Across 114 AI companies in our dataset there are 9,303 active job postings. About 14.5% of those — roughly one in seven — are explicit management titles: Engineering Manager, Director of Sales, VP of Operations, Chief of Staff. In the engineering function specifically, the rate is 8.6%. In research, 3.3%. At the nine pure AI-model labs in our data — the companies that only train models and do not sell applications or infrastructure — it is 3.4% across all functions combined.
The same dataset says something else. Operations hiring is 37.4% management titles. Finance is 35.0%. At the 42 companies that build AI applications without models or infrastructure underneath them, management titles are 16.0% of all hiring. At the 22 pure AI-infrastructure companies, 18.2%. Inside the same industry, two very different companies are being built.
A recent essay by Jack Dorsey and Roelof Botha on the Sequoia Capital site, "From Hierarchy to Intelligence," argues that AI is making middle management obsolete. The essay traces organisational design from the Roman contubernium through the Prussian General Staff to the modern functional pyramid, and argues that at Block — Dorsey's company, 9,000 people before a February 2026 reorganisation that removed roughly 40% of headcount — the structure is being rebuilt around three roles: Individual Contributors, Directly Responsible Individuals, and Player-Coaches. "There is no need for a permanent middle management layer," they write. "Every company will eventually need to confront the same question we did."
The argument is made about Block, generalised to every company. Our data covers neither of those. What it does cover is 114 AI companies — the firms furthest along in using AI internally, and the ones for which the "intelligence replaces hierarchy" claim should be most visible. Here is what we found.
The management layer looks very different depending on what a company sells
The Dorsey–Botha argument lives or dies on a single claim: that AI-native companies are genuinely organising without middle managers. If they are, it should show up in their hiring. It does — but only in one part of the AI industry.
We split the 116 published companies in our dataset by what they build. A company that trains its own models is flagged builds_ai_models. One that sells infrastructure — compute, orchestration, data platforms — is flagged builds_ai_infrastructure. One that ships AI-powered applications is flagged builds_ai_applications. Most companies wear more than one hat. But there are 73 companies that fall into only one category. Their hiring patterns are not subtle.
| Segment | Companies | Active jobs | Mgmt title share | ICs per manager |
|---|---|---|---|---|
| Pure Models | 9 | 177 | 3.4% | 28.5 |
| Models + Apps | 24 | 744 | 9.7% | 9.3 |
| Models + Infra | 10 | 2,586 | 13.5% | 6.4 |
| All three | 5 | 470 | 12.3% | 7.1 |
| Pure Applications | 42 | 2,752 | 16.0% | 5.3 |
| Pure Infrastructure | 22 | 2,154 | 18.2% | 4.5 |
The pure model labs — Thinking Machines Lab, Reflection, Physical Intelligence, World Labs, Liquid AI, Black Forest Labs, Skild AI, and two others — hire about one manager for every 28 individual contributors. Pure application and infrastructure companies hire one for every five. That is a roughly five-fold difference in hiring-measured flatness, inside the same industry, in the same month.
This is consistent with the Dorsey–Botha claim in the narrowest possible reading: companies whose entire product is a research artefact — a model — hire almost no managers. Those companies can run on small groups of senior researchers and engineers because the work they do is legible to other researchers, the organisational coupling is low, and they don't have paying customers to coordinate around. Seven of the nine pure model labs are under 50 active postings. These are not companies at Block's scale. They are small research groups with enough venture funding to hire without much GTM friction.
The moment a company starts shipping products, the picture changes. Pure application companies — Block among them, but also Harvey, EliseAI, Decagon, Sierra, Perplexity — hire at a managerial density five times higher than the labs. Pure infrastructure companies — CoreWeave, Nebius, Crusoe, Vanta, Databricks' infrastructure-heavy peers — are higher still. At that point in the industry, the hierarchy is not disappearing. It is staffing up.
The hierarchy lives in sales, operations, and finance — not in engineering
The essay is mainly about technical work. Dorsey describes an "intelligence layer" that "composes capabilities into solutions for specific customers at specific moments and delivers them proactively." The picture is of an engineering org where the world model absorbs what used to be coordination overhead. Our data on engineering specifically is broadly consistent with that picture.
| Function | Active jobs | Mgmt title share |
|---|---|---|
| Research & Science | 451 | 3.3% |
| Physical Systems | 345 | 3.2% |
| Product | 292 | 4.1% |
| Design & Creative | 194 | 6.2% |
| Engineering | 2,534 | 8.6% |
| Data & Analytics | 180 | 10.0% |
| Security | 310 | 10.0% |
| Sales & GTM | 2,221 | 16.6% |
| Marketing | 393 | 17.0% |
| People & HR | 359 | 18.4% |
| Customer Support | 562 | 20.5% |
| Finance | 283 | 35.0% |
| Operations | 650 | 37.4% |
Across 2,534 active engineering postings at the 114 companies in our dataset, only 8.6% carry a management title. Research hiring is 3.3% management. The technical core of an AI company, in April 2026, is thin on middle management to begin with.
But engineering is 27% of the hiring. The other 73% is a different story. Operations hiring is 37% management titles. Finance is 35%. Customer Support is 20.5%. Sales & GTM is 16.6%. These are the functions that actually coordinate with external parties — customers, regulators, contracts, procurement, facilities, physical infrastructure — and they are hierarchically staffed. The same dataset that shows 3.3% management in research shows 37.4% in operations. These are the same companies.
The Dorsey–Botha essay argues that AI's company world model and customer world model will absorb the information-routing work that middle management exists to do. In engineering, that coordination work is already much less hierarchical than the essay implies — a small fraction of the dataset's management roles. Where the hierarchy actually lives is in the functions where work has to clear human-to-human handoffs that a model cannot yet stand behind: a signed contract, a regulatory filing, a capacity agreement with a data-centre landlord. That is where managers are being hired, and the essay has less to say about those functions than about the engineering one it is mostly written about.
Block's own hiring is not the three-role company
Block is in our dataset. It ships AI applications; it does not train frontier models; it does not sell infrastructure. It is classified Pure Applications, the segment the three-role argument claims to represent. Its current hiring is a useful stress test.
Block has 140 active postings. Of those:
- 114 are Sales & GTM — Account Executives, Partner Managers, Field Sales Managers, Business Development Managers across territories from SMB inbound to enterprise strategic accounts.
- 9 are Customer Support.
- 5 are Operations.
- 2 are Engineering.
- 0 are Product.
- 0 are Research.
Thirteen of the 140 postings — 9.3% — carry an explicit management title. Every one of the thirteen is in Sales & GTM: Head of Enterprise Sales, Head of Channel Sales Enablement, Manager of Field Sales (four of those), Inbound Sales Manager, SMB, and so on. The technical org at Block is, in hiring terms, currently hollow: two engineering postings and zero for product or research. The go-to-market org is expanding and management-heavy.
Two readings of this are possible. The first is that Block's post-February restructuring has genuinely compressed its technical hierarchy — that the engineering team no longer needs engineering managers because the intelligence layer has absorbed that work, and that we are looking at the after-picture of the essay's argument. The second is that Block, in April 2026, is simply not hiring engineers. The 2024 earnings reports describe a company that shrank engineering headcount substantially before writing the essay, not one that replaced managerial work with a model. We cannot distinguish between those two stories from the hiring data alone. What we can say is that Block's visible hiring activity is concentrated entirely in the traditionally-hierarchical function — sales — and that the management titles it is posting are the ordinary ones.
The essay acknowledges that Block is "in the early stages of this transition." Our data supports that caveat: in April 2026, Block's hiring is not yet recognisably the three-role company the essay describes.
Where it gets complicated
Our dataset describes AI companies specifically — 114 with active postings, primarily venture-backed startups and public companies with significant AI operations. It does not describe the broader economy. The Dorsey–Botha essay makes a universal claim ("every company"); our data can test whether AI companies themselves look like they are moving in that direction, not whether the claim holds for pharmaceutical companies, banks, hospitals, or local governments.
There are two methodological caveats readers should weigh. First, we are counting job postings, not filled positions. A posting is a hiring intention at a point in time. It says something about where a company is adding capacity, not about the shape of its existing workforce. A company could in principle be hiring 20% managers while its existing workforce was restructuring to 2% managers. The data cannot rule that out, only note that the postings don't show the transition.
Second, seniority is inferred from job titles, and title conventions are not uniform. Some companies title senior ICs as "Lead" or "Staff" without implying management. Others call people-managers "Principal." We flagged as management any title containing Chief, VP, SVP, EVP, Director, Head of, or Manager/Managing, then excluded common IC "manager" titles — Product Manager, Program Manager, Project Manager, Account Manager, Partner Manager, and others that convention-driven analysis would miss. The net likely undercounts management at companies that use flat titles and overcounts at companies that attach "Manager" liberally to IC roles. The aggregate pattern — that pure model labs hire at one end of the distribution and pure application and infrastructure companies hire at the other — is robust to reasonable changes in the filter.
A third caveat sits outside the data. Our snapshot covers active postings in April 2026. The Dorsey–Botha thesis is forward-looking: the world models they describe are in early deployment. Our data is a measurement of present hiring intent, not a prediction of future structure. If the company world model and customer world model become capable of absorbing meaningful coordination work over the next two years, the hiring pattern will shift. Whether it does, and how fast, is an empirical question that will show up in posting compositions over the coming months.
What the data supports, and what it complicates
1. At pure model labs, management is close to absent from hiring. Nine companies — Thinking Machines Lab, Reflection, Physical Intelligence, World Labs, Liquid AI, Black Forest Labs, Skild AI, and two others — together have 177 active postings, of which 3.4% carry management titles. One manager for every 28 ICs. This is the strongest support in our data for the Dorsey–Botha thesis, and it matches the essay's Roman-contubernium framing: small, flat, cohesive research groups.
2. Engineering and research hiring across AI companies is already flat. Across 2,534 engineering postings and 451 research postings, management share is 8.6% and 3.3% respectively. This means the essay's claim that "AI replaces middle management" has less slack to run in engineering than it implies. The middle-management layer in technical functions at AI companies in April 2026 is already thin, with or without a world model.
3. The hierarchy at AI companies lives outside engineering, in the functions that coordinate with external parties. Operations is 37.4% management titles. Finance is 35.0%. Customer Support is 20.5%. Sales & GTM is 16.6%. The essay's argument has less to say about these functions, and the data says the middle management it wants to eliminate is mostly there, not in the engineering org it mostly describes.
4. What a company builds shapes how hierarchically it hires — by a factor of five. Pure model labs hire one manager for every 28 ICs. Pure applications companies hire one for every five. The phrase "AI-native company" obscures a real structural split: the companies whose product is a research artefact look almost nothing like the companies whose product is a platform customers pay to use. "Every company" is not a useful unit of analysis for this argument.
5. Block's visible hiring is not the three-role company the essay describes. 140 active postings, 81% Sales & GTM, 13 management titles — all in sales. Block may be in transition, but the transition has not yet reached its visible hiring. The strongest available test of the essay's thesis — Dorsey's own company, in the month the essay was published — does not show the structure the essay says it is building.
The Dorsey–Botha essay makes a specific, testable claim and presents Block as its proof point. Our data cannot evaluate Block's internal org chart, but it can evaluate whether the broader AI industry is visibly moving in the direction the essay says it will. The honest answer is: in one corner of it, yes — at small pure research labs, which were already flat. Everywhere else, the hierarchy the essay says is dissolving is where most of the hiring is happening. The intelligence layer has not yet absorbed the coordination work at the company scale where most AI work is now being done.
This article is a commentary on "From Hierarchy to Intelligence" by Jack Dorsey and Roelof Botha, published on the Sequoia Capital site in March 2026. All Applied Methods data is from the dataset as of 18 April 2026 and reflects active job postings at time of analysis. Seniority is inferred from job titles, with explicit exclusions for common IC titles containing "Manager" (Product Manager, Program Manager, Project Manager, Account Manager, and similar); the count will undercount management at companies that use flat titles and overcount at companies that attach "Manager" to IC roles. The dataset covers 114 AI companies with active postings — primarily venture-backed startups and public companies with significant AI operations — and does not describe the broader economy. All roles mentioned can be explored at appliedmethods.ai.