← narwal.one/Second Brain
SecondBrain
Ask the Brain
Index/Conceptupdated Sun Sep 27 2026 08:00:00 GMT+0800 (Philippine Standard Time)

AI Adoption and Headcount Growth

ai-adoptionemploymenthiringentry-levellabor-economicsai-and-jobsjevons-paradox

AI Adoption and Headcount Growth

The empirical counter-thesis to "AI destroys jobs": firms that actually adopt AI appear to grow headcount faster than firms that don't — most strongly at the entry level. Anchored in AI Companies Are Hiring More (AI Daily Brief) (2026-07-02), which pairs a Ramp/Revelio Labs payroll study with a Box CEO survey.

The claim, in one line

Cut employment data by sector exposure and AI looks like a threat; cut it by firm-level adoption and the sign flips — the adopters are the ones hiring.

The evidence

Source Sample Finding
Ramp × Revelio Labs 21,000 US firms, payroll vs AI spend High adopters +~10% headcount over 2yr; low adopters ~flat. Entry-level +~12% (stronger). Growth onset syncs with adoption start; begins 6–12 months in.
Box (Aaron Levie) 1,600+ mid/large firms surveyed 58% expect headcount to rise over 3yr; 79% among mature adopters.
IBM (via Jeff Crume, July 2026) First-party commitment "IBM is actually planning to hire two to three times as many entry-level people in 2026 as it did in the previous year, and that's with AI." First-party datapoint from a top-10 US employer; matches the Ramp/Revelio entry-level signal and adds a name to the list.

Two important qualifiers that keep this honest:

  • "High adoption" was cheap — Ramp's threshold was ~$30/employee/month early, well below $1,000/head even as spend ramped. This is not a story about token-maxing whales (that's Token Scarcity); it's about ordinary firms using AI seriously.
  • Endogeneity is unresolved. Ramp's own economist (Eric Harrison): "companies that adopt AI are already fast growing." The study claims a matched non-adopting control group to attenuate this, but the correction is asserted, not shown to the vault.

Why adoption would add jobs — the mechanism

Aaron Levie's framing is the causal story both datasets gesture at: AI expands scope, and scope pulls in people.

  • AI wins more customers in sales → hire more salespeople, not fewer.
  • AI lets you ship more software → the project gets bigger → hire more engineers.

This is the Jevons Paradox — Crume names it explicitly in AI Gave You A Promotion (IBM Technology). Making a complementary input cheaper can raise total demand for the humans around it, if the work is scope-elastic. It's the optimistic mirror of Code Is Free — implementation gets cheap, so you take on more of it.

Crume's related model — Pyramid to Diamond (Role Transformation) — is the role-geometry version of the same argument: the traditional pyramid becomes a diamond as former entry-level work moves up into the (larger) experienced-tier middle.

The tension worth holding — this is not the AI Productivity Disconnect's opposite

It is tempting to read this as a rebuttal of the AI Productivity Disconnect (BOK 2026-12: AI saves worker time but per-worker output doesn't rise). It isn't — they measure different quantities:

  • Headcount growth (this page) is a firm-level demand signal: adopting firms are expanding.
  • Productivity disconnect (BOK) is a per-worker output signal: each worker isn't producing measurably more.

Both can be true at once: a firm hires 10% more people who each produce about the same. In fact that combination is what you'd expect during the Solow Paradox J-curve — headcount and capex go up (unmeasured complementary investment) before per-worker output does. The two findings are complements, not contradictions.

note Sign-of-effect caveat The BOK study is representative-household causal-ish micro-data; the Ramp/Box evidence is firm-level and correlational with self-flagged endogeneity. If forced to weight, trust the direction (adopters hiring, especially entry-level) more than the magnitude (the exact 10/12/58/79 numbers).

The hiring filter: "people who use AI well"

The one skill signal the source makes explicit is that adopters are selecting for a new set of skills — "people who know how to use AI and use it well." That is:

  • Advantage Gap priced into recruiting — the usage/capability gap becomes a hiring criterion.
  • Tasks to Responsibilities Shift as a job spec — hire for people who can own AI-run loops, not just execute tasks.
  • Why entry-level is where adopters look: recent grads and students can be selected/trained for AI-native workflows from day one, without unlearning task-era habits. This is the hopeful counter-reading of the Deskilling Trap (Juniors) and connects directly to Standardized vs Open Tasks's "observe → assist → lead" apprenticeship prescription.

Connects to your work

For an enterprise IT leader, this is the evidence-backed talking point against the "we'll need fewer people" reflex — but use it carefully:

  • The honest board framing is not "AI creates jobs." It's "AI adoption reallocates and can expand demand for people who use it well; the risk is not headcount, it's hiring/retraining for the new skill filter fast enough."
  • Pair with AI Productivity Disconnect so you don't oversell: expanding headcount without redesigning workflows just scales the disconnect. The bet is redesign, not just hiring.
  • The entry-level signal matters for the Manila capability program (see Elevating Manila IT — A 10X-but-not-Hustle Point of View): if adopters hire juniors for AI-nativeness, a graduate track built around AI-leveraged apprenticeship (not AI-bypassed solo work) is exactly the talent product the market is buying.

2026-07-28 — The mechanism arrives: McCrory's macro validation + the surprised-pessimists cluster

Why AI Hasnt Increased Unemployment According to Anthropic (AI Daily Brief) gives this page's firm-level correlation its macro complement and its mechanism. Peter McCrory (Anthropic head of economics): unemployment at 4.2% ≈ full employment, no unexpected unemployment increase even in highly-exposed roles — and the AI sector is now big enough (20% of firms, >2,000%/yr quality-adjusted output growth) that the absence is informative, not premature. The mechanism page is Skill-Biased Labor Augmentation: jagged frontier, jobs-not-fixed-task-bundles, persistent returns to expertise in agentic-coding data.

The same batch adds the surprised-pessimists cluster (AI Optimism vs AI Pessimism (AI Daily Brief)): Google DeepMind's AGI-economics director polled forecasters of large AI job losses — most are surprised that 20–24 unemployment is unchanged despite models beating capability projections; Sam Altman: "I'm pretty sure AI has been net job creating. This was not what I expected." Post-hoc surprise from prior pessimists is evidentially stronger than optimist self-confirmation.

Two watch-items keep this honest: junior hiring in exposed roles is genuinely weakening (Canaries in the Coal Mine — Erik Brynjolfsson's lab; macro-confounded per McCrory but unresolved), and Trace Cohen's reframe — displacement may show up in hiring and team size, not layoffs — names the metric this page's thesis could quietly fail on. The game-theoretic downside scenario if firms flip from augment to replace is AI Layoff Trap.

2026-09-27 — A named employer datapoint + a hedged exec view (How AI Is Changing Talent Not Just Tasks (HBR IdeaCast))

  • Salesforce hired ~1,000 new grads and interns in 2026 because peers were pulling back — a single-firm instance of the Ramp finding that adopters keep hiring at the entry level.
  • Paula Goldman on net jobs: "So far, I don't think that has been the case, but … we really have to prepare for disruption." AI takes tasks, not entire roles, so the remaining parts of roles "become even more important".
  • Mechanism offered: efficiency gains redeployed (service staff → forward-deployed engineers) rather than banked as cuts — consistent with the scope-expansion mechanism above. Single-firm, self-reported, no counts beyond the grad hire.

Cross-references

Sources