March 2016, Seoul. Demis Hassabis watches his program AlphaGo beat Lee Sedol, world champion of Go. Within months, DeepMind hires a group of biologists and builds an interdisciplinary team to tackle an unsolved problem: predicting the 3D structure of proteins from amino acid sequences. In 2020, AlphaFold solves it. In 2024, Hassabis and John Jumper win half the Nobel Prize in Chemistry.
The people who solved protein folding were the same AI researchers who spent years teaching a machine to play a board game. They transferred their method to a domain they didn't know, studied the literature, brought in biologists. As Hassabis told Scientific American: "Something we do uniquely well at DeepMind is mix that together, not just machine learning and engineering." The core competence is transferable. The domain changes.
Hidden families
Cross-referencing O*NET's 18,796 task descriptions with Anthropic's AI penetration data reveals a pattern. The tasks where AI shows no trace cluster into eight families that cut across hundreds of professions: strategic planning and decision-making (2,100 tasks across 600 professions), judgment and quality evaluation (1,600 tasks, 550 professions), context reading and research (1,300 tasks, 470 professions), compliance and governance (1,000 tasks, 370 professions), advisory and client relationships (900 tasks, 320 professions), supervision and people development (680 tasks, 330 professions), stakeholder coordination (660 tasks, 330 professions), negotiation (420 tasks, 240 professions).
The overlap between distant professions is striking. An industrial engineer and a film director share 92% of these skill categories. A CEO and an art director, 91%. A financial manager and a logistics director, 91%. Strip away the productive component that makes them different, and the core is the same: coordinate, evaluate, negotiate, read context, decide, supervise.
The 20th-century production system treats this core as useful but secondary to the "real" function of the role, which is productive output. Sangeet Paul Choudary frames the shift well in Reshuffle: AI does not subtract something and leave a residue. It restructures the system, and value reorganizes around what becomes scarce. When productive components become abundant because AI executes them at decreasing cost, the components of advisory, judgment, coordination, and context reading become scarce. What the system treated as accessory becomes the value core.
Why departments exist
The functional org chart (marketing, engineering, finance, HR) responds to a specific problem: coordinating people with different productive skills is expensive. When producing a campaign requires shared tools, languages, and routines, grouping those who have them makes sense. The marketing department exists because the productive function holds it together.
But that productive glue is separating valuable skills that would be more useful combined. The person in marketing who reads client context and the engineer who evaluates technical trade-offs share a skill: judgment. They work in separate silos and meet in cross-functional meetings, which exist to compensate for the separation the structure itself creates.
AI compresses the constraints that justified that structure. Coordination costs drop: an AI system can orchestrate workflows across people and processes with near-zero friction. Productive specialization costs drop: AI puts the capacity of an engineer, a designer, an analyst into a single system accessible to anyone with sufficient judgment. Span of control expands: one professional augmented by AI can direct processes that previously required an entire team.
What changes
When AI absorbs the productive component of marketing and engineering, the reason those people sat in separate departments disappears. In marketing, someone who reads what clients need and translates it into strategic direction remains. In engineering, someone who evaluates technical constraints and communicates them to decision-makers. In finance, someone who reads numbers and turns them into operational choices. These people share more valuable skills with each other than with their departmental colleagues.
Teams form around problems. A team is the people who can read a specific market, judge an opportunity, and coordinate a response. AI provides the productive capacity: if the problem needs a campaign, it produces one; if it needs a software prototype, it generates one; if it needs a financial model, it builds one. People contribute relationships, experience-calibrated judgment, stakeholder coordination, contextual reading that requires presence and cultural sensitivity.
Roles become fluid. The same person can work on a pricing problem in the morning and a product design problem in the afternoon, because the valuable skill is transferable. The DeepMind team applies deep methodological skill across domains. A professional who can negotiate, read context, and evaluate quality can do the same, with AI filling the productive gap.
Hierarchy transforms. A marketing director and a CTO make sense where marketing and technology are separate functions with distinct production processes. If production is automated and teams organize around problems, you need people who compose the right team for the right problem, allocate attention (the scarce resource), and judge solution quality. The role resembles a director more than a function head.
Size changes meaning. Today, company size corresponds to productive capacity: more people, more output. If productive capacity comes from AI, size corresponds to the relationships, context, and judgment the organization can mobilize. A 10-person agency with deep relationships in three sectors can match the output of a 200-person agency, because AI closes the productive gap. But it operates in those sectors only because it has people with years of relational experience in each. Relationships do not scale with technology. They scale with time.
Three possible forms
If the functional department loses its founding principle, what takes its place?
The organization as value flows. Departments disappear. In their place, flows that run from client problem to delivered solution, with people and AI allocated based on what's needed. You don't exist as "marketing manager" but as a node with a value skill profile: you read a certain type of context, hold relationships in a certain sector, your judgment is calibrated on a certain type of decision. The org chart (who reports to whom) gives way to a flow map: who works on what, with which combination of people and AI, to solve which problem. The map changes every week. The risk is disorientation, because people need belonging and continuity. The counterweight: flows stabilize around client relationships, which by nature require time and constancy. The relationship becomes the stable axis.
Take a law firm. Today it's organized by practice: corporate, litigation, tax, IP. AI absorbs legal research, contract drafting, due diligence, compliance analysis. What retains value is the ability to understand what the client wants, the judgment on which strategy to adopt, the negotiation with the counterparty. These skills don't belong to a practice area: a lawyer who can negotiate and read context can work an M&A deal in the morning and an IP dispute in the afternoon. The firm reorganizes around client relationships: stable teams that follow a client across all their problems, with AI providing productive specialization and people providing experience and continuity.
The organization as nervous system. AI is the nervous system. Its sensory organs are people in context: they talk to clients, frequent a market, read signals. Their work is to perceive, to bring into the organization information that AI cannot collect because it requires presence, trust, cultural sensitivity. AI processes: it takes that information and turns it into analysis, models, prototypes at a speed the human mind cannot match. The response goes back to people, who translate the output into action in the social world, presenting strategy, negotiating, guiding change. Management's role becomes curating connections: ensuring sensory information flows toward processing and processing feeds response. The organization's value is measured by its perceptive quality. How well does it sense the market? How quickly does it understand that something is changing? Production is abundant. Perception and judgment are scarce.
The organization as protocol. The organization stops being an entity with defined boundaries and becomes a collaboration protocol. Like an internet protocol defines rules for data exchange without owning the network's nodes, an organizational protocol defines rules for collaboration between professionals without owning the professionals. It runs on reputation, trust, and shared rules that allow professionals with different value skills to coordinate without hierarchy. AI drives coordination costs low enough to make the protocol sufficient. Size is not the number of employees but the number of nodes in the trust network. Reputation becomes the central organizing mechanism: you're inside the protocol if the network trusts your judgment, outside if it doesn't.
The three forms don't exclude each other. They're three scales of the same transformation: the flow is how work functions inside the organization, the nervous system is how the organization perceives and responds to its market, the protocol is how ecosystems of organizations and professionals function together.
Two layers
Organizations that integrate AI converge toward two coexisting layers. An execution layer, where AI operates within operational flows with growing autonomy. And a value layer, where people work with AI collaboratively, contributing judgment, relationships, and context that AI cannot source on its own.
Putting agents in org charts, giving them manager titles, assigning them to departments: these moves make agents legible inside structures designed to coordinate people with different productive skills. If production is automated, the organizing principle disappears, and adding agents as if they were colleagues moves the problem without solving it.
The shift is from roles to flows, from approval hierarchies to coordination protocols, from productive function as the organizing unit to valuable skill as the organizing unit. Organizations that see it can start building around it now.