AI Already Killed the Modern Company. The Org Chart Just Hasn't Noticed Yet

The comet has already hit. The dinosaurs are still walking around.

That’s the uncomfortable shape of what’s happening to incumbent companies right now. The cause of death has arrived — cheap reasoning, agents that coordinate, APIs that glue any two systems together for pennies — but the visible collapse lags by years. Headcount looks fine. The org chart looks fine. Q3 looks fine. And then it doesn’t, all at once, and everyone acts surprised even though the impact landed eighteen months ago. (The framing — “AI has killed the modern company,” and “the dinosaurs didn’t go overnight” — comes from Peter Diamandis’s Organizational Singularity episode, which is the seed this post is arguing with.)

The interesting claim isn’t “AI is disruptive.” Everyone says that. The interesting claim is why the lag exists — and why the thing dying first is invisible from the inside.

The reason your company exists is quietly eroding #

Start with the economics, because that’s where the structural damage is, not the headlines.

The reason large firms exist at all — Ronald Coase’s answer from 1937 — is that coordinating work inside a company was cheaper than going to the open market for every task: every contract negotiated, every supplier vetted, every agreement enforced. Hierarchy was a transaction-cost-avoidance machine. Build the thing in-house because the meeting is cheaper than the market.

When agents can search, negotiate, and enforce agreements at near-zero marginal cost, that calculus shifts from underneath the firm. The California Management Review’s Berkeley analysis lays this out directly: AI agents undercut the Coasian rationale for the large firm, eroding “the firm’s core reason for existing” even while the headcount and the org chart look completely unchanged (CMR / Berkeley, Apr 2025). A February 2026 paper, “The Headless Firm,” pushes the argument further: autonomous agents absorb the decisions that used to require a layer of middle management, swinging the make-vs-buy decision hard toward modular, AI-intermediated arrangements — redrawing the boundary of the firm without anyone ever calling a board meeting (arXiv, Feb 2026).

This is the part builders should sit with. The damage isn’t a product getting beaten in a bake-off. It’s the load-bearing economic assumption of the org dissolving silently. Nobody schedules a meeting to announce that the reason for the meeting no longer holds.

Why incumbents can’t just “do AI” and fix it #

Here’s the obvious objection: if AI is the threat, incumbents have the most money to spend on AI. So why don’t they just buy their way out?

Because they bolt it onto the wrong thing. An MIT study published in August 2025 — 150 executive interviews, a 350-employee survey, 300 public deployments analyzed — found that 95% of enterprise generative-AI pilots delivered no measurable profitability impact. The bottleneck wasn’t model quality. It was organizational design: generic tools that don’t bend to legacy workflows, dropped into processes engineered for humans (Fortune on the MIT report, Aug 2025). NTT DATA, drawing on prior MIT research, projects a comparably grim band — that 70–85% of enterprise GenAI efforts fail to meet expected outcomes, well above the 25–50% failure rate of conventional IT projects, with the root causes being organizational rather than technical: change fatigue, trust deficits, misaligned incentives (NTT DATA, 2024). (Worth flagging honestly: NTT’s 70% anchors to a pre-generative-AI MIT study, and the 85% is a projection, not a measured outcome — treat it as a directional band, not a precise reading.)

The barrier sits at the top, not the bottom. McKinsey’s 2025 workforce report finds that leadership inertia — not employee resistance — is the binding constraint on scaling AI, with the overwhelming majority of organizations using AI in some form but fewer than one in five having pushed past pilots (McKinsey, Superagency, 2025). (Editorial note before publication: confirm McKinsey’s exact “~90% using / <20% scaled” and the leadership-inertia framing directly in the report — the page was not machine-fetchable during drafting.)

This is the dinosaur lag, mechanized. The incumbent isn’t ignoring AI. It’s adopting AI — into structures that guarantee the adoption does nothing. The cause of death and the appearance of health coexist precisely because the spending is real and the restructuring isn’t.

What the lag looks like when you watch it live #

The cleanest live case is Cursor versus GitHub Copilot.

Anysphere built Cursor — an AI-native code editor — from scratch starting in 2022: no legacy IDE codebase to protect, no installed base to defend, no enterprise-procurement scar tissue. Copilot had every incumbent advantage that matters: first-mover timing, Microsoft’s distribution, and a giant OpenAI investment behind it.

Watch the trajectory. Cursor’s earliest confirmed ARR figure is north of $500M at its Series C in June 2025, then $1B at Series D in November 2025, then $2B+ by February 2026 — the company quadrupling in roughly eight months (Panto / Bloomberg, 2026). By mid-2025 Cursor reported use by over half the Fortune 500; its enterprise page later put the figure at 64% — inside the same large companies where Copilot held the home-court advantage.

Now the honest part, because the disruption story is usually told dirtier than the data supports. Copilot is not losing. A January 2026 JetBrains survey has Copilot still leading developer-at-work usage at 29%, with Cursor at 18% and climbing (same source). The incumbent product is alive and ahead on share. What’s moved is the slope: the AI-native newcomer compounding faster from a standing start than the funded incumbent with the head start. That’s the whole thesis in one chart — the hit has landed, the body is still upright, and the gap that matters (rate of climb) is widening, not closing.

The honest counter-case #

If this only had one side it would be propaganda, so here’s the other side, and parts of it are strong.

Incumbents have real, structural moats — and some will win. Proprietary data at scale, existing customer trust, regulatory relationships, distribution. Microsoft embedding Copilot across Office 365’s roughly 1.2 billion users is a distribution advantage no startup can replicate by building a better product. “AI-native” is not a force field.

Augmentation, not extinction — so far. A 2026 HBR piece argues that firms choosing augmentation over wholesale automation may outperform over the long run: Aon, for one, invested in digital-fluency training instead of headcount cuts and reported productivity gains without layoffs (HBR, Apr 2026). The measured productivity gains from generative AI to date are real but bounded — this looks far more like a steep curve than an overnight cliff for the modern firm.

The “headless firm” can cut the other way. The same Berkeley analysis that undermines Coase also warns that agents can raise organizational transaction costs at the macro level: departments each deploying their own agent stacks breed duplicated effort, conflicting processes, and platform lock-in (CMR / Berkeley, Apr 2025). The dinosaur may adapt before it dies.

And the AI layer itself re-centralizes power. The hyperscalers building and distributing AI are incumbents. Platform dependency pulls toward re-centralization at the same time agents pull toward decentralization — the firms that own the model layer have arguably gotten stronger, not weaker.

And there’s a deeper pattern that should make any AI-native triumphalist nervous: incumbents have survived this movie before. The music industry waved off MP3s because early files sounded worse than CDs and appealed only to fringe users; labels “found it extremely difficult to accept the need for radical rethinking of their business model.” McKinsey’s disruption guide documents the identical “it is not happening to us” story across newspapers, travel agencies, and retail (McKinsey, Incumbent’s Guide). (Editorial note: verify the MP3 / music-industry passage and the cross-industry examples directly in the McKinsey piece before publishing — it wasn’t machine-fetchable during drafting.) The cause arrived years before revenues collapsed — which validates the lag thesis. But “incumbents usually lose to disruption” and “incumbents always lose” are not the same sentence, and the second one isn’t true.

So what do you actually do with this #

If you build, the takeaway isn’t “incumbents are doomed,” because that’s both unfalsifiable and useless. The takeaway is more specific and more uncomfortable:

The thing that dies first is invisible from inside the company — it’s the economic assumption underneath the org chart, not a line on the P&L. By the time it shows up in revenue, the decision window has been closed for a year. So if you’re inside an incumbent, stop measuring your AI program by whether you’ve adopted tools and start measuring whether you’ve restructured the work — because the MIT and NTT numbers say adoption-without-restructuring is the default, and the default fails. And if you’re building the AI-native challenger, the lesson from Cursor is that you don’t win by killing the incumbent’s product. You win by compounding faster from zero while the incumbent’s slope stays flat — and then you let the lag do the rest of the work for you.

The comet doesn’t ask whether you’ve noticed. It already hit.