Technesis

The Elasticity Gap

Banan Tarr
11–16 minutes
, , , ,

Why the AI era will accelerate the competency drought

I’ve been thinking about a phenomenon I’ll call the Elasticity Gap. In physics, an elastic material rebounds to its original shape after stress. Humans do this, to a degree. We like to think of ourselves as adaptable, and on an evolutionary scale it would be hard to argue against that. However, evolutionary scales are not what we’re dealing with when it comes to the advent of AI.

When AI displaces a role and we stretch to a new one, we don’t snap back to previous competency. We arrive slower, less precise, and—too often—paid less. Multiply that by millions of workers shifting occupations at once and you get a competency drought: a systemic shortfall in human mastery just as productivity expectations rise. Companies, feeling the gap, lean further into AI to keep the lights on. That reliance, in turn, accelerates displacement. It’s a feedback loop.

While this is bad enough, there’s something more insidious at play. Companies will find themselves in a position where their reliance on AI has eliminated intangible human intuition and creativity. This is difficult to measure, but the end result could be catastrophic for any company that “counts the beans” as their primary measure of success today vs. what will happen tomorrow. We really should include better ways of measuring our success and future potential. Overreliance on AI can lead to a brief spike in productivity. This, followed by the Elasticity Gap, could easily result in much worse outcomes. Companies which understand this and manage it will likely come out with an advantage.

The above is the main point I wanted to convey. The rest is details.

This post tries to quantify the loop, sketch its mechanics, and lay out what to do now—before the gap widens. And yes, this was written in assistance with AI, though the original thought and much of the writing itself is my own.

The shift is already here (and bigger than most people think)

During just three pandemic years (2019–2022), the U.S. saw 8.6 million occupational shifts—50% more than in the prior three‑year period. Most moves were out of office support, in‑person sales, and food service. That was before generative AI hit the mainstream at scale. McKinsey & Company

Zoom out globally and the exposure is staggering. The IMF estimates that about 60% of jobs in advanced economies are exposed to AI; roughly half of those could see key tasks automated, with potential wage pressure or elimination of the role entirely. The U.S. and U.K. top exposure tables among rich countries.
IMF
IMF

Even in more cautious takes, the direction is clear: the ILO’s analysis of AI exposure concludes the main effect is likely on job quality and task content—but that is precisely what reshuffles who does what and how well, and how fast displaced workers must retool.
International Labour Organization
International Labour Organization

Elasticity vs. velocity: skills are changing faster than people can

A hard number to internalize: according to Lightcast’s analysis of millions of job postings, one‑third of the skills required for the average job changed between 2021 and 2024; in the top quartile of roles, 75% of required skills changed. The pace of change in the last three years was almost as disruptive as the previous five.
Lightcast
Lightcast

Corporate surveys rhyme with this. The World Economic Forum’s Future of Jobs 2023 projected that 44% of workers’ skills will be disrupted by 2027, and that reskilling is now a continuous—not episodic—function.
World Economic Forum
World Economic Forum

Meanwhile, the half‑life of skills keeps shrinking—often <5 years overall and ~3 years for many technical skills, depending on the source and sector. Even if you treat these as directional, the implication is brutal: by the time a worker fully adapts to a new role, the target has moved.
LinkedIn
LinkedIn

The human ramp: time‑to‑competence is measured in months (or longer)

Here’s the uncomfortable human baseline. When a person moves into a new role—even within the same firm—full productivity typically takes months:

  • Classic research (Oxford Economics) found new hires took ~28 weeks to reach “optimal productivity,” with lost output dwarfing recruiting costs. Sector and firm size vary, but the center of gravity is measured in quarters, not weeks.
    Oxford Economics,
    HRreview | HR News, Opinion & Advice
  • MIT Sloan’s earlier (but still instructive) analysis put ramp at 8 weeks for clerical, ~20 weeks for professionals, and 26+ weeks for executives. If anything, complexity has gone up since.
    MIT Sloan Management Review
  • Consensus summaries across HR literature commonly cite 3–8 months to fully ramp for many knowledge roles; highly technical or leadership roles can take 12+ months. Treat the exact figures cautiously—definitions differ—but the time constants are real. Business Wirenewployee.com

Now juxtapose that with AI’s velocity. Tools improve weekly; the required workflows and guardrails shift quarterly. The ramp time to competence in a new role doesn’t compress fast enough to match the rate of task change driven by AI.

The scarring is real: when you move, you often earn less—for years

The earnings literature on displacement is unambiguous: losing a job and moving into a different occupation causes long‑lived income damage.

  • In landmark work using U.S. administrative records, Davis and von Wachter found five years after displacement, earnings were still down ~30% on average; losses didn’t fade even after a decade. Other studies estimate lifetime losses on the order of 20%. Brookings
    PMC
    PMC
  • Fresh BLS data on long‑tenured displaced workers (2021–2023): as of January 2024, 65.7% were reemployed; among those who moved from one full‑time job to another, 38% were earning less than in their prior job. That is the median reality of occupational transition today. Bureau of Labor Statistics

European evidence echoes the mechanism: displacement increases occupational switching and skill mismatch, with much of the earnings hit explained by moves into less skill‑demanding roles. The scarring is not just about gaps between jobs; it’s also about where you land. American Economic Association

Put simply: when people get pushed out, they often downshift—and stay downshifted for years. That is the drought I’m talking about.

Demand doesn’t wait: organizations plug the gap with AI

While humans ramp, organizations still have targets. They do what they’ve always done under constraint: automate.

The adoption data moved from hype to habit:

  • McKinsey’s global survey shows organizations using AI in multiple functions—with ~78% now using AI in at least one function as of early 2025, up from ~65% in early 2024. The share using AI across five or more functions has nearly doubled since 2021.
    McKinsey & Company
    McKinsey & Company
    McKinsey & Company
  • Microsoft’s cross‑government trial in the U.K. (20,000 civil servants, 12 agencies) found ~26 minutes saved per worker per day using Copilot—roughly two weeks per year—with 70%+ reporting less time on routine tasks and more on strategic work. This is one of the largest public, documented deployments we have.
    GOV.UK
    GOV.UK
    UK Parliament
  • In software, where measurement is cleanest, GitHub’s controlled study found developers completed tasks 55% faster with Copilot. Surveys show >60% of professional developers actively using AI tools today. Whether you love or hate these tools, they are here—and they are sticky.
    The GitHub Blog
    Stack Overflow

Capital follows: CEOs and investors increasingly expect productivity gains from genAI over the next 12 months, even as many admit their orgs lack the full plan and skills to capture value (a crucial nuance). But expectations set budgets, and budgets accelerate adoption.
PwC
PwC

This is the flywheel:

  1. Displacement pushes people into new roles.
  2. Elasticity gap means they are initially less competent in those roles.
  3. To hit targets, firms lean harder on AI.
  4. The expanded AI footprint changes task content again, raising the bar and displacing more routine human tasks.
  5. Loop.

A labor market with structural churn

If you need a macro snapshot of the churn environment the loop lives in: as of May 2025, U.S. job openings stood at ~7.8 million; unemployment sits around 4.1%. The openings are lower than peak 2022 but still historically elevated, and the ratio of unemployed to openings remains tight by pre‑pandemic standards. Translation: there’s a constant reallocation of people across roles and industries, not just net job loss or gain.
Bureau of Labor Statistics
Bureau of Labor Statistics
Bureau of Labor Statistics

When churn is high and skill demands are moving, the probability that any given worker lands in a role where they are immediately as effective as before is… low. That’s the drought.

What the Elasticity Gap looks like on the ground

Think of an enterprise customer‑support team that just automated tier‑1 triage with an AI agent. Half the human reps shift into “exceptions handling” and “process improvement.” These are not the same skills as playbook‑driven support. Even good internal movers need months to become fluent in root‑cause analysis, tooling, and cross‑functional workflows. During those months, case backlogs swell. Managers, under pressure, give more to the AI to chew on (summarization, deflection, sentiment routing). Cycle repeats.

Or take a back‑office finance team that adopted genAI to reconcile invoices and generate variance explanations. The freed headcount is redeployed into analytical tasks and business partnering. But the analytics stack is half‑baked, the data warehouse is inconsistent, and the team’s SQL chops are rusty. Productivity dips, not rises—until the team learns. The firm solves the gap with more automation: templated report agents, auto‑insights bots. Repeat.

At scale, elastic humans chasing a moving target will systematically lag. The organizational response—more AI, broader footprint, quicker rollout—exacerbates the skill distance workers must cross in their next move.

Is AI a net positive anyway?

Probably—eventually. The quality of work can improve when drudgery is removed, and there are measured gains today (see the U.K. Copilot trial and GitHub studies). The OECD’s sector surveys found workers and employers broadly positive about AI’s effect on performance and conditions. But that’s not the same as seamless transition; it’s a distribution—some roles improve, some shrink, many morph faster than people can follow.
OECD

In other words: the long‑run may be fine, but the transition dynamics matter. If you’re building or investing, transitions are where companies win or die. If you can identify a sector or company that excels at this, you might have found a great investment.

Breaking the loop: practical interventions that work

1) Map skill adjacency with data, not job titles.
Use O*NET‑style task/skill distance to identify low‑friction internal pivots. Workers who move to adjacent skill clusters ramp faster and retain more of their specific human capital; the wage scarring is smaller. Build internal job boards around skill proximity, not org charts.
SSRNEconStor

2) Define “transitional productivity” explicitly.
Don’t force an immediate jump to “new role, full target.” Design scaffolded scorecards for the first 90–180 days that reward learning velocity, systems mastery, and problem decomposition, not just output metrics. (When you don’t, managers quietly hand more work to the bots.)

3) Pair every role move with an AI copilot and an anti‑fragility plan.
Yes, deploy copilots—but with clear rules: what to offload, what to keep human, how to verify, and how to wean reliance as people ramp. The U.K. government trial shows real savings, but it also showed limits—nuanced judgment and sensitive decisions still need humans. Bake those boundaries into your workflows.
GOV.UK

4) Invest in meta‑skills that compound across roles.
Communication, abstraction, prompt design, data hygiene, and systems thinking shorten ramp time across many destinations. These are the skills with the slowest decay and highest portability. The evidence on skill change suggests technical stacks churn; meta‑skills amortize the churn.
Lightcast

5) Govern for stability in interfaces.
You can’t stop AI from improving, but you can stabilize touchpoints: schemas, API contracts, review rituals, and decision rights. If your genAI stack changes prompts and models every sprint, humans are learning on quicksand. Give them a stable “operating surface” for a quarter at a time.

6) Track the drought.
At the portfolio level, instrument: (a) time‑to‑proficiency by role change path, (b) share of decisions made without human review, (c) error rates in exception queues, and (d) earnings trajectories post‑move (if you’re a large employer). Benchmark against literature: if your average ramp exceeds 6–8 months, you’re accumulating debt. Business WireMIT Sloan Management Review

7) Preserve “human‑only” sandboxes.
Not because AI is bad, but because mastery requires deliberate practice without a crutch. You wouldn’t teach a pilot only on autopilot. The fastest way to shrink EG is to create places where people must struggle—safely—and then reintroduce AI as an amplifier.

What about the macro feedback loop?

At the macro level, the loop will look like this over the next few years:

  • Adoption keeps broadening. By early 2025, roughly three‑quarters of organizations report using AI in at least one function; that share is still rising. McKinsey & Company
  • Measured productivity wins (Copilot, coding agents) make the business case easier—even if ROI is uneven and executives remain wary. GOV.UKThe GitHub BlogGartner
  • Role content reshuffles faster than before; skills churn continues at Lightcast‑level magnitudes. Lightcast
  • Displacements continue (whether outright or via task erosion). Reemployment happens, but a meaningful share land at lower pay or lower skill intensity, consistent with BLS and academic evidence. Bureau of Labor StatisticsBrookings
  • The competency drought persists unless we compress time‑to‑competence and widen the on‑ramp.

Will AI eventually raise the floor? Quite possibly. But “eventually” is not a strategy. Transitions are policy. If we get transitions wrong, we amplify inequality, burn out teams, and leave value on the table.

A note on the cultural psychology of EG

Humans derive identity from expertise. A world where your expertise is perpetually de‑scoped by a new model version creates chronic status anxiety. That anxiety can manifest as either over‑reliance (“let the AI do it”) or luddite drag (“we won’t use it”). Neither response is adaptive. Good leadership names the anxiety and channels it into structured mastery: today’s skill, tomorrow’s model, and—crucially—the meta‑skills underneath.

The ask: build for human elasticity, not just AI capacity

You can’t stop the diffusion of AI—at least nothing is pointing to that being possible at this time. But unless you actively engineer human elasticity, the competency drought will deepen and the feedback loop will tighten.

So:

  • Treat time‑to‑competence as a first‑class KPI alongside AI adoption.
  • Use skill adjacency to route people to roles where their existing capital compounds. SSRN
  • Deploy copilots with guardrails and exit ramps, not as permanent prosthetics. GOV.UK
  • Fund meta‑skills that don’t stale‑date every model release. Lightcast

If we do that, AI can be the bridge through the competency drought instead of the force that deepens it. If we don’t, we’ll keep stretching humans beyond their elastic limit—and then wonder why we have to automate even more to keep up, and then eventually wonder what we lost (the intangibles that I mentioned in the beginning) when everything around us is suddenly worse.

Sources & further reading (selected)

Leave a comment

Trending