AI-Native Organization

Sources#

• 2026 State of Scaling: The Great Sorting
• Engines of Growth: Global Startup Trends Report
• Garry Tan: Own Your Intelligence
• Indian AI Coding Startup Emergent Becomes a Unicorn with $130M Series C
• Startup ARR is less secure than ever, new research shows
• State of AI 2026: The Builder's Economy
• The Human-AI Substitution Principle: When will you be replaced by AI in your organization?
• The ICONIQ Pacesetter Index
• The New Physics of Business — Garry Tan, Y Combinator
• The state of AI in 2026: On the road to ROI

Summary#

Garry Tan's thesis (AI Engineer talk, July 2026, practitioner-opinion): the components of an organization that used to require hiring — roles, org charts, processes, performance reviews — can now be encoded as markdown files that agents execute. An AI-native company is not a company that uses AI; it is a company shaped like this from day one: a thin team, skills for every recurring function (sales, support, ops, finance), and engineers hired specifically to maintain the skills and do the work the skills can't do yet. "We've been building organizations this whole time, but we didn't have a management layer — now that's what we have."

The org mapping#

The heart of the talk — each organizational primitive has a markdown counterpart:

Organizational primitive	Agent-infrastructure counterpart
Employee	Skill file — one capability, one job, written clearly enough to execute (Agent Context Files)
Org chart	Resolver table — a task comes in, the resolver decides who handles it and where it goes (the "load test.md when altering tests" table)
Internal process	Filing rules — whether the resolver is actually routing in compliance
Performance review	Trigger evals — a test that the right skill actually gets loaded when needed (Evals as Product Spec)

"When you sit down with Claude Code or Codex, you're not writing software. You're hiring, training, and managing a workforce made of markdown."

The organization of one (August 2026)#

Tan's Startup School keynote (Garry Tan: Own Your Intelligence, practitioner-opinion) runs the identical mapping — skill file = employee, resolver = org chart — but points it at a single person rather than a company, and that reframing is the update worth recording: "before you ever incorporate anything, before you have a co-founder or a logo or a deck, you can already be running an organization. An organization of one plus your agents. You are the founder and the entire management layer… and the head count under you is now whatever you decide it is."

Two consequences he draws:

• The org precedes the company. The AI-native shape is no longer something a startup is founded into; it is something an individual is already running before incorporation. This is Founder as Agent Orchestrator pushed one step earlier in the lifecycle, and it is what makes the ownership question on Owning Your Externalized Cognition load-bearing — if the org is one person plus their markdown, the org chart and the person's career capital are the same files.
• Software stops being precious. "You can build exactly the tool you need for the audience of one in a weekend." His revision of the canonical advice: scratch your own itch not because it might be a market but "because scratching itches is nearly free," and you'll learn which tools-for-one were companies when other people start asking to borrow them. The problem-selection version of Implementation Abundance Inverts Product Work.

The revenue-per-head exhibits are restated unchanged from the July talk (Emergent 15 people at $15M ARR, Retell $60M at ~40, the Winter 25 batch quarter with 95% AI-generated codebases), so they gain repetition but no new evidence — the third-party check below still governs. He again explicitly disclaims causation.

The claimed new physics: revenue per head#

Tan's evidence that companies built this way operate in a different regime (all figures his, unverified):

• Emergence/Emergent (AI app builder, YC S24): public launch → nine figures of ARR in eight months; 15 people at $15M ARR.
• Retell (YC W24): $60M revenue with ~40 people.
• Winter 25 YC batch: a quarter of companies had codebases 95% AI-generated — and became "the fastest-growing, most profitable batch in the history of YC."
• "That kind of revenue per head did not exist before. Not in software, not in oil, not in railroads."

He explicitly does not claim causation ("I can't prove that the AI-generated code caused the growth") — the claim is that the fastest-growing founders treat AI as a workforce, not autocomplete. Note the aggregate counter-signal at AI Investment Story, Not Efficiency Story: across 50K+ companies AI firms currently show lower revenue-per-employee than non-AI peers; Tan's examples are the tail, not the mean (the same average-vs-tail split flagged on Founder as Agent Orchestrator).

Population context, self-reported (AWS, June 2026). Between Tan's ~$1M/head anecdotes and Emergence's ~$394K median sits AWS's Engines of Growth founder survey: 55% of AI-natives report earning $400K+ revenue per employee (vs 34% of startups globally and 23% of large enterprises). It reads as a middle instrument — a self-reported distribution clustered around the same ~$400K threshold Emergence's cap-table median lands on, but framed as ahead-of-baseline where the cap-table data finds AI behind matched non-AI peers. So the three RPE readings triangulate rather than settle: practitioner anecdote (tail, rosiest) → founder survey (population self-report) → cap-table receipts (matched-segment, soberest). The direction-of-comparison conflict is handled as an instrument split on AI Investment Story, Not Efficiency Story, not averaged.

The Emergent figure, third-party checked (flag, don't smooth)#

Emergent is now the wiki's one revenue-per-head claim with an outside verification trail — the Emergent entity page carries the full detail; the load-bearing summary:

• The company is real and large. TechCrunch (2026-07-15) reports Emergent became a $1.5B unicorn on a $130M Series C, with company-reported $120M ARR (+70%/4mo) and 200K+ paying customers — directionally confirming Tan's "different physics" framing.
• But the record per-head number compresses on inspection. At scale, $120M ARR / ~200 employees ≈ $600K/head — below Emergence Capital's $100M+ top-decile AI-company RPE of $960K (AI Investment Story, Not Efficiency Story, empirical), and roughly half the ~$1M/head implied by Tan's own "15 people at $15M" snapshot. The headline extreme belongs to the 15-person moment; it regresses toward the mean as headcount scales — the investment-phase-staffing signature AI Investment Story, Not Efficiency Story measures across 50K+ companies.
• Contradiction, weighted by tier. Tan (practitioner-opinion, transcript that self-flags the company name as an ASR guess) places Emergent "out of Summer 24"; TechCrunch (vendor-claim; the round and founders are verifiable) says the Jha brothers co-founded it June 2025. On the founding date TechCrunch is the more authoritative source; the ARR/headcount snapshots are compatible earlier-vs-later moments rather than a strict conflict. Full disposition on Emergent.
• Weighting caveat. TechCrunch's ARR/headcount/customer figures are themselves company-self-reported (vendor-claim), so the $600K/head is checked against an empirical benchmark but rests on an unaudited numerator; the Retell half of Tan's claim ($60M at ~40 ≈ $1.5M/head) still has no third-party corroboration.

The numerator's durability, which no RPE instrument on this page measures. Every reading above — Tan's anecdotes, AWS's survey, Emergence's cap-table medians, Emergent's $120M — divides ARR by headcount and treats the numerator as an annuity. TechCrunch (2026-09-03) reports a survey finding that argues it is less annuity-like than the acronym implies: 77% of 150 enterprise IT professionals say they re-evaluate their AI vendors every six months or on a rolling basis, which Madrona calls a "fast in, fast out" dynamic against enterprise SaaS's "moat of inertia" (full treatment, including why this is not the same as a churn rate, on Seven Powers Applied to AI). Two reasons this is recorded as a caveat and not a correction. It is practitioner-opinion — a VC's survey reaching the wiki through a news article, with no churn, renewal or retention figure anywhere in it — and its population may not even be these companies': Emergent's 200K+ paying customers are overwhelmingly non-technical individual and prosumer accounts, not enterprise procurement committees on a semi-annual review. This is explicitly not an answer to the revenue-per-head question below, which asks for third-party corroboration of the per-head levels; it flags a different variable — that two firms with identical RPE can hold revenue of very different quality — and the instrument that would test it (cohort retention by headcount band) does not exist in the corpus.

Not just engineers#

The extension Tan says most engineering talks miss: at YC the transformation runs through media staff, event staff, and finance — "people who have never opened a terminal in their lives are building skill files and cron jobs." One finance staffer collapsed ~100 Excel workbooks into a single app built with OpenClaw and YC's company brain: "She's not a programmer. She's a manager of agents now." The claim generalizes Founder as Agent Orchestrator beyond founders and beyond engineering: "It's not just 400x engineers. It's one company that operates at the level of 400x everyone else." This is Printing Press Software Democratization observed inside one institution.

The restructuring, surveyed (ICONIQ, Q2 2026)#

Tan asserts the AI-native org shape from a stage; ICONIQ Growth's State of AI 2026 measures the intent to restructure across ~305 executives at AI-building software companies (empirical survey; forward figures prediction-grade). It confirms the reshaping is about role composition, not just team size, which is the part Tan's "different physics" leaves vague:

• 78% are restructuring, split by kind (single-select, N=302): 45% plan a different mix of roles — "fewer operational, more AI-fluent talent" — with no net headcount reduction; 33% plan a smaller team given AI efficiency gains; only 9% backfill-only, 9% hiring faster, 4% unchanged. The plurality is a composition shift, not a cut.
• Function-level headcount diverges by role (single-select, N=303): R&D/Engineering, Sales, and Product & Design are net-growing; Customer Success and Marketing are mixed; Customer Support and G&A are net-shrinking. The org isn't flattening uniformly — it's reallocating heads from operational/support toward build and revenue functions, echoing the engineering-heavy allocation Emergence measures on cap tables.
• Two named new engineering roles are the growth categories: forward-deployed engineers (~50% of companies scaling FDEs as a permanent GTM motion, monetized as revenue drivers — see AI Product Economics Maturation) and AI safety / trust & reliability engineers (companies "planning to hire more"). G&A shows the sharpest substitution: operators report removing finance-ops and order-management roles as agentic workflows took over, redirecting budget to strategic and AI-specific functions — "the bar for any new G&A hire is rising: roles are added only where AI cannot yet cover the work."
• The org runs flatter and more cross-functional: cross-functional structure is more common at AI-heavy companies (42% at 50%+ AI-revenue vs 35% below); at $100M+ scale, 72% of 50%+-AI-revenue companies run on just 1–4 management layers vs 56% of peers; and high-growth firms are widening spans of control (first-line R&D managers with 7+ reports 21%→30%, GTM managers 26%→33%, 2025→2026). One early-stage operator "collapsed PM and designer into single-person product ownership" — the role convergence Tan's skill-file mapping assumes, observed in the wild.

This is survey-of-intent, not a headcount census, and the forward hiring plans are self-report — but it is the first population-scale evidence that the AI-native restructuring is a role-composition event (toward AI-fluent/build/revenue roles, away from operational/support), not merely leaner teams.

The flattening bullet is also the vault's only confirming evidence for a formal prediction. Banerjee & Singh's HAT substitution model (arXiv 2607.20781, July 2026, practitioner-opinion — a formal model with no data) derives organizational flattening as its prediction P3, from a mechanism: human managerial cost escalates super-exponentially with depth at (1 + r_0 D)^{D−i+1} while AI coordination cost is assumed to escalate more slowly, so at the optimum depth falls (D^(AI)* ≤ D*) and — since level sizes are the product of spans — holding execution capacity fixed forces spans to widen. The predicted signature is exactly the two numbers above: fewer management layers and wider spans, together. Of that model's seven predictions this is the only one the vault can corroborate at all; the rest are untested, contradicted, or resolved by measurement instrument rather than by the economy. Three caveats keep it from being a clean confirmation: the ICONIQ cut is cross-sectional (AI-heavy companies vs peers at one moment, not the same firms before and after), it is self-reported by AI builders rather than AI adopters, so selection on company age and founding shape is uncontrolled, and the widening spans run straight into the oversight ceiling the same vault measures — review capacity does not expand when output does. The correlation matches; the mechanism is not identified.

The intent, checked against the receipts two months later (ICONIQ State of Scaling, September 2026)#

The survey above measures what ~305 executives plan. 2026 State of Scaling (September 2026, empirical) is the same publisher's operating-data report, and its headcount chapter is the closest thing available to a check on that intent. The intent does not show up in the receipts.

• Adoption is effectively total and the daily-use rate is high (Annual Growth Operating Trends Survey, n = 38, February 2026): employees with AI-tool access run 99% S&M / 98% R&D / 97% G&A, and daily users 74% / 86% / 70%. The report's headline "70%+ are daily active users" is the G&A floor, not the average — R&D is 86%. Note the n: this is 38 companies, bolted onto a 137-company operating dataset, and it is the source of every AI-adoption figure in the report.
• The functional mix did not move (p.44, average headcount distribution by function, 2022-23 → 2026). Under $100M: S&M 48 → 50 → 50 → 50%, R&D 38 → 38 → 37 → 41%, G&A 14 → 13 → 13 → 9%. Above $100M: S&M 39 → 49 → 45 → 47%, R&D 39 → 38 → 41 → 39%, G&A 22 → 13 → 15 → 14%. ICONIQ's own conclusion is blunt: "headcount distribution across functions remains largely unchanged, suggesting the broader workforce model has yet to meaningfully shift" and, elsewhere, "while AI is improving productivity, it has not yet meaningfully reshaped operating models."

This is the directly relevant test of the 45%-plan-a-different-role-mix finding, and it fails it — with one caveat that survives. If a plurality of AI-builders were executing a composition shift from operational toward AI-fluent roles, the S&M/R&D/G&A split is the coarsest place it would appear, and across four years it is flat to within a few points. The one movement that is in the predicted direction is real but small and thin: G&A under $100M falls 13% → 9% between 2025 and 2026, which is precisely the function the survey's operators said they were removing (finance-ops, order management), and R&D rises 37% → 41% in the same step. The 2026 column rests on n = 7 company-quarters under $100M and n = 8 above it, so a four-point move is one or two companies. Call it consistent with the survey's direction and nowhere near confirmation of it.

Why the two can both be honest. A three-bucket function split is a very coarse instrument: the survey's claim is about which roles inside S&M and R&D are being hired (AI-fluent versus operational, FDEs versus traditional CSMs), and that substitution is invisible to a chart that only counts the buckets. The composition shift Tan's mapping predicts is intra-functional; ICONIQ measures inter-functional. So this does not refute the restructuring thesis — it establishes that four years into the AI era, the org chart's top-level shape has not changed, which is the ceiling any claim about reshaped operating models now has to argue past.

The mechanism the report does supply is about backfills, not restructures. Under "Perspectives from the ICONIQ Network," the observed behaviours are: attrition "increasingly being used as an opportunity to avoid, delay, or down-level backfills"; "no-backfill policies… explicitly tied to AI-driven productivity improvements in sales, content, operational, and analytical functions"; and "hiring freezes… used while companies evaluate AI-driven productivity gains before approving incremental headcount." Emerging titles named: AI Business Analyst, AI Integration Engineer, GTM AI leader. That is a quieter and more plausible transmission path than a restructure — the org changes shape by not replacing people, which takes as long as attrition takes and would not yet be visible in a four-year function chart. It is also unquantified: ICONIQ gives no share of companies running such a policy, and these are an investor's qualitative observations of its own portfolio.

Tension: the employee metaphor#

Tan's central metaphor — skill files as employees, agents as a workforce you hire, train, and manage — is structurally the exact framing AI Employee Framing (Kropp et al., HBR May 2026, empirical, n=1,261) tested against: employee-framing measurably cut personal accountability (−9pp), raised escalation (+44%), and reduced error-catching (−18%), with no adoption gain. The reconciliation is the same one Founder as Agent Orchestrator needed: Tan uses the metaphor as an org-design blueprint (encode roles as files, test them with evals — artifacts, not anthropomorphized coworkers), while Kropp measures the psychological framing of agents as org-chart peers. Encoding a role as an auditable markdown file may even be the accountability-preserving form of the metaphor — the "employee" is a versioned document with a test suite, not a "Kevin" whose mistakes belong to no one. Neither source engages the other; the synthesis is this wiki's.

Relation to the complements thesis#

Tan's sharpest line — "the 2x people and the 100x people are using the exact same Claude. Same weights, same context window, same API. The leverage is not in the weights, it's in how you wire the work" — is a practitioner restatement of Organizational Complements to AI (same model, 99.8% vs 63.3% vs 16.5% usage across populations; the gap must be complements). The org mapping above is his answer to which complements: encoded procedure (skills), routing (resolvers), and verification (trigger evals). His "never do one-off work — skillify it" discipline is the normative version of the systematization margin Agentic Work Systematization measures (skill use 5.4%→26.6% of weekly-active Codex users, Mar→Jun 2026): "The organization that captures what it learns like this gets smarter every single day. The one that doesn't wakes up every morning with amnesia."

The redesign claim at population scale (McKinsey, August 2026)#

The org-as-artifact mapping above is a frontier-accelerator picture. The state of AI in 2026 (McKinsey / QuantumBlack, 2026-08-25, empirical but self-reported, n=1,719) supplies the population version of its weakest-sounding premise — that the organization itself, not the tooling, is what gets rebuilt — and finds it is the single largest thing separating the organizations reporting financial impact from everyone else.

Among AI high performers (6% of respondents; at least 5% of EBIT attributed to AI plus "significant" value), ~72% report having fundamentally redesigned workflows because of AI use, against ~25% of all others — the widest gap in the survey's practice chart (Exhibit 11, a dumbbell chart with no printed labels; gridline-read, approximate), ahead of transformative ambition (~62% vs ~15%) and senior-leader role modeling (~65% vs ~32%). The prose gives the unapproximated version: nearly three-quarters of high performers report fundamentally redesigning workflows, up from 55% a year earlier, against one-quarter of other respondents. Tara Balakrishnan's closing commentary states the constraint in this page's own terms — "the limiting factor is increasingly the organization's ability to absorb change."

Two things it does not supply, both relevant here: no role or occupational taxonomy of any kind, and no revenue-per-head figure. The survey can say that organizational redesign travels with reported AI value; it cannot say what the redesigned organization is staffed with.

Connections#

• The Solo-Founder Shift — the "organization of one" measured, and reframed as a waypoint: solo-founded companies are the fastest to add a second person (median 399 days vs 480 for co-founded teams), so the org of one is a stage passed through rather than a destination
• Agent Context Files — the substrate: the org mapping is the context-file pattern promoted from configuring one agent to encoding a whole company
• Excellence as an Operating System — the big-company original of the operating norms AI-native startups converge on: Lenny's observation that the early Netflix culture deck (agency, talent density, top-of-market pay) reads like a description of how the top AI labs now run
• Agentic Work Systematization — the measured counterpart: Tan's "skillify it" rule is the normative form of the systematization margin OpenAI's Codex telemetry tracks
• Evals as Product Spec — trigger evals as performance reviews: the eval-as-spec idea applied to the org's own routing layer
• AI Employee Framing — the empirical counter-evidence to the workforce metaphor; see the tension section above
• Founder as Agent Orchestrator — the founder-scale version of this role shift; Tan extends it to every employee ("everyone at YC is a manager of agents now")
• Role Averaging, Not Role Elimination — the role-composition half of the restructuring, at population scale: ICONIQ's 45%-plan-a-different-role-mix (not a net cut) is Ambrosino's "averaging, not elimination" measured across ~305 AI-builders
• Organizational Complements to AI — the economics frame Tan's "leverage is not in the weights" restates; the org mapping names the complements. Also where the HAT substitution model lives: the 1–4-management-layers and widening-spans figures above are the vault's only confirming evidence for that model's flattening prediction (P3)
• Human-AI Accountability Redesign — the ceiling on the widening-spans finding: oversight capacity does not expand when output does, so "fewer layers, more reports each" is a structure that has to be paid for in redesigned review, not just declared
• Forward-Deployed Engineering as a Delivery Layer — the first of the two new engineering hiring categories examined as an industry structure rather than a headcount line: what the role is, the 5–10x fully-loaded-cost gate that rations it to large accounts, the ~1 GM / 1 FDE Lead / 2–3 AEs / 8–10 FDEs per-market template, and who ends up employing it
• AI Product Economics Maturation — the survey-scale corroboration of the restructuring (45% role-mix shift, function-level reallocation, FDE + AI-trust roles, flatter/wider-span orgs across ~305 builders) and the unit-economics side of the same "proving AI pays" deck; Ramp's 350 versioned reusable workflows are the "skillify it" discipline measured in one company
• AI-Native Startup Lifecycle — Anthropic's stage-by-stage playbook for the same target; Tan adds the org-primitive mapping and YC-portfolio revenue-per-head claims
• AI Investment Story, Not Efficiency Story — the aggregate counter-signal: AI companies on average show lower revenue-per-employee; Tan's Emergent/Retell figures describe the tail
• Printing Press Software Democratization — non-engineers building skill files is the democratization thesis observed inside YC's own staff
• Returns to Expertise in Agentic Coding — Tan's 400x self-report vs the measured 2× actions / 5× output expertise premium; see that page's evidence note on the gap between telemetry and self-assessment
• Latent vs. Deterministic Space — the companion engineering discipline from the same talk: knowing which side of the org's computation belongs to the model and which to code
• LLM-as-Compiler Knowledge Base — the company brain (library + librarian) is the AI-native org's memory layer; Tan's GBrain is an in-the-wild instance of the compiled-wiki pattern
• Emergent — Tan's headline revenue-per-head exhibit, now a third-party-checked $1.5B unicorn; the datapoint on which this thesis's per-head claim partly holds and partly compresses
• Owning Your Externalized Cognition — the same mapping read from the encoded person's side rather than the employer's. If a skill file is an employee, it is also somebody's judgment written down, and Tan's August 2026 keynote argues the file's value accrues to whoever holds the repo — the tension this thesis generates and does not resolve
• Implementation Abundance Inverts Product Work — the "software isn't precious anymore" corollary: when building the tool is nearly free, problem selection rather than implementation is what's scarce
• Garry Tan — the thesis's author; OpenClaw — the harness YC runs it on
• Standardize the Infrastructure, Not the Tools — the platform decision underneath the org design: standardize the layer beneath the tools rather than the tools, on the reasoning that nobody yet knows which model or workflow wins
• The Committed-Artifact Chain — the same markdown-as-org-substrate bet reached from the software lifecycle instead of from operations, and by a vendor rather than an investor. Both make version-controlled plaintext the thing agents execute and both put the enforcement in configuration rather than in process; they differ on what sits at the top of the mapping — skill files as employees and resolver tables as org charts here, versus committed stage artifacts (intent.md → spec.md → plan.md → the PR) whose merges are the handoffs there. Tan's trigger evals and Anthropic's continuous-evals play are the same instrument pointed at the same target: does the right configuration actually load and still work after it changed.
• Skill Lift — the performance review's missing half. Trigger evals test that the right skill loads; NVIDIA's with/without-skill ablation tests whether the loaded skill changed the outcome, and reports it per product — the first instrument in the corpus that scores an "employee" on results rather than on being summoned

Open Questions#

• Tan's revenue-per-head figures (Emergent ~$15M ARR at 15 people, Retell $60M at ~40) are stated from stage without sourcing. Do third-party data (Carta/Standard Metrics cohorts, press-verified ARR) corroborate record revenue-per-head at AI-native YC companies, or do these examples regress toward the AI Investment Story, Not Efficiency Story mean on inspection? (Partially answered: TechCrunch, 2026-07-15 corroborates Emergent as a real, fast-growing $1.5B unicorn — $120M company-reported ARR, 200K+ paying customers — so the direction holds. But the record per-head claim compresses: at scale it is ~$600K/head (200 employees), below the $100M+ top-decile AI-company RPE of $960K AI Investment Story, Not Efficiency Story and about half the ~$1M/head of Tan's own 15-people/$15M snapshot — the per-head extreme is a low-headcount-phase artifact that regresses as the company staffs up. Caveats keeping this open: TechCrunch's figures are themselves company-reported vendor-claim, not Carta-audited, and the Retell half ($60M at ~40 ≈ $1.5M/head) remains unverified. AWS's June-2026 founder survey adds a population reading (55% of AI-natives self-report $400K+/head) but it is self-report, not the cap-table/press verification this question asks for. See Emergent.) Sharpened by a peer-group benchmark, and it points against the claim (2026-09-22): The ICONIQ Pacesetter Index (empirical, quarterly operating financials 2024 – Q2 2026, not a survey) is the closest thing yet to the third-party cohort data this bullet asks for. It benchmarks revenue per FTE for companies that are AI-forward and top-quartile three-year growers — Tan's exhibits' own peer group — at $655K median / $890K top quartile for $100M+, and $75K / $115K / $225K medians in the three smaller bands. Two consequences. Emergent's $600K/head lands marginally below the median of its peer group, not at a record. And the sub-$10M median of $75K/FTE says the fast-growing AI cohort is, at small scale, less productive per head than conventional software — so Tan's "15 people at $15M ARR" ($1M/head) is an extreme outlier against its own cohort rather than a new normal. Still not a full answer: ICONIQ selects its cohort with no control group and is benchmarking its own portfolio in part, the Retell half remains unverified, and this is a cohort distribution rather than the per-company verification the bullet requests.
• The org mapping predicts a testable staffing signature: AI-native companies should hire engineers to maintain skills rather than function-specific staff. Does job-posting data show a "skill maintainer / agent ops" role emerging as a distinct hiring category? (Partially answered: ICONIQ, Q2 2026 confirms the composition shift the signature predicts — 45% of ~305 AI-builders plan a "different mix of roles (fewer operational, more AI-fluent talent)," function-level headcount reallocates toward R&D/Product/Sales and away from Customer Support/G&A, and G&A operators are "removing finance-ops and order-management roles… redirecting budget to strategic and AI-specific functions." It also names the concrete new engineering hiring categories: forward-deployed engineers (~50% scaling as a permanent motion) and AI safety / trust & reliability engineers. What it does not supply is the specific "skill maintainer / agent ops" title from job-posting data — ICONIQ measures function-level headcount intent and two named roles, not an occupational taxonomy. The direct test (a "skill maintainer / agent ops" posting category) still needs job-posting/occupational-emergence data, e.g. the un-ingested arXiv 2606.22769 "Agent Systems Engineer" signal from the 2026-07-21 research pass. See the restructuring section above and AI Product Economics Maturation for the FDE detail.) Checked against the largest available population instrument and still not answered (2026-09-22): The state of AI in 2026: On the road to ROI (empirical, self-reported, 1,719 respondents in 97 nations) publishes no occupational or role taxonomy at all — its workforce section forecasts headcount levels by function (39% expect an AI-related decline in the coming year) and never asks what roles are being created. The nearest row in its practice chart runs the other way: "completed a strategic workforce planning exercise" is reported by only ~24% of AI high performers and ~13% of everyone else (Exhibit 11, gridline-read, approximate), the second-narrowest gap in the chart. So at population scale the exercise that would produce a named skill-maintainer role is the practice organizations are least likely to have done, including the ones attributing EBIT to AI. The direct test still needs job-posting or occupational-emergence data. Three more names, and a negative result on the coarser instrument (2026-09-22): 2026 State of Scaling: The Great Sorting names AI Business Analyst, AI Integration Engineer and GTM AI leader as emerging titles in ICONIQ's portfolio — closer in spirit to the skill-maintainer signature than anything previously logged, since all three sit between a function and the AI systems it runs. But they arrive as an investor's uncounted qualitative observation, with no share of companies, no posting volume and no definition, so they are anecdote, not the occupational data the bullet specifies. The same report supplies the negative result at the level it can measure: functional headcount distribution is flat across four years (S&M/R&D/G&A ≈ 50/41/9 under $100M, 47/39/14 above), with ICONIQ concluding the workforce model 'has yet to meaningfully shift.' That is a coarse instrument — an intra-functional role substitution is invisible to a three-bucket chart — so it bounds rather than answers: whatever role emergence is happening is not yet large enough to move the buckets. The direct test is unchanged.
• Is the encoded-role form of the employee metaphor actually accountability-preserving, as the synthesis above suggests, or do Kropp-style framing effects attach to skill-files-as-employees too once teams talk about them that way? No study has tested framing effects on artifact-level anthropomorphism.

Sources#

• The New Physics of Business — Garry Tan, Y Combinator — Garry Tan, "The New Physics of Business," AI Engineer, 2026-07-17 (practitioner-opinion)
• Garry Tan: Own Your Intelligence — Garry Tan, "Own Your Intelligence," YC Startup School, 2026-08-06 (practitioner-opinion; auto-caption transcript): §"Building a Company of One" — the organization-of-one reframing, the audience-of-one software corollary, and the revenue-per-head exhibits restated without new evidence
• Indian AI Coding Startup Emergent Becomes a Unicorn with $130M Series C — TechCrunch (2026-07-15, vendor-claim): the third-party-checked Emergent figures behind the revenue-per-head subsection
• Engines of Growth: Global Startup Trends Report — AWS Startups, Engines of Growth (June 2026, self-reported vendor survey): the population RPE reading (55% of AI-natives report $400K+/head) that sits between Tan's tail anecdotes and Emergence's cap-table median
• State of AI 2026: The Builder's Economy — ICONIQ Growth, State of AI 2026: The Builder's Economy (2026-07-08, empirical): §"Talent & Organization" — the restructuring-by-kind survey (45% role-mix / 33% smaller-team, N=302; image_000056), function-level headcount (N=303; image_000061), cross-functional/management-layer/spans-of-control cuts, and the FDE spotlight — the population evidence behind the restructuring section and the staffing-signature partial answer
• The ICONIQ Pacesetter Index — ICONIQ Venture & Growth, The ICONIQ Pacesetter Index, 2026-09-17 (empirical, not a survey: quarterly financial and operating data 2024 – Q2 2026 from public software companies plus ICONIQ's own venture and growth portfolio). Cited here only for the revenue-per-FTE row by ARR band ($75K / $115K / $225K / $655K median; $890K top quartile at $100M+), used to benchmark the record-revenue-per-head claim against its own peer cohort. Selection caveat: a Pacesetter is selected on top-quartile three-year growth plus ICONIQ's own AI-Native/AI-Driven label, with no control group; COI: the sample is partly the publisher's portfolio. Parse note: the benchmark table exists only as a page image, transcribed and re-verified cell-by-cell at compile time; figures rounded to the nearest 5. Instrument provenance at ICONIQ
• The state of AI in 2026: On the road to ROI — Dan Tinkoff, Lieven Van der Veken & Michael Chui with Tara Balakrishnan, The state of AI in 2026: On the road to ROI (McKinsey / QuantumBlack, 2026-08-25, empirical, self-reported; online survey, 1,719 participants in 97 nations, fielded May 4 - June 8 2026, GDP-weighted). Cited here for Exhibit 11's workflow-redesign and strategic-workforce-planning rows (gridline-read from an unlabelled dumbbell chart — approximate), the prose three-quarters/one-quarter redesign figures, and Balakrishnan's absorption-capacity commentary. COI: McKinsey sells the transformation work this finding implies demand for.
• The Human-AI Substitution Principle: When will you be replaced by AI in your organization? — Banerjee & Singh, arXiv 2607.20781 (2026-07-22; practitioner-opinion, formal model, no empirical data): §4.2.1 Theorem 4 and §4.2.4 Corollary 1 (depth reduction D^(AI)* ≤ D* and the implied span widening, via the level-size identity e_i = ∏ s_ℓ), §4.1.3 Theorem 3 (super-exponential human cost escalation with depth), §5.6 Table 3 prediction P3 and §5.6.3 (its stated empirical signature). Cited here only as the theory the ICONIQ layer/span figures happen to match; full treatment and the P1–P7 ledger at Organizational Complements to AI
• Startup ARR is less secure than ever, new research shows — Julie Bort, TechCrunch, 2026-09-03 (practitioner-opinion, press reporting of two unread VC surveys): cited here only for Madrona's 77% semi-annual vendor re-evaluation figure (n=150 enterprise IT professionals) as a caveat on the durability of the ARR numerator in every revenue-per-head figure above — not as evidence about per-head levels, and not a churn measurement. Full evidence handling at Seven Powers Applied to AI
• 2026 State of Scaling: The Great Sorting — ICONIQ Venture & Growth, 2026 State of Scaling: The Great Sorting (September 2026, empirical; 52-page PDF, docling-parsed). Cited here for the AI-adoption-by-function table (p.41 — access 99/98/97%, daily users 74/86/70%, from the bolted-on n = 38 Annual Growth Operating Trends Survey, February 2026, not from the 137-company operating dataset), the functional headcount distribution by year (p.44, read off the page image in a two-pass; n = 7-8 company-quarters in the 2026 column) and the ICONIQ-network backfill observations. Selection, COI and the company-quarter n trap at ICONIQ