Skip to main content

The workforce map: which roles in your organization change, which grow, and which end

The short answer

A role is a bundle of tasks, and that is the unit of analysis

AI does not replace job titles; it automates some tasks, augments others, and leaves the rest alone. Decompose every role into tasks before drawing any conclusion about the role, or you will both overcount and undercount the impact.

What you will be able to do

  • Decompose any role in your organization into its core tasks, because AI acts on tasks, not on job titles, and a role is a bundle of tasks that AI touches unevenly.
  • Classify each task by what AI actually does to it (automate, augment, or leave unaffected), using evidence about the specific system, not a general feeling that "AI is coming for this job."
  • Roll up task-level classifications into a defensible fate for each role: it ends, it changes, or it grows, and state the threshold logic that decides which.
  • Detect the roles that AI grows, which is the category leaders miss most often because growth is quieter than loss and shows up as new oversight, judgment, and exception-handling demand.
  • Produce a workforce map: a named, inspectable artifact that lists each affected role, its task-level evidence, its fate, the headcount and timing, and what the organization owes each person, feeding the rest of Module 9 and the Module 13 dossier.
  • Defend each entry on your map against the challenge it will face from a hostile board, an employee representative, or a works council, because a map that has not survived attack is a wish list. (see Topic 9.5)
  • Distinguish an honest workforce map from a dishonest change narrative, using the BT disclosure as a positive anchor and knowing where the dishonest version collapses. (see Topic 9.6)
  • Assign each ended role a path (attrition, redeployment, or redundancy) and a timing band, because the fate word alone hides the most important human decision and the sequence in which the organization must act.
  • Record the jurisdiction of the people on each row, because the same fate carries different legal obligations in different countries and a global map that ignores this understates what the organization owes where the law binds hardest. (see Topic 9.5)

The ideas, one by one

Every task gets one of three strict labels

Automate (no human on the routine path), augment (human stays in the loop), or unaffected. The strictness of the automate-versus-augment line is what makes the map survive the accusation that you exaggerated.

Roles have three fates, decided by a stated rule

A role ends when its automated tasks were the core and the residual does not justify a job; changes when the task mix shifts but a real job remains; grows when demand rises or new oversight work attaches. Write the roll-up reasoning down.

Growth is the category you will miss

Loss is loud and growth is quiet. Run the growth question on every role: oversight work, concentrated exception work, and demand expansion. A map that shows only cuts produces a plan that removes the capacity you are about to need.

Build the map from deployed behavior, never vendor claims

"The system can do this" is about the frontier; "our system reliably does this without a human" is about your organization. Only the second belongs on the map, and the gap between them is where overstated cuts come from.

The honest picture beats the palatable one

Softening "ends" into "changes" is a lie that fails at the first attack and pushes the organization into consultation on a false premise. Put the fate honestly and put the humanity in the obligation column.

The obligation column is what makes it leadership, not accounting

A fate without a stated obligation to the person is a cut list in a governance costume. Every "ends" and "changes" carries what the organization will do for the person, which the rest of Module 9 then executes.

The map is a dated forecast, and it is a chained artifact

Date it, state its assumptions, review it, and expect to be wrong, as BT's own revision of its AI-impact estimate shows. It feeds the hard conversation, the literacy rollout, the consultation, and the change narrative, and it is audited in the Module 13 dossier.

Every fate has a date, and sequencing is a moral choice

A role ending next month and one ending in five years demand different responses. The humane sequence usually stands up the growing roles and reskilling pipelines first, so displaced people have somewhere to go before they are displaced, not after. Add a timing band to every row.

An "ends" is not one thing

It plays out as attrition, redeployment, or redundancy, each with different obligations. Name the path on every ended role, because "ends by attrition over three years" and "ends by redundancy next quarter" are different commitments wearing the same word.

The same fate is a different obligation in different jurisdictions

Record where the affected people sit, because a group-level fate can trigger mandatory consultation in some countries and not others, and a global map that ignores this understates its obligations where the law binds hardest.

Build the map from deployed behavior, and follow the failure path

When unsure between automate and augment, ask who gets the call when the task goes wrong; if a human catches every routine failure, the task is augmented. Accountability for failure is the task that leaves a role last.

Growth must be evidenced, not asserted

A "grows" row names its mechanism (a specific oversight duty, a measured escalation increase, a demand elasticity), held to the same evidentiary standard as an "ends." Hand-waved growth deserves the skepticism it gets; evidenced growth is what makes the map credible.

The map is the hinge from systems to people

It is where AI governance stops being technical and becomes leadership. A truthful map gives the rest of Module 9 a foundation; a distorted or skipped one poisons every conversation, training, and consultation that follows.

Listen to it

The same lesson, talked through by two hosts as episode 68 of the podcast.

Real cases

These are documented cases used to show the analysis, not to predict your organization. Each is cited; each is used for a specific analytical point.

A note on how to read these: in each case, look past the headline to the analytical move it illustrates. The point is never "here is what happened to a famous company"; it is "here is the reasoning step you must be able to perform on your own organization." A learner who collects the anecdotes but not the moves has missed the topic.

Example 1: BT Group's honest disclosure (this topic's anchor). In May 2023 BT announced up to 55,000 job cuts by 2030, with the then-CEO Philip Jansen attributing about 10,000 to AI, concentrated in customer service and network management (Forbes; CNN Business, both 18 May 2023). The analytical value is the specificity: BT named a function and a scale rather than hiding behind transformation language. It is also a lesson in humility about numbers: by June 2025 the successor CEO Allison Kirkby indicated AI might drive the company "even smaller" than the original plan (The Register, 16 June 2025). A workforce map is a dated forecast, and BT's own revision proves it.

Example 2: Clerical and data entry roles at the macro scale. The World Economic Forum's Future of Jobs Report 2025 (January 2025, survey-based projection) identified clerical and secretarial workers as the largest projected absolute decline, with bank tellers, postal clerks, and data entry clerks among the fastest-declining roles. The analytical point is task composition: these roles decline fastest because a very high share of their core tasks (transcription, routine lookup, standardized processing) fall cleanly into the "automate" bin, leaving little irreducibly human residual. Roles decline in proportion to how automatable their task bundle is, which is exactly what the task-based view predicts.

Example 3: The care and frontline growth the same report projects. The same WEF report projected some of the largest absolute growth in nursing, social work, personal care, delivery, and construction roles. These grow because their core tasks (physical presence, in-person judgment, human trust, bodies in space) sit heavily in the "unaffected" bin, and because AI-driven productivity elsewhere can expand demand for them. The lesson for your map: the roles AI leaves alone can grow precisely because the technology cannot reach their core, and a map that only looks at what AI touches will miss them entirely.

Example 4: The IBM back-office hiring pause (referenced, owned elsewhere). In 2023 IBM's then-CEO said the company would pause hiring for roughly 7,800 back-office roles it expected AI and automation to absorb over time. This is a workforce decision expressed as a hiring freeze rather than a cut, and it is the anchor for the 90-day priorities decision in Module 0 (see Topic 0.4), so it is only referenced here. The analytical note worth carrying: pausing hiring is itself a workforce-map action (a role fate of "ends by attrition"), and honest maps distinguish attrition from redundancy because the human obligations differ.

Example 5: Augmentation that looks like automation. The Presto Automation drive-thru case, where the United States Securities and Exchange Commission found that offshore humans handled a large majority of orders behind an "autonomous AI" claim, is owned by the vendor-interrogation topic (see Topic 3.3). It appears here as the cautionary example for the automate-versus-augment line: a task marketed as automated was in fact heavily human. If you had built a workforce map from that vendor's claims, you would have ended roles that the system could not actually cover. The map is only as honest as the task-level evidence under it.

Example 6: The oversight role that AI creates. High-risk AI systems under the EU AI Act carry a legal human oversight obligation (owned in the risk-classification topic (see Topic 5.3)). The analytical point for the map is that this obligation is new, ongoing human work: someone must be positioned to understand, monitor, and override the system. That work is a growth entry on the workforce map, frequently attaching to the very team whose routine tasks the system automated, which is how a single deployment both ends old tasks and creates new ones on the same desk.

Example 7: The skills-transformation finding as a map input. The World Economic Forum's 2025 report projected that about 39 percent of workers' existing skill sets would be transformed or become outdated over 2025 to 2030 (a projection, so treat it as emerging). The analytical value is that this is a "changes" statement at the macro scale: it says the dominant fate is not roles ending but roles transforming, which reskilling has to serve. A map that shows mostly "ends" and little "changes" is out of step with the broad evidence and is worth re-examining, because for most organizations the transformation of surviving roles, not their elimination, is the larger and harder workforce challenge. The number is not your organization's number, but the shape of the finding (transformation over elimination) is a useful sanity check on the balance of your map.

Example 8: Attrition as a fate, expressed as a hiring pause. The IBM back-office hiring pause of 2023, in which the company said it would pause hiring for roughly 7,800 roles it expected AI and automation to absorb (owned by Topic 0.4 (see Topic 0.4)), is the clean real-world example of an "ends by attrition" fate rather than an "ends by redundancy" one. No one was dismissed; the reduction was designed to come through natural turnover and a hiring freeze over time. The analytical lesson for the map: the same role fate ("ends") can be realized through radically different human paths, and an honest map names which path applies, because "ends by attrition over several years" and "ends by redundancy next quarter" are different commitments with different obligations even though the fate word is identical.

Example 9: The multi-year horizon as a deliberate choice. BT set its workforce reshaping over a seven-year horizon (2023 to 2030) rather than compressing it, and the successor leadership continued to treat the pace as a variable rather than a fixed cut (The Register, June 2025). The analytical point is that horizon is a decision, not a given: a longer horizon opens attrition and redeployment and phased consultation, while a compressed one forces redundancy. Reading the BT case as a lesson in timing, not just in scale, is what separates an analyst who copies the headline number from one who understands why the number was spread across years.

Example 10: The macro shape as a sanity check, not a source. The WEF 2025 projection of simultaneous large displacement and larger creation (about 92 million and 170 million by 2030, a projection) is worthless as a source for any specific role on your map, because your organization is not the world economy. Its analytical use is as a sanity check on the balance of your map: if your map shows only elimination and no transformation or growth, it is out of step with the broad evidence that transformation dominates and creation exceeds displacement, and that mismatch is a prompt to re-examine whether you ran the growth and change analysis honestly or stopped at the losses.

Where people go wrong

  • "Will AI replace [job title]?" as the framing. This is the foundational error, because the job title is the wrong unit of analysis. AI acts on tasks, and every role is a bundle of tasks that AI touches unevenly. Reasoning at the title level produces both false losses (killing roles whose residual tasks still justify a person) and false safety (keeping headcount for roles that have quietly hollowed out). Always decompose to tasks first.
  • Confusing automation with augmentation. A task where a human still verifies, handles exceptions, or owns the outcome is augmented, not automated. Counting augmented tasks as automated is the single largest source of overstated job-loss numbers, and it is exactly the gap that the Presto and Nate enforcement cases exposed between "autonomous AI" claims and human-in-the-loop reality (see Topic 3.3) (see Topic 4.1). Be strict: automated means no human on the routine path.
  • Building the map from vendor claims instead of deployed behavior. "The system can do this" is a marketing statement about the frontier; "our deployed system reliably does this without a human" is a statement about your organization. Only the second belongs on the map. A map built on capability claims ends roles the system cannot actually cover.
  • Softening "ends" into "changes" to avoid the conversation. This is the most damaging mistake because it is a lie that fails at the first attack and sends the organization into consultation on a false premise. If a role's core tasks are automated and the residual does not justify a job, the honest label is "ends," and the humane response is in the obligation column, not in relabeling the fate.
  • Only mapping the losses. Growth is quiet and loss is loud, so first-draft maps almost always miss the roles AI grows: oversight work, concentrated exception work, and demand expansion. A map that shows only cuts produces a plan that removes the capacity the organization is about to need. Run the growth question on every role.
  • Treating the map as a one-time slide instead of a dated forecast. A workforce map is a forecast and forecasts age. BT's own 2023 estimate of about 10,000 roles to AI was later signaled as possibly understating the impact (The Register, June 2025). Date the map, state its assumptions, and set a review cadence, or it becomes a quiet lie as the technology and business move.
  • Skipping the obligation column. A fate without a stated obligation to the person is a cut list wearing a governance costume. The "what we owe the person" column is what makes the map a leadership instrument rather than a spreadsheet, and it is what Modules 9.2 through 9.5 execute.
  • Assuming automation is neutral to the survivors. Automating the routine 80 percent leaves humans doing the hard, high-risk 20 percent, which is more stressful and more skilled, not easier. A changed role is often a harder job than the one it replaced, and a map that treats "changed" as "no big deal" underestimates the reskilling and support burden.
  • Collapsing the three ways an "ends" plays out. "Ends" can mean attrition (not refilling as people leave), redeployment (the person moves to a growing role), or redundancy (the person leaves the organization). These carry completely different obligations, and a map that writes "ends" without naming the path hides the most important human decision. Name attrition, redeployment, or redundancy for every ended role.
  • Ignoring where the people sit. The same fate carries different legal obligations in different jurisdictions: a "changes" affecting a group in Germany or France may trigger mandatory works-council or committee consultation before you may act, while the floor is lower elsewhere. A single global map that ignores jurisdiction understates the obligations exactly where they bind hardest (see Topic 9.5).
  • Decomposing roles from job descriptions instead of observed work. A job description is an aspirational document, often years out of date, written for hiring rather than for accuracy. Building the task list from it alone produces a map that is confidently wrong. Decompose from what people actually do, verified with the people who do it, not from what the role was once written up to be.
  • Reasoning from the incumbent's skill instead of the task mix. "This person is too talented to be affected" confuses a property of the person with an analysis of the tasks. A talented person can hold a highly automatable role; the fate follows from the tasks, and the person's talent belongs in the redeployment conversation, not in the fate.
  • Mapping only the present state. A role can be stable today and hollowing on a two-year view because deployments roll out over time. A map built only against how the systems behave today will keep discovering fates after the window to handle them humanely has closed. Decompose the role against the deployment roadmap, and mark a role trending toward "ends" while there is still time to reskill.
  • Treating "changes" as automatically kind. A role can change so far that it becomes a job the current incumbent cannot or will not do. Heavy change is a real transition needing real support, and for some people the honest answer is redeployment to different work rather than forcing them through a change they did not choose. The obligation column must be honest about changed roles, not only ended ones.

Questions people ask

What is workforce map?
A dated, inspectable artifact that lists each role affected by an organization's AI systems, decomposed into tasks, with each task classified by what AI does to it, each role assigned a fate (ends, changes, or grows), and each fate carrying the organization's obligation to the person. Produced in this topic, consumed across Module 9, and audited in the Module 13 dossier.
What is task-based view of automation?
The settled expert framework (associated with the economist David Autor, 2015) that technology substitutes for human labor in some tasks and complements it in others, so the effect on any occupation depends on the mix of tasks inside it, not on the job title. The reason a role is the unit that gets a fate but a task is the unit of analysis.
What is task decomposition?
Breaking a role into the five to ten concrete tasks that actually make up the job, stated as action verbs (answer, check, calculate, de-escalate, approve), so that AI's uneven effect on different tasks can be seen. The foundational step of building a workforce map. More on Task decomposition
What is automate (task fate)?
The label for a task the deployed AI system performs end to end with no human on the routine path. The strict test is deployed behavior, not vendor claims, and not tasks where a human still verifies or handles exceptions (those are augment).
What is augment (task fate)?
The label for a task where AI changes how the human does it but the human stays in the loop, doing the judgment, the exception, the sign-off, or the relationship. Augmented tasks change a role rather than end it and are where reskilling concentrates.

Keep going

Watch the lesson and prove it

Proving it happens inside the program: watch, explain it in your own words, and it goes on a record employers can check.

Watch the lesson and prove it →