Suppose you could turn an experienced employee’s output into a capability that anyone on the team could replicate; a process that once required years of experience suddenly simple to execute. No single point of failure, more customers benefitting from what that one person knows, and the organization gains capacity seemingly overnight. Now consider a question that lives alongside that opportunity: Who will develop the expertise needed once that method is no longer sufficient?
Making knowledge reusable is valuable, no question; but building an organization that can continuously improve what it knows is an even greater ambition.
Three recent developments bring that distinction into focus: executable research, the economics of expertise, and the design of work itself. Together, they offer a useful starting point for a leadership argument: as AI extends the reach of established expertise, companies should deliberately strengthen the human and organizational capabilities that renew it. That is a practical expression of the thesis: “Human Value Moves Upstream™.”
A New Way to Put Knowledge to Work
On September 16, Nature published research on Paper2Agent, a framework that makes research papers and associated resources accessible through AI agents, including executable tools where suitable code permits. In a set of 100 computational-biology papers, 74 were successfully converted and 99% of the proposed tools passed automated validation. (Nature: Paper2Agent).
That result has a specific scope; but it doesn’t demonstrate that a company can capture all the tacit knowledge of its experts. The researchers also distinguish faithful execution from the validity of open-ended analysis and retain human responsibility for evaluating scientific evidence. The possibility I take from it is an enterprise inference: some established methods could become easier for others to apply with less dependence on the original expert’s availability. That could expand access to useful capabilities. It also creates a design choice: Will the company use that access mainly to repeat what it already knows how to do, or to help people investigate problems it previously lacked the capacity to address?
Technology doesn’t decide which ambition deserves the investment.
Reusing Expertise Changes Its Economics
A September 15, 2026 Brookings analysis draws on the economic literature to explain why counting tasks exposed to AI is insufficient. Automation can change the expertise required in the remaining work, with different implications for pay, entry into occupations, and employment; but the analysis doesn’t support a universal promise that expertise becomes more valuable as AI advances (Brookings: Workforce policy for the age of AI). For leaders, that’s a reason to inspect the actual work rather than reassure people with an abstract promise of advancement.
If AI makes a previously scarce method widely accessible, the enterprise may benefit while the person associated with that method loses some of their economic advantage. The benefit could also reach people who previously lacked access to specialist support. Neither possibility tells us whether employees are developing stronger judgment or gaining meaningful opportunities. These are separate questions that deserve separate answers.
Protecting a specialist’s bottleneck is not the objective, expanding capability is. But sharing the specialist’s methods and developing their next contribution should be considered together. For the individual, that contribution might involve diagnosing unfamiliar problems, testing the method’s limits, integrating knowledge from another field, or developing a better approach. For the organization, it means deciding which of those contributions matters and giving someone the resources to pursue it. “Higher-value work” needs an identifiable purpose, a user, and a result. Otherwise, it’s a comforting phrase attached to a reduction in tasks.
Where Human Value Should Move
At Whole Enterprise™, our thesis that Human Value Moves Upstream™ describes the movement toward shaping purpose, asking consequential questions, designing work, and exercising judgment about what intelligence makes possible. In this context, it means helping people influence how the enterprise’s expertise develops and where it’s applied. That can happen at many levels.
An employee may notice that a standard recommendation repeatedly fails a particular group of customers.
A developing analyst may identify a question the existing model never considered.
An experienced team member may recognize that a once-useful assumption no longer fits the market.
For those observations to create value, the company needs a way to investigate them and change its methods. It also needs people who can distinguish a meaningful exception from noise. This is why I would be cautious about defining the future human role simply as supervising AI. Review without knowledge becomes a formality. Knowledge without the authority to challenge or change the process becomes frustration. The contribution has to be designed into the work. Nor does this argument require a claim that AI cannot help improve methods. It can participate in generating alternatives, testing assumptions, and analyzing results. The human responsibility, then, is to establish what improvement means, judge the evidence, and remain accountable for the consequences.
Human value moving upstream should mean more agency in that process, supported by greater capability.
Design an Organization That Learns from Using Its Knowledge
McKinsey’s September 14, 2026 article on scaling agentic AI recommends redesigning workflows and specifying changes to roles, accountability, and decision rights. Its case rests on practitioner experience rather than a controlled causal study. (McKinsey: Stacking the odds). My extension is to make the renewal of expertise an explicit part of that redesign.
Consider a hypothetical manufacturer preparing customer proposals. Suppose an AI system can apply a tested estimating method using approved cost and capacity data. (This is an illustration, not a reported implementation.) The immediate opportunity is to produce more consistent estimates with less preparation time. The larger opportunity might be to explore alternative configurations, identify recurring sources of uncertainty, and improve the company’s understanding of which commitments it can fulfill profitably. An experienced estimator could spend more time investigating unusual requirements and improving the method while a newer colleague could learn by explaining where an estimate is most vulnerable before comparing their reasoning with the system’s output. Operations could report where actual delivery differed from the assumptions used in the proposal. None of that follows automatically from faster estimating; it requires time to investigate discrepancies, access to outcome data, and an accountable person who can approve changes to the method. It requires incentives to surface weaknesses rather than quietly work around them.
If employees are measured only on proposal volume, those activities may become the work everyone agrees matters and nobody has time to do. The design task is to connect execution with learning: use the method, observe the result, examine the difference, and decide what should change. That is how reusable expertise could become a foundation for compounding capability. The compounding remains a hypothesis to test through outcomes, not a benefit to declare at deployment.
Don’t Automate Away the Route to Judgment
The same example exposes a second issue. Preparing estimates may be part of how new employees learn which assumptions matter. If the preparation disappears, leaders should examine what learning disappears with it. That doesn’t mean preserving repetitive work for its own sake; it means identifying its developmental function and designing a better way to serve it. People could reason through selected cases before seeing an AI recommendation. They could investigate disagreements, review unusual outcomes with an experienced colleague, and gradually take responsibility for more complex decisions. The test is whether they’re developing usable understanding. Finishing more work with assistance and understanding the work more deeply are different outcomes; and both can be worth measuring.
Experienced employees need a credible path. Asking them to share knowledge while offering no clarity about their future role creates an obvious tension. Leaders should explain what contribution they will need next, how people can prepare for it, and what opportunities actually exist. True, some roles may shrink or disappear, and a commitment to human agency doesn't justify pretending otherwise. It does; however, require candor about the transition and serious attention to the choices available to affected people. A person’s dignity does not diminish because one of their capabilities becomes less scarce.
Start with One Expertise Map
Choose one workflow that matters to customers or business performance. Bring together an experienced practitioner, someone learning the work, and the leader accountable for its result. Map its tasks and judgments into three categories:
AI commoditizes: work becoming easier to obtain at acceptable cost and quality.
AI amplifies: contributions that could become more capable, more useful, or possible for the first time through partnership.
AI cannot yet reliably replace: contributions still requiring human involvement under the conditions in which you operate.
Treat the categories as judgments to validate. Test against representative work, including difficult cases. Record uncertainty. Separately specify where accountable human authority should remain even as technical performance improves. For each important method you intend to automate or make reusable, answer four questions.
What new contribution will people make? Identify a specific unmet need. In the manufacturer example, it might be investigating uncertain commitments or evaluating alternatives that previously went unexplored.
How will they develop the required expertise? Specify practice, feedback, protected time, and access to experienced colleagues. Make development part of the redesigned workflow.
Who can improve the method? Assign responsibility for investigating failures, approving revisions, and retiring assumptions. Give employees a usable route to challenge the output.
What evidence would show that capability is growing? Measure more than production speed. Examine quality, rework, total cost, the usefulness of newly possible work, and people’s ability to recognize when the method should not be trusted.
Use a bounded pilot to test these choices before making broad promises. If the system creates more correction work than it saves, or the proposed new contribution has little value, revise the design. The purpose is to establish a repeatable way of learning what better partnership requires.
The Leadership Opportunity
These sources support the questions, not a guarantee of the answers. They do not establish that every organization can turn its knowledge into agents or that every worker will benefit from doing so. My position is that leaders should pursue a larger outcome than extracting more output from accumulated expertise. They should build organizations in which using that expertise helps people and systems become more capable over time, and that requires human development, work design, leadership, and technology to advance together. It connects the promise of reusable knowledge to the responsibility of keeping the organization able to adapt.
Before scaling the next AI-enabled method, ask: Are we building the capability to renew what we know—or only a faster way to repeat it?
