Workforce and Organization: Experienced Judgment Is the Constraint

If you want to receive this in your inbox, subscribe here.This is the third post in our series that creates a new structure for enterprise technology going forward (where AI is a player, not the basis).

First, introducing the taxonomy.

Second, discussing what is business value beyond AI.

In 2017, Attention Is All You Need introduced the Transformer architecture that became foundational to today's generative AI. One of its breakthroughs allowed AI to pay attention to many parts of the information at once rather than processing them one after another. Nine years later, machine attention can operate at a scale human attention never could. As AI moves from generating answers to executing work, the enterprise constraint moves with it. Experienced human judgment becomes harder to scale than machine execution.

Generative AI dramatically expanded the amount of cognitive work machines can produce. Agents are now extending that capability from generating outputs to executing work. Deloitte found that 61% of surveyed leaders expect most agents used by their organizations to become generally autonomous within four years, while 74% expect nearly half of their business processes to be rebuilt around them. Enterprises can potentially create machine execution faster than they have ever been able to add human labor.

Experienced judgment becomes the scarce resource in that environment.

For decades, enterprises depended on human linchpins who connected fragmented applications, reconciled inconsistent data, resolved exceptions, and understood how the business actually worked when documented processes were insufficient. Agents can increasingly perform much of that execution. They cannot automatically replace everything embedded in the person performing it.

Process knowledge, business rules, histories, and known exceptions can increasingly be captured as enterprise context and made available to machines. Experiential judgment develops differently. Routine work surfaces exceptions. Handling exceptions creates knowledge about how customers, systems, processes, and the enterprise behave when reality diverges from expectations. Accumulated knowledge becomes experience, and experience allows an operator to exercise judgment when the next situation does not fit the rules.

Automation can interrupt that progression. Agents performing routine work also encounter the exceptions through which people historically learned. As agents improve, they will resolve more of those exceptions themselves. Humans increasingly see only what machines recognize and choose to escalate, potentially losing exposure to anomalies the machines fail to recognize at all.

As agents take over more work, they can reduce the opportunities people have to develop the experience required to supervise them.

The evidence is beginning to expose the problem. PwC's analysis of more than one billion job advertisements found that AI-exposed entry-level roles are seven times more likely to require skills traditionally associated with senior employees, including judgment and leadership. BCG found that half of surveyed senior executives are already observing de-skilling and more than 60% believe it will become a material threat within three to five years.

Two forces could consequently move in opposite directions. Autonomous execution expands while the mechanism that historically produced experienced operators contracts. Enterprises could improve productivity today while reducing their future supply of people capable of recognizing when automated execution is wrong.

That judgment will be required throughout the organization. The accounts-payable specialist who understands why one supplier behaves differently, the customer-service operator who recognizes that an unusual request signals something larger, and the security analyst who sees a pattern across individually legitimate events possess forms of experiential judgment. Calling their future role human oversight understates the requirement. Presence does not create judgment. The enterprise needs people who understand the work well enough to recognize what the machine does not.

Thousands of agents can operate simultaneously, continuously, and at machine speed while experienced human judgment remains finite. Compute can be added, models replicated, and agents instantiated rapidly. Experienced operators cannot. The relevant workforce question increasingly becomes how much autonomous execution the enterprise can responsibly support with the judgment capacity available to supervise, challenge, audit, and intervene.

Machines will inevitably perform some of that supervision. Agents can monitor agents, evaluate outputs, enforce policies, detect anomalies, and escalate exceptions. The volume of machine activity will make this necessary. It solves part of the monitoring problem without resolving the judgment problem.

Treating guardian agents as the answer risks putting the fox in charge of tending the henhouse. Executing and supervising agents can share models, data, context, assumptions, or reasoning weaknesses. Machine supervision can reduce what requires human attention while still failing to recognize that the automated system itself is wrong. Independent experienced judgment remains necessary precisely because machines increasingly participate in both execution and supervision.

Enterprises have rarely treated that judgment as capacity because they historically acquired it indirectly. People worked, encountered exceptions, solved them, accumulated knowledge, and became experienced operators. Agentic AI can break that mechanism while simultaneously increasing demand for its output.

Judgment capacity therefore needs to become part of workforce planning: where experienced judgment resides, how it is produced, where autonomous execution requires it, and how much is available relative to machine execution. This does not require another KPI. It requires understanding judgment as a finite enterprise capability that must be developed, preserved, and allocated deliberately.

Security and data architecture influence that capacity as well. Agents need access to systems, information, and authority to act; experienced operators need visibility into what agents did and the context behind consequential decisions. Tight controls can prevent both machines and humans from obtaining enough context to work effectively, while excessive access increases security, privacy, and compliance exposure. Identity, permissions, and data flows increasingly determine productive capacity rather than merely protecting it.

The workforce itself may become more fluid around these constraints. Permanent employees can retain institutional context, accountability, and strategically important judgment while agents provide scalable execution. Experienced specialists may increasingly supply judgment for projects, transformations, audits, and unusual situations where permanent capacity is unnecessary. Agentic execution changes the economics of expertise enough to make it worth planning for.

Workforce and Organization occupies this role in the 2027 Enterprise Technology Taxonomy because Business Value establishes the outcomes the enterprise needs while the workforce determines whether it has the productive capacity to create them. Enterprise Intelligence and Context provides the knowledge humans and machines require; Architecture, Infrastructure and Modernization provides technical capacity; Trust, Identity and Control establishes authority and acceptable boundaries. Judgment increasingly connects all of them.

Over the next 24–36 months, boards and executive teams need to understand where experienced judgment resides, whether automation is disrupting its development, which judgment must remain inside the enterprise, and how much autonomous execution existing judgment capacity can responsibly support.

Attention Is All You Need helped make machine attention scalable. Intelligence can now scale. Execution can scale. Machine supervision will increasingly scale. Experienced judgment does not scale the same way, and enterprises may simultaneously be automating the work through which people acquire it.

The agentic enterprise may ultimately be constrained less by how many agents it can deploy than by how much autonomous execution its experienced judgment can responsibly support.

The human-machine bridge behind that constraint deserves deeper examination. Later this year I will explore how enterprises develop judgment when machines increasingly perform the work that once created experience, how human and machine supervision should interact, and what a deliberate judgment architecture could look like.

Reading Resources

Vaswani et al., “Attention Is All You Need,” 2017.

Deloitte, “AI Agents Are Only the Beginning,” August 12, 2026.

McKinsey & Company, “The Agentic Organization: Contours of the Next Paradigm for the AI Era,” 2026.

McKinsey & Company, “Building Expertise in the Age of AI: Who Trains the Next Generation?”, July 2026.

PwC, “2026 Global AI Jobs Barometer,” June 15, 2026.

BCG, “When Everyone Uses AI, Companies Risk Losing Critical Skills,” June 17, 2026.

Deloitte, “Decision-Making in the Age of AI,” 2026 Global Human Capital Trends.

PwC, “AI Agents as Workforce Counterparts—What Governance Should Look Like,” July 17, 2026.

Esteban Kolsky, “Revaluing the Human Linchpin in Enterprise Operations,” April 2026.

Esteban Kolsky, “The CEO's AI Dashboard — Talent,” 2026.

What's your take? We are fostering a community of executives who want to discuss these issues in depth. This newsletter is but a part of it. We welcome your feedback and look forward to engaging in these conversations.

If you are interested in exploring the full report, discussing The Board offering further, or have any additional questions, please contact me at [email protected].