work and meets your needs, record what the flagged task involves before considering a costly course or career pivot. If the role is already shrinking or no longer workable for your health, income, or family responsibilities, those facts may justify preparing a broader move; the score does not cancel them. Start with a workflow record: how often the task occurs, what information goes in, what exceptions make it harder, what an error could affect, how much review it takes, and who remains accountable. Imagine a support worker whose checker flags drafting routine replies. Ask whether the work also includes unusual cases, checking account details, resolving complaints, and owning the final response. This is a practical investigation sequence, not a validated assessment. The International Labour Organization's *Generative AI and Jobs: A Refined Global Index of Occupational Exposure* combines task-level data, worker assessments, expert input, and model predictions to estimate occupational exposure. The report says job transformation is more likely than job disappearance because most occupations include tasks requiring human input. That global finding explains why a task signal deserves attention, while leaving an individual workplace outcome unresolved.

Sources: Generative AI and Jobs: A Refined Global Index of Occupational Exposure

What does a high-pressure flag establish—and what does it leave unknown?

A high-pressure flag establishes that the named task deserves examination under the checker’s criteria. It does not establish that an AI system can reliably complete the workflow, that your employer has adopted it, or that your job will disappear. Treat the result as a screening signal: useful for choosing what to inspect, too limited to decide your employment future. Check which task description and assumptions produced it, then compare them with your actual work.

Several different questions sit behind a single flag. Capability asks whether a system can assist with or perform a defined task under particular conditions. Exposure estimates how closely tasks match capabilities or use patterns. Observed use records interactions with a tool, but not necessarily in your workplace. Employer adoption means an organization has brought a tool into a recurring process. Task redesign concerns changes to responsibilities, handoffs, review, or workload. Labor demand concerns hiring across a market; displacement concerns workers losing roles. Evidence for one question does not automatically answer the next: a demonstration can show capability without proving routine use, and exposure does not measure a person’s chance of losing a job.

The 2025 ILO working paper, “Generative AI and Jobs: A Refined Global Index of Occupational Exposure,” illustrates what an exposure estimate can contain. Researchers collected 52,558 data points on perceived automation potential for 2,861 tasks, drawing on a representative sample of 29,753 tasks in Poland’s occupational classification and a survey of 1,640 employed people across broad occupational groups. They compared assessments with international expert input and used model predictions to extend task scores across the classification. The result is a four-gradient index of occupational exposure across countries and sectors. The ILO says job transformation is more likely than outright disappearance because most occupations include tasks requiring human input. This is a global, modeled assessment, not observation of your employer’s decisions or a personal forecast; its methods do not establish whether a specific workplace adopted a tool.

The U.S. Bureau of Labor Statistics makes a related distinction in “Artificial Intelligence (AI) exposure categories.” For its 2025–35 projections material, BLS combines three theoretical exposure measures with two measures based on observed AI use mapped to occupational tasks or work activities. Theoretical exposure asks whether AI could assist with or complete some work; current-evidence measures summarize observed interactions mapped to tasks. BLS says these measures do not directly observe whether workers in an occupation used AI on the job. Exposure is not a forecast of job loss, productivity gains, automation probability, or wages. The categories compare occupations; they do not show that an individual employer approved a tool or reassigned work.

A score compresses a task description and a method. It cannot settle whether the label fits your inputs, exceptions, quality or accountability. A high category does not mean an absolute level of pressure: BLS compares occupations using selected data. Its measures combine capability ratings and observed interactions. Exposure identifies a question for investigation, not a conclusion about adoption or staffing. Read the flag alongside its task definition and limits, then seek local evidence about whether work has changed.

Sources: Generative AI and Jobs: A Refined Global Index of Occupational Exposure; Artificial Intelligence (AI) exposure categories

Why can one task be exposed while a role still has useful work?

A task can be technically exposed while the surrounding role retains exception handling, domain judgment, relationship work, or accountability. A job is a bundle of activities, not one output. A system may draft a routine response or summarize familiar material more readily than resolve an unusual case with incomplete context. Someone may need to check facts, interpret policy, explain a decision, handle an exception, and accept responsibility. These duties do not make a role immune; they change which parts of the workflow are candidates for assistance and what work remains.

A high-pressure flag tells you to inspect the task, not treat its label as a verdict on the occupation. Ask how much of your week it takes, whether inputs are predictable, how often cases depart from the template, what mistakes cost, and who reviews the result. Removing minutes from a small task may leave substantial work intact. If the task is central and costly to review, its change may matter more. These are questions about actual job design; a checker cannot answer them from an occupational description alone.

Labor projections reinforce this distinction but offer no personal guarantee. In “AI impacts in BLS employment projections,” the U.S. Bureau of Labor Statistics discussed occupations potentially susceptible during its 2023–33 projection period. It projected lawyer employment to grow 5.2% and paralegal and legal assistant employment 1.2%. These different paths among occupations that can include document review do not show AI caused either trajectory or establish what happened to an incumbent. They are U.S. occupation-level projections from that cycle, not a local forecast for readers elsewhere.

Demand can shift separately from task exposure. The OECD brief “How is AI changing the way workers perform their jobs and the skills they require?” analyzed online vacancy data across ten countries, comparing country-specific base years with 2021–22. It found skill mentions changed in highly AI-exposed occupations, while management and business skills remained highly demanded; declines in some skill demand at more exposed workplaces were relatively small. Vacancy wording is indirect: it records employer mentions in postings, not each worker’s duties, hiring outcomes, or AI’s causal effect. It complicates a replacement story but cannot establish that a particular role is stable.

A field study shows why workers can experience the same tool differently. “Generative AI at Work” examined staggered access to a conversational assistant among 5,172 customer-support agents at one business-process software company. The tool suggested replies while agents remained responsible for conversations. Researchers reported 15% more issues resolved per hour on average, with larger gains for less experienced and lower-skilled agents. For the most experienced and highest-skilled agents, they found small speed gains alongside small quality declines. One company and support workflow cannot forecast other occupations or staffing effects. It shows an average result can conceal differences by experience and task.

Together, these findings support a narrower conclusion: exposure can coexist with useful work, while the mix and value of that work may change. Expertise may help someone spot a flawed answer or judge an exception, but the study does not establish that experienced workers are protected; its quality result is a reason to measure outcomes. Projected occupational growth likewise does not prove an employer will retain every task or role. Map routine outputs, exceptions, verification, relationships, decisions, and accountability, then look for evidence that those responsibilities are shifting at your workplace. The task score starts the inquiry; the work around the task determines what it means for you.

Sources: AI impacts in BLS employment projections; How is AI changing the way workers perform their jobs and the skills they require?; Generative AI at Work

How do you tell tool capability from a change in your workplace?

A demonstration answers a narrow question: can a tool produce a plausible result for a task under conditions? It does not answer whether your employer has approved it, whether colleagues use it repeatedly, or whether it has changed anyone’s responsibilities. To tell capability from workplace change, look for an authorized workflow that recurs and for observable changes in task volume, handoffs, review ownership, or staffing. A demo or your own experiment is a reason to ask better questions, not proof that the role has been redesigned. The distinction matters because early use can be deliberately peripheral. In the OECD’s 2024 survey of small and medium-sized enterprises using generative AI, respondents said it was used more for simple, one-off, less important tasks than for complex, recurring, important ones. Only 29% of these AI-using SMEs reported use in their company’s core activities. Many used it across a mixture of task types. This is evidence about surveyed SMEs and their reported use, not a forecast for larger employers, every country, or your organization. Still, it shows why a tool’s ability to draft a familiar document does not establish that a central work process has been handed over to it. Moving from an impressive output to routine work involves practical gates. Someone must permit the use; the data must be available under acceptable rules; the result must be dependable enough for the task; and the tool must fit existing systems and handoffs. Someone also needs time to review errors, authority to act on the output, and responsibility when it is wrong. These are questions to investigate locally, not a universal checklist proven to block adoption in every workplace. A polished answer on a public demonstration may rely on clean inputs and omit the corrections, unusual cases, and sign-off that shape the real workflow. Start with evidence you can observe or ask about. Is there written guidance naming an approved tool and the information that may be entered? Is a manager or process owner responsible for the workflow? Has the task been repeated, rather than tried once? Do colleagues now pass different materials to one another, or does the same person still check and approve each result? Are fewer requests arriving, or has only the first draft become quicker? Has anyone changed the responsibilities, workload expectations, or staffing plan? Record what you know, what you heard, and what remains unconfirmed; a rumor and a documented change are not equal signals. The OECD report *Using AI in the Workplace* summarizes worker and employer surveys and discusses both reported benefits and concerns in surveyed settings. Such reports can help identify questions about performance, work experience, oversight, and implementation. They cannot establish that a particular task in your workplace has been adopted, nor that reported experience caused a staffing decision. A worker saying a tool helps with a task is evidence of reported use or experience; an employer changing a process is a separate observation. Neither should be silently converted into a prediction about your job. If you cannot identify approval, repeat use, a process owner, and a concrete change in duties or workload, treat the checker flag as a prompt to learn, not a workplace verdict. Ask a question about policy and ownership, and watch whether the workflow changes over time. Do not test with confidential information in an unapproved service. If your employer authorizes a trial, clarify permitted data, review responsibility, quality expectations, and a stop condition before using it. This keeps your next step tied to evidence while protecting both the work and the people whose information it may contain.

Sources: Generative AI and the SME workforce; Using AI in the Workplace

What should a safe task test measure beyond speed?

If policy and data rules allow a bounded trial, compare the whole task cycle rather than asking only how quickly a tool produces a first answer. Count completion time, correction and review time, exception handling, error severity, output quality, handoffs, and who remains accountable. Before starting, define what would count as useful improvement and what result would make you stop. A faster draft is not a successful redesign if checking it consumes the saved time, errors become harder to catch, or responsibility becomes unclear. Start with a baseline from ordinary work, not a memorable easy case. Note the task’s usual inputs, steps, and completion time, including checking, correction, missing information, and handoffs. If the organization permits it, select a small set of representative, low-risk cases. Use an approved tool and data, and keep the comparison fair: similar cases, the same definition of a finished task, and consistent quality criteria. Never put confidential or personal information into an unapproved service. If permissions or data protections are unclear, ask the process owner before experimenting. Agree on review ownership. The person checking should know what can go wrong, how to verify key facts, and when to reject or escalate an output. Record exceptions as well as routine completions: cases needing extra context, a human decision, customer clarification, or correction may be where apparent time savings disappear. Track whether review shifts to another colleague or a faster step creates work later. The study “Generative AI at Work” helps explain why speed alone is insufficient. It examined the staggered introduction of a conversational assistant using data from 5,172 customer-support agents. The Stanford Digital Economy Lab’s research record reports a 15% average increase in issues resolved per hour, but substantial differences by experience: less experienced and lower-skilled agents improved speed and quality, while the most experienced and highest-skilled agents saw small speed gains alongside small quality declines. This was one company’s customer-support deployment, not a forecast for another role, and its average cannot tell a worker how their cases will behave. It does show how a single time-saved figure can conceal different quality effects. The OECD’s “Unlocking productivity with generative AI: Evidence from experimental studies” synthesizes bounded experiments in which outcomes depend on task fit, worker skills, and the ability to evaluate output. The OECD report “Using AI in the Workplace,” based on worker and employer surveys in manufacturing and finance across seven countries, describes reported benefits alongside concerns such as work intensity and data collection. The survey accounts are not causal estimates of productivity or job effects. A practical trial sheet can capture four things: the baseline and cases; permitted tool and data conditions; time, quality, error, exception, review, and handoff observations; and a decision rule agreed in advance. Widen a trial only if it meets the same quality threshold without unacceptable errors, hidden review burden, or unclear accountability. Stop or redesign it if those conditions fail. These are proposed safeguards for a local decision, not a validated assessment method. A small approved test can show whether a particular workflow merits redesign, training, or discussion with a manager. It cannot establish organization-wide adoption, a staffing change, or job security. Treat the record as evidence about one task under specific conditions, then use it to ask what work remains and what skills are needed to review it well.

Sources: Generative AI at Work; Unlocking productivity with generative AI: Evidence from experimental studies; Using AI in the Workplace

What local evidence should move you from redesign toward a career change?

A high-pressure flag becomes a reason to prepare for a wider move when it is joined by repeated, observable changes in the role: a core responsibility is removed or reassigned across work cycles; an approved system becomes part of the routine process; workload, handoffs, review ownership, or staffing changes are documented; or new requirements exceed a learning step you can reasonably take. These are decision triggers, not proof of a coming layoff. They suggest the job design deserves attention while you still have options. Separate that evidence from signals that invite investigation but cannot carry the decision by themselves. A checker score describes task-level change pressure under its stated assumptions. A polished demonstration shows that a capability may exist. A colleague’s report, a rumor about a planned rollout, a single vacancy, or a broad occupational projection may prompt a question, but none establishes that your employer adopted a system, removed work, or intends to reduce a role. The International Labour Organization’s “Generative AI and Jobs: A Refined Global Index of Occupational Exposure” combines task-level assessments, expert input, and model predictions to estimate exposure across occupations. Its global measure describes where work may be transformed; it does not observe your team’s implementation or predict your employment outcome. Use the flag to identify what to inspect, then look at your workflow. A practical threshold is a pattern, not one announcement. Note whether the same responsibility has moved away from people repeatedly, whether a tool is approved and used in ordinary cases, whether remaining work has changed in volume or difficulty, and who checks and signs off. Ask whether the organization is changing role descriptions or training people for a revised process. A formal process change or recurring reassignment is more informative than a demo or speculation. Even stronger evidence describes a local change, not its eventual effect on headcount. Record what happened and when, and set a review point after another work cycle or formal decision. There is an alternative to assuming that automating one task means the whole role is being removed. The OECD report “The Impact of AI on the Workplace: Evidence from OECD Case Studies of AI Implementation” examined nearly 100 qualitative case studies in finance and manufacturing across eight countries. The cases showed different mechanisms: AI complemented some work, automated portions of other tasks, and sometimes shifted workers toward surrounding responsibilities. The report found reorganization more prevalent than displacement in the cases studied; it also documented exceptions, including reduced employment in some settings through slower hiring or attrition. These findings help explain ways a workflow can change, not how often each outcome occurs across employers. The cases are not a representative forecast and do not establish what your organization will choose. Do not wait for an AI signal if the role already fails a practical test. A salary that no longer covers essential costs, an unsustainable schedule, health limits, caregiving responsibilities, location constraints, or a role contracting before a tool appeared can make a broader move reasonable now. Those conditions are valid reasons in their own right. Where possible, compare options while preserving income: identify adjacent responsibilities that use your domain knowledge, check local requirements and openings, and estimate training time and cost before committing. If evidence is mixed, begin a reversible preparation step and decide what change would justify a larger one. The decision rule is simple: investigate a lone flag; prepare when local changes form a pattern; change sooner when the work no longer fits your constraints.

Sources: The Impact of AI on the Workplace: Evidence from OECD Case Studies of AI Implementation; Generative AI and Jobs: A Refined Global Index of Occupational Exposure

Which move fits: redesign, adjacent work, or a larger change?

The right move depends on what is changing around the flagged task and what you can afford to risk. Redesign is the proportionate first option when the broader role still contains valuable responsibilities, the workflow can be examined under workplace rules, and no stronger local evidence says the role is contracting. An adjacent move makes sense when it carries forward domain knowledge while shifting the mix of duties. A larger career change deserves priority when the role is already losing core work, cannot meet your income or health needs, or no longer fits family or location constraints. Compare the options as decisions with different costs and reversibility, not as a ranking of occupations that are supposedly safe. Redesign is often the easiest to reverse: you can learn where a tool helps, where it fails, and which responsibilities still require your judgment. It usually asks for the least immediate disruption, but can leave you exposed if the employer is already consolidating duties. Adjacent movement may protect more income continuity than a full reset when the experience transfers, though the fit must be checked against real openings. A larger change can address a poor fit directly, while demanding the most planning around savings, training, and transition time. It may preserve income and expertise, but employer controls, limited approved tools, or changed expectations about volume can still make it risky. A better task metric does not prove that the organization will retain the same role or headcount. An adjacent move can preserve more of what you already know than starting over. A person who understands a regulated process might investigate whether local roles need exception review, output checking, client explanations, or handoff coordination. These are directions to investigate, not promised openings. Ask whether actual postings or conversations with people doing the work show these duties, what credentials they require, and whether the salary range, location, schedule, and workload are workable. Domain knowledge may transfer; a specific tool, data, communication, or process skill may be the gap. Identify it before purchasing training. A larger change may be rational even if the checker flagged only one task. Perhaps several important responsibilities have already been reassigned, or the remaining role has become unsustainable because of pay, health, caregiving, schedule, or location. Those reasons count independently of AI. The reverse also holds: a worrying score is weak grounds for abandoning a role that still meets your needs when there is no observed change beyond the task signal. Compare your salary floor, benefits, location, training time, and tolerance for uncertainty. An attractive path can still fail your constraints. Evidence about AI-supported work also argues against assuming that one intervention benefits every worker equally. In the peer-reviewed field study “Generative AI at Work,” a conversational assistant was introduced to 5,172 customer-support agents at one company. Issues resolved per hour rose by 15% on average, but effects varied: less experienced and lower-performing workers gained more, while the most experienced and highest-skilled workers saw small speed gains alongside small declines in measured quality. This supports testing a tool on a task; it cannot predict adoption, staffing, or how your skills will be valued. It concerns one assistant in one support setting, and productivity is not employment. The OECD report “Using AI in the Workplace” draws on surveys of workers and firms in manufacturing and finance across seven countries. It describes workers reporting benefits such as improved performance and enjoyment, alongside concerns about work intensity, data collection, and inequality. These are reports from selected sectors and countries, not causal proof or a portrait of all knowledge workers. Together, they suggest assessing work quality and distribution as well as speed. Ask what support, review ownership, and expectations accompany approved learning. Do not do unauthorized work to demonstrate enthusiasm. Training should answer a verified requirement. If your current organization has an approved tool and you need to use it in an existing field, a short course or guided practice may be enough to learn the workflow, data safeguards, and output checks. A project using permitted, nonconfidential material can show that you can frame a task, assess an answer, and document corrections. Self-study can fill a narrow gap cheaply, though feedback and hiring signal may be limited. A certificate is relevant when local employers request it; it does not itself establish hiring value. Consider a degree when the target role requires its depth or credential. For technical transitions, check prerequisites and curriculum before enrolling. No route guarantees employment. A practical comparison can fit on one page. For each path, record: which responsibilities it preserves or changes; the evidence that those duties exist locally; entry requirements; likely cost and time; expected income continuity; and effects on location, health, and family. Mark unknowns plainly. For redesign, the next evidence might be approval to test one low-risk workflow with clear quality review. For an adjacent move, it might be several current postings that match your experience and constraints. For a larger change, it might be confirmation that the role is contracting or that another field better meets a non-AI need. These are inputs, not universal thresholds. Choose redesign when the role remains workable and a safe, bounded investigation is possible. Choose adjacent work when local evidence supports a role that uses your experience while changing the task mix. Prioritize a larger change when the current role is already contracting or fails practical needs that matter now. If two paths remain plausible, preserve options: map the task, verify workplace policy, and investigate one adjacent role before committing to expensive training or resigning. Revisit the choice when responsibilities, requirements, or constraints actually change. That keeps the response proportionate without mistaking a task-pressure signal for either reassurance or a prediction.

Sources: Generative AI at Work; How is AI changing the way workers perform their jobs and the skills they require?; Using AI in the Workplace

What is your next action this week?

This week, write down one recurring task the checker flagged: who reviews the result, and what happens if it is wrong. Then check your workplace’s rules and whether an approved tool is already used in that workflow. Do not put confidential material into an unapproved service. The checker identifies a task to examine; it cannot establish employer adoption or predict your job outcome. The ILO’s Generative AI and Jobs: A refined global index of occupational exposure frames exposure around potential task change, not a forecast that a particular worker will lose a job.

If policy permits, test only a small, low-risk part of the workflow and record the full cycle: time to produce, time to check and correct, exceptions, quality,. Set a review date, or name the concrete change that would prompt an earlier review, such as repeated reassignment of core duties or a documented shift in workload. If useful work remains and your constraints favor continuity, investigate redesign first. Compare an adjacent move or larger change against local requirements, income needs, location, training time, health, and family responsibilities when the evidence or your circumstances call for it. The free checker offers task-level change-pressure signals, not validated displacement odds. If you need a comparison of several paths, the paid roadmap considers your constraints; it does not guarantee employment or income.

Sources: Generative AI and Jobs: A Refined Global Index of Occupational Exposure

Questions readers ask

Does a high-pressure checker result mean my job is at risk?

No. It is a task-level change-pressure signal, not a validated probability of displacement. It does not establish reliable workflow performance, employer adoption, changed staffing, or an individual outcome.

When should I consider a move beyond redesigning the task?

Consider preparing for a wider move when core responsibilities are repeatedly removed or reassigned, workload or staffing changes are documented, requirements exceed a manageable learning step, or the role no longer meets your practical needs.

Sources and notes

  1. Generative AI and Jobs: A Refined Global Index of Occupational Exposure

    Supports the distinction between modeled occupational task exposure and an individual employment forecast.

  2. Artificial Intelligence (AI) exposure categories

    Supports the distinction between theoretical occupational exposure measures and observed AI use.

  3. AI impacts in BLS employment projections

    Supports the point that projected outcomes differed among potentially susceptible occupations and do not isolate AI as the cause.

  4. How is AI changing the way workers perform their jobs and the skills they require?

    Supports the account of changing skill mentions in online vacancies and the limits of vacancy data as evidence about an individual worker.

  5. Generative AI at Work

    Supports the reported average productivity result and differences by agent experience in one customer-support field study.

  6. Generative AI and the SME workforce

    Supports the survey finding that use among surveyed AI-using SMEs was more common for simple tasks than core activities.

  7. Using AI in the Workplace

    Supports discussion of reported workplace benefits and concerns in surveyed manufacturing and finance settings.

  8. Unlocking productivity with generative AI: Evidence from experimental studies

    Supports the synthesis that experimental productivity results depend on task fit, worker skills, and ability to evaluate output.

  9. The Impact of AI on the Workplace: Evidence from OECD Case Studies of AI Implementation

    Supports examples of different workplace implementation mechanisms, while the case studies do not establish how common each outcome is.

  10. How is AI changing the way workers perform their jobs and the skills they require?

    Supports comparison of adjacent career paths against changing skill demand in online vacancies, while vacancy data does not establish an individual worker's hiring outcome.

Apply this to your own work

See the whole job market at once.

Explore which occupations AI may reshape, then turn the signal into a practical response.

Explore the job map