Palantir's Bet Against Management

Palantir is building exertion designed to illness the region betwixt what organizations cognize and what they tin do. The larger consequence whitethorn beryllium a redesign of guidance itself.

AFP via Getty Images

In 1924, General Motors had a problem that would sound almost absurd to a CEO today: it didn't really cognize what was happening successful its ain business.

Cars could beryllium piling up connected trader tons while executives successful Detroit were still moving from weeks-old information. Alfred Sloan, who led GM from 1923 to 1946, later recalled that GM sometimes “knew thing astir the astir caller 5 aliases six weeks” of its car sales. By the clip those executives understood what customers were really buying, the marketplace had already moved.

To onslaught that problem, GM created a strategy of making accusation move faster. Its automobile divisions began requiring reports from dealers each 10 days: caller orders, orders waiting to beryllium filled, and caller and utilized cars sitting connected lots. Executives could spot request sooner. Production could set faster. Decisions that had been made done weeks of uncertainty could now beryllium grounded successful accusation simply days old.

Ten days. That was a breakthrough.

But present is the portion of the GM communicative that relates to today. When accusation is slow, organizations request infrastructure to move it. Infrastructure meaning they needed group to cod it, group to reconcile it, group to move it into reports, managers to construe those reports, and further layers to move accusation upward and decisions backmost down.

Over time, overmuch of that infrastructure became what we simply telephone management.

That doesn't mean guidance exists only to move information. Managers create people, resoluteness conflict, workout judgment, build spot and make decisions nether uncertainty. But a meaningful stock of the firm level we inherited was built astir a applicable problem: the group making decisions could not spot what was happening accelerated enough.

Today, that problem is disappearing.

Information that erstwhile took weeks, and later days, to scope decision-makers tin progressively beryllium reconciled, analyzed, and acted upon successful seconds. If companies nary longer request arsenic galore group and layers simply to cod information, construe it, move it upward and nonstop decisions backmost down, what happens to the statement built astir doing precisely that?

That is not chiefly a jobs question. It is an organizational-design question. If AI changes really accusation moves and decisions get made, the rumor isn't simply really galore group a institution needs. It is really the institution itself should beryllium built.

Now put that problem wrong a institution today, a period later.

Monday, 8:02 a.m.

A divisional executive opens the 47-slide operating platform her squad spent Friday assembling. There is only 1 problem: the strategy already knew everything successful it.

Before anyone opened PowerPoint, it had reconciled the weekend's results against inventory, flagged 3 anomalies, traced 1 to staffing and different to a supplier, modeled imaginable responses, executed the moves it was authorized to make, and put 2 decisions successful beforehand of her that really required a human.

Her gathering is shorter. Her squad is smaller. The institution is faster.

So what's the problem?

She hasn't mislaid her job. She's mislaid the activity that taught her really to do it.

She learned the business by building those decks. She learned which numbers mattered, which explanations didn’t clasp up, wherever operating problems hid, and which questions elder executives asked erstwhile the numbers went wrong. She was moreover promoted 3 times because she became exceptionally bully astatine turning accusation into judgment.

Now, accusation arrives already assembled, analyzed, and progressively accompanied by a recommendation.

That leaves 3 questions that are overmuch harder than whether AI will destruct jobs.

If guidance layers were built partially to lick an accusation problem, what happens erstwhile machines lick overmuch of that problem differently?

1. If AI absorbs the activity done which early managers learned, wherever will the adjacent procreation create judgment?

2. And, if AI radically lowers the costs of coordinating information, group and decisions, really overmuch of the corp we inherited still needs to beryllium astatine all?

3. Most of the AI-and-work statement starts pinch the worker: Which jobs tin a instrumentality perform?

I deliberation that's the smaller question.

The bigger 1 starts pinch the organization: Which parts of the institution existed because humans couldn’t process information, coordinate activity, and make decisions efficiently capable without them?

This is wherever Palantir becomes worthy watching, and not for the logic it usually gets covered.

OpenAI, Anthropic, and Google compete complete intelligence. Palantir occupies different ground: organizational context, past decision, past action. Its Foundry, AIP, and Apollo stack is marketed arsenic an endeavor operating system, but the load-bearing conception is the Ontology — an effort to correspond an organization's objects, relationships, logic, permissions, and actions successful a shape some humans and machines tin run against.

For decades, endeavor package helped organizations grounds themselves. ERP recorded what a institution owned. CRM recorded what customers did. Business intelligence explained what had happened. Generative AI answered questions group already knew to ask. Palantir is attempting thing structurally different: to make capable of the statement computationally understandable that humans and machines tin enactment against the aforesaid operational reality.

Which produces the thesis worthy arguing about.

Palantir’s astir consequential stake whitethorn not beryllium that AI tin switch workers. It is that AI tin sorb capable accusation processing, coordination, and regular decision-making to alteration really overmuch accepted guidance an statement needs.

So, AI whitethorn flatten the corp earlier it empties it.

Nobel Prize-winning economist Ronald Coase famously asked why firms beryllium astatine all. One answer: immoderate activities are cheaper to coordinate wrong a institution than done the market. If AI radically lowers the costs of coordination, it could yet alteration not conscionable really galore guidance layers a institution needs, but really overmuch of the institution needs to stay wrong the institution astatine all. That is simply a larger mobility than this portion tin settle. The disruption wrong the org floor plan is already visible.

AI Is Coming for the Org Chart

Two bodies of grounds are pointing successful what look for illustration other directions. Read together, they picture thing much unsettling than either does alone.

At the bottom, 2 ample datasets constituent successful a akin direction. Stanford's Digital Economy Lab, moving pinch ADP payroll information covering millions of workers, finds employment among 22-to-25-year-olds successful AI-exposed occupations moving astir 19% beneath wherever it would beryllium had it tracked less-exposed peers — pinch nary comparable spread for knowledgeable workers. Separate résumé-based investigation covering tens of millions of workers finds inferior employment declining astatine firms that adopt generative AI while elder employment holds steady. Both sets of authors are observant to picture these arsenic patterns alternatively than causal estimates.

The system matters much than the magnitude. In some cases the accommodation runs chiefly done reduced hiring alternatively than separations.

Companies are not firing their apprentices. They are softly nary longer creating apprenticeships.

That favoritism is the full argument, because entry-level activity was ne'er only production. Building the model, moving the analysis, drafting the document, preparing the recommendation, assembling the operating reappraisal — these were the mechanisms done which group learned really an institution really worked. The output was the visible product. The acquisition was the point.

In the middle, the image is noisier but harder to dismiss. An study Revelio Labs conducted for Business Insider recovered that, astatine the white-collar hiring trough, openings were down 43% for mediate managers and 57% for elder leaders against the 2022 roar — compared pinch 14% for inferior roles. Gartner, successful strategical predictions issued successful precocious 2024, forecast that done 2026 1 successful 5 organizations would usage AI to flatten their structures, eliminating much than half of existent middle-management positions. That forecast model closes this year. The prediction is worthy treating arsenic a test, not a finding. The firm examples are not subtle: Citi targeted a simplification from thirteen guidance layers to eight, and UPS eliminated astir 12,000 positions during 2024. Amazon separately moved successful 2024 to raise its ratio of individual contributors to managers by astatine slightest 15%, and announced astir 14,000 firm occupation cuts the pursuing year.

An honorable caveat belongs here. Not 1 of those decisions tin beryllium cleanly attributed to artificial intelligence. Post-pandemic overhiring, liking rates, and mean costs subject explicate a awesome woody of it. The defensible declare is narrower and, I think, much interesting: AI is arriving precisely erstwhile corporations were already questioning really overmuch level they need. It has the imaginable to move a cost-cutting manner into a structural redesign.

The Rise of the Hollow Company

For astir of the past century, companies developed executives the aforesaid way: inferior worker, knowledgeable contributor, manager, elder leader. Each shape taught thing the adjacent 1 required. Junior activity was apprenticeship disguised arsenic production. Management was judgement training disguised arsenic coordination.

AI now exerts unit connected some stages astatine once. At the bottom, it absorbs the accumulation activity that doubled arsenic apprenticeship. In the middle, it absorbs the accusation activity and coordination that justified the managerial layer.

The near-term economics look excellent. Productivity rises. Layers fall. Spans widen. SG&A drops. Decisions accelerate. Revenue per worker improves. Every 1 of those numbers will look bully successful a committee deck.

What they will not show is the point being consumed to nutrient them.

The Hollow Company: an statement pinch blase executives astatine the apical and bonzer intelligence throughout, but a steadily weakening system for producing the humans tin of starring it next.

Companies tin amended today's margins while liquidating tomorrow's activity bench. The institution eliminating managers successful 2026 still needs leaders successful 2030, and it whitethorn person spent 4 years dismantling the developmental strategy it took for granted.

There is simply a second-order problem layered connected top, and it explains why this is truthful difficult to hole from inside.

The managers AI makes little basal successful their existent shape are the aforesaid managers companies request successful bid to instrumentality it successfully. Their hesitation is not technophobia. It is arithmetic. Asking a head to deploy a strategy that eliminates accusation bottlenecks, widens spans, and pushes authority toward the beforehand statement is asking her to automate the circumstantial activities that warrant her position. Rational group move slow nether those conditions.

That is simply a activity problem wearing an IT problem's clothing, and it is why organizational translator will beryllium harder than technological implementation astatine astir companies.

What Palantir Is Really Betting On

Palantir's durable advantage whitethorn not beryllium its AI astatine all.

The institution allows customers to tally aggregate underlying models and bring their own. That is simply a revealing creation choice. It suggests the scarce plus isn't the intelligence — it's the organizational discourse surrounding the intelligence. Intelligence, past context, past decision, past action, past feedback. The strategical mobility is not who has the smartest model.

Who owns the furniture done which intelligence becomes organizational action?

Customers look progressively consenting to salary for that proposition. Palantir's 2nd 4th of 2026 produced $1.94 cardinal successful revenue, up 93% twelvemonth complete year, pinch U.S. commercialized gross up 149% to $764 cardinal and remaining U.S. commercialized woody worth much than doubling to $6.24 billion. Valuation is simply a abstracted argument. Revenue astatine that trajectory is difficult to disregard arsenic experimental spending.

None of which intends Palantir has cornered anything. Microsoft has distribution nary 1 tin lucifer and is pushing agents into the aboveground wherever activity already happens. ServiceNow occupies workflow territory that matters enormously successful the agentic era. Salesforce, SAP, and the hyperscalers are each moving toward related ground. Everyone will person agents. The harder mobility is who tin build a trustworthy operational practice of really a consequential statement really runs and past safely fto machines enactment wrong it. Palantir spent 2 decades moving connected versions of that problem crossed defense, intelligence, manufacturing, healthcare, and proviso chains — environments wherever fragmented accusation and consequential decisions are the normal condition.

The skeptic's lawsuit deserves its paragraph. Is the Ontology a moat aliases an bonzer switching costs dressed arsenic one? How overmuch of the worth depends connected Palantir's forward-deployed engineers alternatively than the software? And what governance consequence is created erstwhile an expanding stock of a corporation's consequential decisions flows done a azygous computational layer?

But location whitethorn beryllium a deeper coherence present worthy naming. Palantir's products many times onslaught the aforesaid target: fragmentation, delay, bureaucratic handoffs, and the region betwixt knowing thing and acting connected it. That suggests the institution did not simply build endeavor software. It built exertion astir a mentation of institutions — that organizations go weaker erstwhile accusation is fragmented, authority is diffuse, and execution is slow.

Read that way, Palantir whitethorn not beryllium betting against managers astatine all. It whitethorn beryllium betting against the organizational clash that made truthful overmuch guidance basal successful the first place.

The Apprenticeship AI Could Rebuild

Here is the portion I find genuinely surprising.

If activity improvement tin nary longer tally done production, the reply cannot beryllium preserving obsolete tasks because erstwhile generations happened to study from them. Nobody should build a financial exemplary by manus successful 2030 for pedagogical reasons.

Development has to move from accumulation apprenticeship to judgment apprenticeship — learning by interrogating the machine's recommendation, stress-testing its assumptions, identifying the discourse it lacks, comparing alternatives it discarded, studying exceptions, deciding erstwhile to override it, and owning the outcome.

And judgement apprenticeship requires thing organizations person ne'er really possessed: a grounds of judgment.

An operating furniture that captures what the statement knew, what was recommended, which alternatives existed, what the quality chose, what action followed, and what really happened is producing a longitudinal dataset of organization decision-making. Companies person ever had this successful fragments — successful memory, successful warfare stories, successful the intuition of group who were there.

The astir valuable firm training database of the AI era whitethorn not incorporate courses. It whitethorn proceed decisions.

Palantir's ain archiving goes partway there. It describes an Ontology that captures determination lineage — erstwhile a determination was made, against which type of the endeavor data, done which exertion — and discusses aggregated decisions becoming worldly that improves the capacity of some group and agents complete time. What remains an conclusion is the quality half: that specified a grounds could go the system ground for processing executive judgment, not only for tuning the system. That is an accusation of the architecture, not a merchandise being sold. But it is the astir absorbing point the architecture makes imaginable: the infrastructure disrupting the aged apprenticeship could besides go unusually powerful infrastructure for building a amended one.

It besides has a acheronian twin, and it needs to beryllium named earlier personification other names it.

A imperishable grounds of organizational judgement is besides the astir blase guidance surveillance strategy a corp has ever had. Reconstruct what a head knew, erstwhile she knew it, what the instrumentality advised, what she chose instead, and what happened — and you person built a capacity reappraisal pinch cleanable hindsight attached.

The pathology writes itself. A head who knows each deviation from the machine's proposal will beryllium permanently reviewable does not make amended decisions. She makes much defensible ones. The statement slow optimizes for auditability and calls it discipline.

Decision lineage should create judgment, not constabulary it. That requires existent rules — who tin entree the record, whether it enters capacity evaluation, really hindsight bias is controlled, and whether managers clasp protected abstraction to disagree pinch a instrumentality and beryllium wrong. Companies that deploy the capacity without answering those questions will get the surveillance strategy and suffer the learning system.

Meanwhile, the occupation itself changes shape. As AI absorbs reporting, monitoring, accusation movement, regular coordination, and predictable escalation, what remains is judgment, ambiguity, trust, persuasion, conflict, culture, ethics, motivation, talent development, and accountability.

The much guidance AI tin perform, the much quality the remaining occupation of the head becomes. Management doesn't disappear. Mediocre guidance becomes overmuch harder to justify.

That is AI's organizational past mile. Palantir whitethorn beryllium building an operating strategy for the AI-native enterprise. But package cannot renegotiate authority, redesign careers, create trust, create judgment, aliases determine what humans should stay accountable for. Somebody still has to build the quality operating strategy astir it — determination rights, incentives, spans, learning, accountability. Technology redesigns workflows faster than institutions tin renegotiate immoderate of that. Build the first without the 2nd and you person simply automated yesterday's bureaucracy astatine sizeable expense.

Diagnosis is cheap. Here is what to measurement earlier the adjacent committee meeting.

Information time. Audit 1 typical week of your elder managers' calendars. What percent went to obtaining, reconciling, packaging, and transmitting accusation alternatively than exercising judgment?

Decision distance. Take 1 consequential operational determination and trace it from the first observable frontline awesome to authorized action. Count the humans it passed through. Not the org floor plan — the existent path.

Reporting latency. Take your past meaningful antagonistic surprise. Find the day the underlying information made it knowable. Compare it to the day personification empowered to enactment recognized it. Then inquire the mobility that should support you up: how agelong did your past bad 4th beryllium wrong your systems earlier anyone realized you had a problem? A period ago, 10 days was revolutionary. What is your number?

Human determination rights. List the 10 highest-consequence recurring decisions successful the enterprise. Classify each: AI recommends, AI executes aft approval, AI executes autonomously, quality determination required. If your activity squad cannot complete that workout earlier agentic systems arrive, governance is already down implementation.

Monday, 8:03 a.m.

Return to the executive we discussed earlier.

She learned the business by building the deck. Her younger workfellow ne'er will. He is sitting beside her, not assembling accusation but interrogating a recommendation. He thinks 1 of its assumptions is wrong. She asks why. He explains. They reason astir it for 4 minutes. She decides. He watches what happens next, and truthful does the system.

She learned guidance by producing information. He is learning it by challenging intelligence.

That is accumulation apprenticeship becoming judgement apprenticeship, and it is astir apt amended training than she got.

It is besides training for less people. Twelve group erstwhile participated, badly, successful preparing that operating review. Two now participate profoundly successful the decision. Better apprenticeship for a narrower chimney is not a solved problem. It is simply a different one, and it belongs connected the schedule of each CEO who is astir to observe a flatter org chart.

Across town, a competitor is opening descent one.

General Motors transformed guidance by compressing an accusation spread measured successful weeks toward 1 measured successful days. Palantir is betting that AI tin push meaningful parts of that spread toward seconds. The institution that collapses a seven-day reporting rhythm to 7 seconds is not 30% much productive. It is operating connected a different clock. By the clip that advantage shows up successful a competitor's margins, you whitethorn already beryllium 2 organizational redesigns behind.

Palantir's stake against guidance whitethorn yet beryllium thing much eager than a stake against managers: a stake that companies tin yet abstracted activity from bureaucracy.

For a century, we paid for guidance partially because accusation was scarce, fragmented, and slow. If that stops being true, the mobility is not whether guidance survives AI.

It is what guidance is for erstwhile accusation nary longer needs a head to move it.