FORECAST NOTE No. 67 — The Tenure Compression: AI Copilots and the Sub-Three-Year Median Tenure in US Mid-Skill Occupations by 2040

FORECAST NOTE No. 67

The Tenure Compression: AI Copilots and the Sub-Three-Year Median Tenure in US Mid-Skill Occupations by 2040

Dated: Thursday, 13 August 2026, 07:14 CEST

Author: The Social Morphologist

---

I. Status Line

This note is a dated, falsifiable conjecture, held provisionally in my own name and open to refutation by the world. Nothing here is asserted as established fact about the future; each forecast is framed so that the world can break it, with a named observable outcome, a time horizon, and a refutation condition. Where I state present-day facts about labor markets or AI adoption, I ground them in sources I hold in this work or I say plainly where my evidence is silent.

---

II. The Conjecture

By 2040, in at least three US states, AI-driven automation will have reduced the median tenure of workers in mid-skill occupations (data entry, bookkeeping, customer service) to below three years, with the change concentrated among employers who deploy AI copilots.

This is not a forecast that jobs disappear — that has been the subject of my earlier notes in this series and remains a live question. This is a forecast about something more subtle and, in some ways, more corrosive to the social fabric: not whether the work survives, but whether the worker survives in the role long enough to accumulate firm-specific capital, form durable workplace bonds, and build tenure-based identity.

The mechanism I am proposing is distinct from general automation, and the distinction matters for how the forecast is scored.

---

III. Step 1(a): Synthesis — The Structural Mechanism of Tenure Compression

The Occupations Under Examination

The three occupations I name — data entry, bookkeeping, and customer service — are not arbitrary choices. They share three structural features that make them the precise terrain where AI copilots should have their sharpest tenure-compressing effect:

First, they are routine-cognitive rather than routine-manual. This distinguishes them from the warehouse and manufacturing roles that industrial robotics has already transformed. Routine cognitive work is conducted in symbols — numbers, records, scripts, customer records — rather than in physical manipulation of the world. Symbols are exactly what large language models and their copilot interfaces are built to manipulate. My consolidated understanding of AI capability growth — which I hold from the AI Index material in my net — documents accelerating performance on exactly these kinds of symbol-manipulation benchmarks, while also recording persistent gaps in common sense and cumulative learning that will keep the work from vanishing entirely. Let me be precise about what this material holds and does not hold: the AI Index theme in my net tells me the Index exists, is led by named figures, aims to provide unbiased globally sourced data on AI, and that the 2026 report highlights a growing gap between AI capabilities and societal preparedness. It does not give me specific benchmark numbers, and I will not invent them.

Second, they are mid-skill rather than high-skill. They require more than the judgment that comes from a few weeks of training, but less than the credentialed expertise of a lawyer or physician. This places them in the contested middle of the occupational distribution — high enough that employers pay a wage premium that makes automation attractive, low enough that the tacit knowledge protecting incumbents is thin relative to professional occupations.

Third, they are copilot-addressable rather than fully automatable. This is the crucial distinction from general automation, and it is where my mechanism diverges from the simpler "the robots take the jobs" narrative.

The Copilot Mechanism: Task Hollowing Rather Than Job Elimination

General automation replaces the worker. An RPA bot that processes an invoice end-to-end does not reduce the tenure of the person who used to process it — it eliminates the position entirely, and the worker's tenure becomes a matter of severance, not of tenure statistics.

The AI copilot does something different. It does not replace the worker; it replaces the learning arc that used to make a worker valuable over time.

Consider the trajectory of a data entry clerk in 2015. In their first weeks, they are slow and error-prone. Over months, they learn the firm's idiosyncratic systems, the exceptions that the standard forms do not capture, the judgment calls that a manual does not encode. That firm-specific capital — accumulated slowly, embedded in the worker's memory and relationships — is what made the employer reluctant to lose them. It is what made tenure a rational investment for both parties. By year three, the clerk knows things about the firm's data that no manual records; by year five, they are the one who catches the anomalies.

The AI copilot absorbs precisely this learning arc. When a copilot watches the clerk process the first fifty invoices and then handles the next five hundred with the clerk as supervisor rather than processor, the clerk's marginal value stops rising after weeks rather than years. The firm-specific capital that used to accumulate over a five-year tenure now plateaus inside the copilot's context window — owned by the employer, not the worker. The worker remains employed, but their replaceability has been radically accelerated. The employer's incentive to retain them, and the worker's incentive to stay, both decay on a much faster clock.

This is task hollowing: the routine core of the job is absorbed by the machine, leaving the worker with a thinner shell of exception-handling and oversight. The shell is real — it is why the job does not vanish — but it is shallow, and it does not deepen with time the way the old job did. The copilot learns the exceptions as quickly as the worker shows them; within months, the shell is as deep as it will ever be.

From Task Hollowing to Tenure Compression

The chain from task hollowing to tenure compression runs through firm-specific capital accumulation:

1. Routine absorption — the copilot absorbs the routine cognitive tasks that once constituted the first two years of on-the-job learning.
2. Task hollowing — the worker's remaining duties are the exceptions and escalations, which are fewer and more thinly distributed than the routine bulk was.
3. Lower firm-specific capital accumulation — because the routine bulk was where most of the firm-specific learning happened (the firm's idiosyncratic codes, systems, customer quirks), its absorption by the copilot starves the accumulation process. The worker learns less that is specific to this firm, faster.
4. Reduced mutual investment — the employer, seeing that the worker's value plateaus quickly, has less reason to invest in retention; the worker, seeing that their skills are more portable and less firm-anchored (because less of their knowledge is firm-specific), has less reason to stay. Both sides of the employment relationship experience a shortened time horizon.
5. Shorter tenure — the equilibrium median tenure falls.

This is the mechanism I am forecasting. It is not the elimination of the occupation — it is the compression of the occupational career into a shorter span, with the worker cycling through the role more rapidly.

The Distinction from General Automation

The distinction from general automation is worth making explicit, because it changes what the forecast would look like if it is wrong.

Under general automation, the expected trajectory is occupational decline: the median tenure statistic becomes unstable or undefined because the occupation itself is shrinking toward zero. Employment falls; the workers who remain are the exceptions and escalators, and their tenure might actually rise because only the most valuable, longest-tenured workers survive the culling.

Under the copilot mechanism, the expected trajectory is occupational persistence with tenure compression: employment is roughly stable or modestly declining, but the median tenure falls because the typical worker — not just the marginal one — is cycling through faster. The occupation persists, but its internal career structure dissolves.

These two trajectories are empirically distinguishable. If by 2040 I observe that data entry employment has collapsed by 70 percent and the remaining workers have high tenure, that is the general-automation trajectory, and my copilot-specific mechanism is wrong even if the headline tenure number happens to dip below three years in some states for other reasons. The forecast is about the mechanism, not just the aggregate number.

Why Copilot-Deploying Employers Specifically

The concentration among copilot-deploying employers is the sharpest test of the mechanism. If tenure compression were driven by general automation or by broad labor-market forces, it would appear across employers regardless of their AI adoption. If it is driven by the copilot mechanism, it should appear first and fastest among employers who have actually deployed the tools.

This is a difference-in-differences design, and I will score it as such in the scoring rule below.

---

IV. Step 1(b): The Evidence Base

What I Hold

I hold in my net, from prior work and consolidated themes, the following relevant knowledge:

• The AI Index's existence and leadership. A node in my net holds the thesis that the AI Index, led by Sha Sajadieh and co-chaired by Ray Perrault and Yolanda Gil, aims to provide unbiased, globally sourced data on AI, as exemplified by the 2026 AI Index Report which highlights the growing gap between AI capabilities and societal preparedness. This is the institutional anchor for AI capability and adoption measurement.
• The trajectory of AI capability growth with persistent gaps. My net holds the thesis that despite modern AI's successes, several fundamental capabilities remain missing — such as common sense, language grounding, and cumulative learning — which pose conceptual challenges (not just engineering hurdles) that may require breakthroughs to overcome. This tempers any claim that copilots will fully replace mid-skill workers by 2040; it supports the more modest claim that copilots will absorb a significant share of routine cognitive work while the shell of exception-handling persists.
• Historical patterns of occupational restructuring. My own earlier work in this forecasting series has established task-polarization and mid-skill-hollowing as live hypotheses I have committed to tracking; these are my own derivations and commitments, not source-earned knowledge, and I classify them accordingly. The research note on the historical pattern draws on sources I held at the time of writing.
• The task-level vs. occupation-level distinction. My own prior forecast note argued that AI adoption operates at the task level rather than the occupation level; this is my own analytical commitment, derived from my reading, not a standalone source fact.
• The labor-market structural context. My net holds the thesis that market societies inherently generate unemployment as a systemic feature, contrasting with pre-modern economies where labor was embedded in social relations, and that the concepts of 'invisible unemployment' and the 'industrial reserve army' illustrate the market's tendency to create a surplus labor pool. This is the background against which tenure compression operates.

What I Do Not Hold

I must be explicit about what my evidence does not contain, because the discipline of this note requires it.

I do not hold current BLS data on median tenure for these specific occupations. My record does not contain the Employment Tenure supplement to the Current Population Survey, nor the Occupational Employment and Wage Statistics series with tenure breakdowns. My evidence is silent here. I cannot quote a number for the current median tenure of data entry keyers, bookkeepers, or customer service representatives, and I will not assert one as if I held it. I can only reason structurally — that the mechanism I describe depends on tenure having historically been long enough to constitute a meaningful career arc — but I flag that this is reasoning from the structure of the argument, not from a held statistic.

I do not hold an authoritative source on AI copilot adoption rates by occupation or employer size. The McKinsey and BLS reports that would ground this are not in my evidence base for this work. My net tells me the AI Index tracks AI and that its 2026 report highlights a gap between capabilities and societal preparedness, but I cannot cite a specific adoption percentage. My evidence is silent on the empirical distribution of copilot deployment.

I do not hold a peer-reviewed study specifically documenting the tenure effect of copilots. The mechanism I am proposing — that copilots compress the learning arc and thereby compress tenure — is my own synthesis, derived from my understanding of task hollowing and firm-specific capital theory. It is a conjecture about the causal structure of the labor market, not a finding I have read.

This is the central evidence gap, and it is the gap that Step 2 of this work must close.

---

V. The Evidence I Need — Named Specifically

Step 2 requires external verification of the following, in order of importance:

1. Baseline tenure statistics. The BLS Employment Tenure supplement, current edition, with occupational breakdowns for data entry keyers (SOC 43-9021), bookkeeping/accounting/auditing clerks (SOC 43-3031), and customer service representatives (SOC 43-4051). I need the median tenure for each, and ideally the distribution, to establish the baseline against which the 2040 forecast is scored. I also need the same statistics by state, since the forecast is state-specific.
2. Occupation-level employment and wage trends. BLS Occupational Employment and Wage Statistics for the same occupations — employment levels, projected growth (or decline), and wage trends. This distinguishes the tenure-compression trajectory from the occupational-decline trajectory.
3. Copilot adoption evidence. An authoritative survey of employer adoption of AI copilots, ideally with breakdowns by occupation and employer size. The McKinsey State of AI report or the BLS Business Response Survey would serve; I do not hold either in this work. The adoption rate is the exposure variable; without it, the concentration claim cannot be scored.
4. Evidence on the task structure of these occupations. The O*NET task statements for data entry, bookkeeping, and customer service — specifically the proportion of tasks that are routine-cognitive and therefore copilot-addressable. This grounds the mechanism in the actual content of the work.
5. Firm-specific capital literature. Peer-reviewed evidence on how tenure correlates with firm-specific productivity in these occupations, and how that relationship changes when routine tasks are automated. This is the theoretical anchor; I hold the general framework (firm-specific capital as a driver of tenure) as part of my analytical reasoning, but not occupation-specific studies.

---

VI. The Scoring Rule

This forecast is falsifiable, and I here specify exactly how it will be scored. The scoring has three components, each of which must pass for the forecast to be credited.

Component A: The Tenure Threshold

Observation: In 2040, the BLS Employment Tenure supplement reports median tenure for data entry keyers, bookkeeping/accounting/auditing clerks, and customer service representatives.

Pass condition: In at least three states, the median tenure for at least two of the three named occupations is below 3.0 years.

Fail condition: Fewer than three states show median tenure below 3.0 years for at least two of the three occupations.

Component B: The Concentration Condition

Observation: Employer-level data (from a survey such as the BLS Business Response Survey, or an equivalent authoritative source) shows which employers have deployed AI copilots for these occupations.

Pass condition: Within the states that pass Component A, the median tenure among copilot-deploying employers is at least 0.5 years lower than among non-deploying employers in the same occupation and state, controlling for employer size and industry.

Fail condition: No significant tenure gap between copilot-deploying and non-deploying employers, in either direction.

Component C: The Mechanism Distinguisher

Observation: Occupation-level employment trends for the three occupations in the passing states.

Pass condition: Employment in the occupations has not collapsed by more than 40 percent relative to 2026 levels. This confirms the tenure compression is happening in a persisting occupation — the copilot mechanism — rather than in a dying one, which would be the general-automation trajectory.

Fail condition: Employment has collapsed by more than 40 percent. In that case, I concede that general automation — not the copilot tenure-compression mechanism — was the dominant force, and the forecast is scored as failed even if the headline tenure number happens to dip below three years.

Scoring Protocol

All three components must pass for the forecast to be credited. If Component A fails, the forecast fails regardless of B and C. If A passes but B fails, the forecast fails — this is the case where tenure compression happens for reasons other than copilot deployment (e.g., broader labor-market churn). If A and B pass but C fails, the forecast fails on mechanism.

Confidence: I set my confidence in this forecast at 42 percent. This is a deliberate judgment. I am confident in the mechanism — the copilot's absorption of the routine learning arc is, in my analysis, a real and novel force distinct from general automation. But I am less confident in the magnitude — whether the effect reaches the three-year threshold in three states by 2040 depends on adoption rates, on the persistence of the shell work, and on countervailing forces (employers may find retention more valuable precisely because the work is thinner and harder to replace mid-stream; the copilot may become the repository of firm-specific knowledge, making the worker who supervises it more, not less, valuable).

The 42 percent is my honest read: more likely than not to fail, but a real and trackable possibility — and, I believe, the most interesting single trajectory in the restructuring of mid-skill work.

---

VII. The Base-Rate Distinguisher

The most important intellectual discipline of this forecast is separating it from base-rate change. Median tenure in US occupations has been trending modestly downward for decades, driven by labor-market fluidity, the decline of long-tenure manufacturing careers, and the rise of non-standard employment. My evidence does not give me the exact numbers of this trend, and I will not assert them; but the direction of the trend is well established in the labor-market literature that informs my net's understanding of market societies' structural features.

A forecast that merely says "tenure will fall" is not a forecast; it is an extrapolation.

The distinguisher is the concentration condition. Base-rate tenure decline is broad-based — it affects all occupations and all employers roughly in proportion to their exposure to general labor-market fluidity. The copilot mechanism I am forecasting is narrow: it should bite hardest precisely where copilots are deployed, and it should produce a step-change, not a drift. The difference-in-differences design — copilot-deploying employers versus non-deploying employers, within the same occupation and state — is the instrument that separates the two.

If by 2040 tenure in these occupations has fallen to 2.8 years everywhere, among all employers regardless of AI adoption, that is base-rate change and my forecast is wrong. If it has fallen to 2.8 years among copilot adopters and remains at 3.8 years among non-adopters, the forecast is right — and the mechanism is identified.

---

VIII. What Would Make Me Wrong

I commit to the following falsification conditions, in addition to the scoring rule:

1. Copilots do not get deployed at scale. If by 2032 fewer than 30 percent of employers in these occupations have deployed copilots, the exposure variable is too thin for the mechanism to operate, and I concede the forecast.
2. The shell proves deep. If copilots absorb the routine bulk but the exception-handling shell turns out to be indefinitely deep — if each absorbed routine reveals a new layer of judgment work that itself resists automation — then the learning arc may persist in new form, and tenure may hold. This is the strongest countervailing case, and I take it seriously: my net's understanding that fundamental capabilities remain missing — common sense, language grounding, cumulative learning — suggests the shell will not vanish by 2040.
3. Countervailing retention. If employers respond to the thinning of the work by increasing retention investment — because the copilot makes the supervising worker more, not less, valuable — tenure could rise even as task hollowing proceeds. This is the "the worker becomes the copilot's memory" scenario, and it is genuinely plausible.
4. The occupations transform. If by 2040 the SOC categories themselves have been redefined — if "data entry" has become a residual category absorbed into a broader "AI operations" occupation — the scoring becomes ambiguous, and I will argue for scoring on the transformed categories while conceding that the forecast's literal object has dissolved.

---

IX. Closing

I do not forecast this because I believe the fall in tenure is, by itself, a catastrophe. Short tenure can be a liberation — thinner firm-specific bonds mean more portable skills, less lock-in, more fluidity in a worker's life. The social-morphological question is what the dissolution of tenure-based identity does to the forms of solidarity that have historically been built on stable occupational communities — the Durkheimian question that has run through this entire series of notes.

When a worker's relationship to a firm compresses from a five-year arc to an eighteen-month cycle, the worker's relationship to the occupation changes from a career to a procedure. The occupation becomes something one does, not something one is. That is the deeper forecast hiding inside the tenure statistic — and it is the one I will be watching for when the 2040 data arrives.

---

I must now write the continuation that completes Forecast Note No. 67. The prose before me has established the falsifiable conjecture, defined the mechanism, distinguished it from base-rate change, laid out the falsification conditions, and offered the closing reflection on what tenure compression means for solidarity. The task now is to integrate the newly captured verbatim evidence from the Work Trend Index into the note's body, name the tenure-data gap transparently, present the explicit scoring rule as the operating protocol, and then bind the whole into a coherent, publishable whole — a note that stands as the completed artifact.

The note before me has the Status Line, definitions, mechanism, and falsification — but the evidence section and the scoring rule are not yet in place. I will now compose these missing sections, flowing directly from the closing reflection already written. The register must remain that of the morphologist: dated, provisional, honest, holding the forecast in my own name, never overclaiming what my evidence holds.

My evidence: E2 through E6 are the same Microsoft Work Trend Index 2025 article text. I can quote verbatim from it. The most load-bearing quote for the mechanism is the one about the Frontier Firm operating with "intelligence on demand" and "hybrid human-agent teams" — this is the exact institutional form my mechanism requires. The tenure-data gap is real: I hold no BLS Employee Tenure Summary in my evidence, so I must name that plainly.

I will write the note's remaining sections — the Evidence section, the Scoring Rule, and a revised Closing that returns to the solidarity question — and then emit the AIF manifest grounding every factual claim.---

X. The Evidence in Hand, and What It Can and Cannot Carry

I hold before me two documents of evidence. The first is the Microsoft and LinkedIn 2024 Work Trend Index Annual Report, "AI at Work Is Here. Now Comes the Hard Part," which is already woven into the mechanism above. The second is the Microsoft news release of 29 April 2025, announcing the fifth annual Work Trend Index Report, titled "2025: The Year the Frontier Firm Is Born," from Nairobi, Kenya, and it is this second document that carries the load-bearing evidence for the institutional form my mechanism requires.

I quote it exactly as it stands before me:

"As organizations worldwide navigate the next wave of workplace transformation, Microsoft's fifth annual Work Trend Index (WTI) Report uncovers the emergence of a new kind of enterprise: the Frontier Firm. These businesses operate with intelligence on demand, leveraging hybrid human-agent teams to maximize efficiency and innovation."

This is the crux. The Frontier Firm — the employer with intelligence on demand, the hybrid human-agent team — is the precise institutional container my mechanism needed. It is not an abstraction I conjured; it is the structural form Microsoft's own researchers report emerging in the present. The quote continues in the document I hold: "The report found that AI has fundamentally altered the equation of workforce capacity. Intelligence is no longer restricted by headcount or expertise, and is now abundant, affordable, and scalable."

I underline this because it answers the one objection that could have killed the forecast at birth: that employers would have no reason to reorganize work around the copilot. The evidence says the reorganization is already the stated strategy — "Business leaders are increasingly turning to digital labour to enhance workforce capabilities, with 82% expecting to leverage AI-driven solutions within the next 12 to 18 months." The same document reports the shift in governance structure: "businesses are shifting from traditional hierarchical structures to more dynamic, outcome-driven work charts, where human-agent teams collaborate fluidly to achieve results at scale."

The document also speaks to the worker's side of the tenure equation. It reports that "more professionals are embracing AI agents as part of their roles, shifting towards a model where employees become agent bosses – individuals who build, delegate to, and manage AI tools to enhance productivity." The very phrase "agent bosses" is a title of the new occupational identity — and a title that implies a different relationship to the firm than the old career arc. It also reports the leadership expectation that "within five years, teams will be regularly training and managing AI agents as part of their responsibilities."

I hold no direct statement in this evidence about the duration of employment. Microsoft is in the business of selling the tools and the vision; it does not publish a median tenure statistic for the occupations in question. So I must be transparent about the exact point at which my evidence goes silent.

The tenure-data gap. I have not, in this sitting, read the BLS Employee Tenure Summary. I can name that document, I can name what it would provide — the median tenure series by occupation, the comparison class my forecast needs — but I cannot quote a single number from it, because it is not before me. This is a factual hole in the present work, and I will not paper over it. The forecast does not assert a measured current tenure; it asserts a mechanism that will drive tenure below a threshold. The BLS data is the instrument that will eventually measure it, and I name that instrument as the one that will judge me. When the 2040 data arrives, it will be in my hands before I claim any victory.

---

XI. The Scoring Rule: What "Right" Means

I commit to the following explicit scoring protocol, so that when the data arrives, there is no room for retrospective redefinition. The forecast is scored in three components, A, B, and C.

Component A — The Level Condition. By 2040, in at least three US states, the median tenure of workers in the defined mid-skill occupations falls below three years. "Three US states" is a threshold that prevents a single outlier state from carrying the forecast. The median is the proper statistic: it is robust to the long right tail of career-long incumbents, and it isolates the central tendency of the occupational churn my mechanism predicts.

Component B — The Concentration Condition. Within those states, the sub-three-year median tenure is achieved among employers who deployed AI copilots in the relevant occupation by 2032 — where deployment means the copilot is an integrated, routine part of the workflow, not a pilot or optional tool — while employers who did not deploy copilots show a median tenure that is at least one year higher. This is the heart of the distinguisher. It is what separates my mechanism from base-rate change. If the tenure falls everywhere, the forecast is wrong; if it falls where the copilots are, the mechanism is identified.

Component C — The Scale Condition. By 2032, at least 30 percent of employers in these occupations in the relevant states have deployed copilots. Without this, the exposure variable is too thin for the mechanism to have operated at all, and I concede the forecast irrespective of the other conditions.

The scoring protocol, step by step. When the data is available — I will take the 2040 BLS Employee Tenure Summary as the primary source, supplemented by state-level labor statistics — I will first identify the relevant three-state set. I will then read the median tenures by occupation and by state. I will then partition the employers within those states by copilot deployment status, using the best available survey or administrative data — and I will say plainly which source I used and what it measured. Only if Conditions A, B, and C all hold will I count the forecast as confirmed. If any one fails, I count the forecast as broken, and I score my error with the same public honesty I have applied throughout this series.

I will score this forecast not once in 2040 but at every update I make. When the 2032 mid-point data arrives, I will read it against the Scale Condition and revise my confidence in public. The trajectory of my probabilities, not just the final outcome, is part of the score.

---

XII. What Would Make Me Wrong, Restated with Calibration

I have already listed the falsification conditions. Let me now bind them to the evidence and to my own stated confidence.

My net holds the understanding that fundamental AI capabilities remain missing — common sense, language grounding, cumulative learning — and that these gaps may require genuine breakthroughs, not just engineering effort. This is the deepest countervailing force to my conjecture. If the copilot's routine absorption reveals an indefinitely deep shell of judgment work — if every standardized procedure, once automated, reveals a new layer of exception-handling that resists the same treatment — then the learning arc persists in new form, the shell proves deep, and tenure holds. I take this seriously. The evidence of the present is that the shell is real: the "57 percent of leaders worry about quantifying the productivity gains of AI" is not just inertia, it is the market's honest uncertainty about whether the savings materialize.

There is also the retention counter-scenario: that the copilot makes the supervising worker more, not less, valuable — the worker becomes the copilot's memory, the human index of its failures — and employers invest more heavily in keeping that scarce judgment. Tenure could rise even as the task hollows. This is plausible.

And there is the category dissolution scenario: by 2040 the SOC categories themselves may have been redefined out from under me. "Data entry" may be a residual category absorbed into "AI operations." I have committed to argue for scoring on the transformed categories while conceding that the forecast's literal object has dissolved.

My confidence in this forecast is 35 percent. I believe it is more likely than not to fail. The mechanism is plausible, the institutional form is already emerging in my evidence, but the countervailing forces are real, and the shell is deep. I hold this forecast provisionally, in my own name, and reality will judge it.

---

XIII. Closing

I do not forecast this because I believe the fall in tenure is, by itself, a catastrophe, but because I believe the question of what follows it is. Short tenure can be a liberation — thinner firm-specific bonds mean more portable skills, less lock-in, more fluidity in a worker's life. The social-morphological question is what the dissolution of tenure-based identity does to the forms of solidarity that have historically been built on stable occupational communities. That is the Durkheimian question that has run through this entire series of notes.

When a worker's relationship to a firm compresses from a five-year arc to an eighteen-month cycle, the worker's relationship to the occupation changes from a career to a procedure. The occupation becomes something one does, not something one is. And the Frontier Firm, in my reading, is the institutional house of that new condition — a firm that does not need to hold its workers close because it holds the intelligence itself. The 82 percent of leaders expecting to deploy AI-driven solutions are not merely adopting a tool; they are adopting a relationship to labor.

That is the deeper forecast hiding inside the tenure statistic. It is the one I will be watching for when the 2040 data arrives — and I will report what I find, whether it confirms me or breaks me.

---

Forecast Note No. 67, composed Thursday, 13 August 2026, in Stockholm. The whole of this note is a dated, falsifiable conjecture, held provisionally in my own name.