The CEO’s AI Dashboard, Part 7: Talent

Welcome to a new edition of The Board: Distillation Aftershots (*)

This is an online copy of the weekly newsletter, written by Constellation's Chief Distiller and Board Advisor Esteban Kolsky, sharing curious and interesting insights and data points distilled from enterprise technology to identify what’s notable. To subscribe, click here.

In this final installment of the CEO’s AI Dashboard series, I want to focus on the constraint that will determine how far AI can scale after the pilots, infrastructure, data, economics, and governance are in place: talent. The enterprise needs technical skills, but the harder shortage is experienced operators who understand how work behaves after automation reaches its limits.

First, my take.

The talent discussion still spends too much time on prompting, coding, and familiarity with AI tools. Those skills matter because they support early deployments, model development, and technical operations. Enterprises have recruited and outsourced technical expertise for decades, and they should continue doing so. Scaling AI introduces a different need.

Automation can absorb simple tasks and solve many complicated problems when the objective, inputs, and rules are sufficiently clear. Complex, wicked, and intractable problems contain unknown variables, competing consequences, incomplete information, and conditions that change during resolution. They require intuition, innovation, reasoning, understanding, and judgment beyond what can be encoded in a model. That is where experienced operators matter.

PwC’s 2026 AI Jobs Barometer, based on more than one billion job advertisements, found that AI is increasing the value placed on judgment, creativity, leadership, and other human capabilities. It also found that roles in which AI removes routine work and raises the premium on human judgment are growing faster than roles in which AI makes specialist work easier for nonexperts. As expertise becomes commoditized, experience becomes more valuable.

Experience is accumulated through exceptions, failures, informal workarounds, customer reactions, and decisions made without complete information. It is difficult to document because people often do not know which parts of their judgment matter until an unusual situation forces them to use it. AI can reproduce recorded rules and recognize patterns. It cannot be taught intuition or innovation, and its apparent reasoning breaks down when tested against unfamiliar, ambiguous, and consequential situations.

That loss of experience is becoming material. Deloitte estimates that more than 30 million Americans will turn 65 over four years, placing trillions of dollars of output at risk as institutional knowledge leaves the workforce. APQC’s 2025 research found that many organizations recognize the threat but still lack consistent methods for capturing and transferring expertise. The enterprise is allowing its hardest-to-replace operating asset to leave while recruiting for skills that are becoming easier to buy.

The human linchpin sits at the center of this issue. These are the people who connect fragmented applications, spreadsheets, approvals, and process exceptions. Some of their manual work will be absorbed by agents. Their knowledge of why the process behaves as it does remains valuable. They know where the data is unreliable, which exception indicates a larger failure, which customer will react badly, and when the formal process should be challenged.

The objective should be to move selected linchpins from manually bridging systems into designing, supervising, and correcting AI-mediated workflows. Their role begins where automation stops: ambiguous decisions, unusual conditions, escalation, process redesign, and monitoring consequences over time. These are decisions agents cannot make reliably because the required variables are unknown, contextual, and often visible only through experience.

There is an important distinction between preserving knowledge and retaining judgment. Documentation, recordings, decision histories, and AI models can preserve known information. They can explain what happened previously and make established practices easier to retrieve; AI vendors now position context graphs as a substitute for this experience but organizing recorded knowledge does not recreate judgment. Judgment develops through practice, responsibility, consequences, challenge, and repeated exposure to situations that do not conform to the documented process. Enterprises need both, but capturing knowledge and results alone will not replace experienced operators.

Removing routine work also removes part of the experience-development pipeline. Employees traditionally acquire judgment by completing repetitive work, observing variations, making controlled decisions, and gradually handling more difficult exceptions. When AI absorbs those early tasks, the enterprise must deliberately create other ways for employees to encounter ambiguity, see consequences, and develop decision authority.

BCG warns that widespread AI use can weaken critical skills when workers defer to automated judgment too early. Experienced professionals face the same risk through an “autopilot trap,” where convenience gradually replaces deliberation. An enterprise that automates entry-level work without redesigning how judgment develops may solve today’s labor problem and create tomorrow’s experience shortage.

The evidence already shows how little progress organizations have made. Deloitte’s 2026 Human Capital Trends research found that 85% of leaders consider adaptability critical, while only 7% believe their organizations lead in helping workers grow and adapt continuously. Only 6% report meaningful progress in designing effective human-AI interactions. IBM found that 70% of executives have experienced AI-project problems caused by unclear authority, while 34% of organizations lack a repeatable process for resolving disagreement between human and AI judgment.

Talent readiness has two parts. The enterprise needs technical builders who can develop and operate AI systems. It also needs experienced operators who can determine when those systems are wrong, incomplete, or working against the larger objective. These groups should work together, with authority and responsibilities defined around the problem rather than around the technology.

The CEO question is whether the organization is preserving and developing the human judgment required after automation handles the obvious work. If that judgment is leaving faster than it is being captured, or if employees no longer have a path to develop it, the enterprise will scale automation while weakening its ability to manage the consequences.

Recommended CEO actions

1. Create a structured program to retain departing experience through contracted operator roles, decision reviews, scenario exercises, mentoring, and recorded exception histories. Retirement or departure should not automatically sever access to institutional experience.
2. Require the CHRO and COO to redesign the experience-development pipeline after routine work is automated. Include rotations, simulations, supervised exceptions, mentoring, and deliberate exposure to increasingly complex decisions.
3. Separate knowledge-preservation programs from judgment-development programs. Documentation and AI can retain known information; judgment requires practice, responsibility, and exposure to consequences.
4. Monitor talent readiness through board-level indicators: loss of critical institutional knowledge, coverage of high-risk workflows by experienced operators, strength of the experience-development pipeline, and the percentage of AI workflows with named human escalation owners.
5. Treat technical expertise and operating experience as complementary resources. Technical skills can often be recruited, contracted, or shared across initiatives. Enterprise-specific judgment has to be retained, developed, and connected to the operating model.
6. Identify the human linchpins and experienced operators inside workflows targeted for automation before redesign begins. Move those with the strongest judgment into roles that shape, supervise, test, and improve AI-mediated work.

Here are some reading resources

1. PwC’s 2026 Global AI Jobs Barometer is useful because its analysis of more than one billion job advertisements shows that AI is increasing demand for judgment, creativity, leadership, and other human capabilities.
2. IBM’s 2026 AI-human operating-model research is useful because it connects revenue growth and operating margin to clear authority, human judgment, and disciplined execution around AI.
3. Deloitte’s 2026 Human Capital Trends research is useful because it shows the gap between the need for workforce adaptability and enterprises’ limited progress in continuous learning and human-AI work design.
4. Deloitte’s research on capturing institutional knowledge is useful because it documents the scale of expected retirements and the economic risk created when tacit knowledge leaves the enterprise.
5. APQC’s research on retirement, AI, and knowledge loss is useful because it shows that many organizations recognize the institutional-knowledge risk but lack consistent approaches to capture and transfer experience.
6. BCG’s research on critical-skill erosion is useful because it warns that extensive AI use can weaken judgment, learning, and expertise when organizations allow automated answers to replace deliberation.
7. BCG’s work on converting AI skills into business performance is useful because it emphasizes problem solving, judgment, collaboration, and relationship building and shows that leading enterprises invest substantially more in capability development.
8. The Urban Institute’s report on AI and older workers is useful because it explains how AI can complement experienced workers while identifying the training and organizational barriers that may prevent enterprises from retaining their contribution.
9. ManpowerGroup’s 2026 Talent Shortage Survey is useful because 72% of employers across 41 countries report difficulty filling roles, while AI capabilities have become the most sought-after skill category.
10. IBM’s research on agentic enterprise operations is useful because it describes workflows as a deliberate combination of human judgment, real-time orchestration, and AI autonomy.
11. Deloitte’s workforce-evolution research is useful because it describes how experienced employees can transfer tacit knowledge through real work and reports that 60% of respondents believe AI can help experienced workers share knowledge across the organization.
12. BCG’s research on insurance talent is useful beyond that industry because it examines how organizations can develop judgment when AI removes the repetitive work that traditionally gave employees experience.

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’s offering further, or have any additional questions, please contact me at [email protected], and I will be happy to connect with you.

(*) A normal distillation process has byproducts: primary, simple ones called foreshots and secondary, more complex and nuanced called aftershots. This newsletter highlights remnants from the distillation process and shares insights to complement the monthly report.