
When work changes before the catalog does
Imagine an architecture whose titles remain stable even as exceptions, decisions, and tasks have shifted: the work changed before its catalog did.
Imagine a team adopting a tool that can prepare an analysis that previously took several hours.
For the first few weeks, the role seems unchanged. People still have the same job titles, and the catalog retains the same skills.
Yet the work has already shifted. Checking assumptions becomes more important, exceptions arrive sooner, and conversations with other departments change. Producing output accounts for less of the role; framing questions and taking responsibility for decisions account for more.
Change emerges at the edges
Formal documents describe the stable core of a role. Transformations begin at its interfaces: a new approval, data arriving from a different source, an exception requiring judgment, or coordination that did not exist before.
Ask what has started taking more time, what now takes less, and where corrections accumulate. Also observe which decisions have changed hands. A task can remain on the list while losing its professional substance because the system does almost everything and the person intervenes only in difficult cases.
Autor, Levy, and Murnane showed that technology can replace routine tasks and complement nonroutine cognitive activities. Their analysis examines the US labor market in earlier decades; it does not predict the fate of a particular occupation or a current tool. It offers a useful unit of observation. Change happens in the mix of tasks, not necessarily in the immediate disappearance of the job. Autor, Levy, and Murnane, 2003 (opens in a new tab).
A provisional signal is not yet a skill
The team might respond by creating a new entity called “AI oversight.” The name records what is new but does not yet explain which distinct capability it represents. The work may require evidence validation, control design, domain knowledge, or decision-making under uncertainty: existing capabilities in a different combination, with different weights.
Before expanding the catalog, keep the signal with its context. Describe the situation, the decision affected, the frequency, the risk, and who encounters it. Compare cases over a period long enough to establish whether this is a local exception, a process change, or a reusable capability.
The architecture needs a provisional space for observation without turning every signal into a permanent category. Otherwise, every technology, tool, or practice becomes a skill whose name ages faster than the work.
Relationships confirm the pattern
A new capability deserves its own entity when it retains a distinct outcome and appears in more than one context. It must also be possible to develop or observe it, and it must change relevant decisions. It also needs a clear relationship with what already exists.
Sometimes no new skill emerges. What emerges is a different relationship among several skills. A previously secondary capability becomes an entry requirement; another shifts from execution to oversight. A third stops differentiating people because the system now incorporates it.

We can now revisit the team in the opening example. If different departments begin reviewing automated outputs, they may share criteria for evidence and escalation. If each needs entirely different knowledge and only the tool is shared, creating a common category could conceal more than it connects.
Change the architecture in layers
First, update examples and evidence so that the existing definition represents current work. Then review relationships, levels, or profiles if relative importance has changed. Create, split, or retire entities when those actions are necessary for a decision the model must support.
The NICE Framework change process shows how a public architecture can receive requests with context and review them before incorporation. Its domain is cybersecurity, but maintaining a traceable pathway prevents the catalog from depending solely on a major update. NIST NICE (opens in a new tab).
Every change needs a date and a defined effect. If a skill changes scope, do not reinterpret historical results as though they had measured the new standard. Preserve the version used and determine which evidence remains current.
Watch what disappears, too
Teams readily discuss emerging capabilities and say less about capabilities whose demand has shrunk. Keeping them as core requirements can direct development toward work that no longer creates value, exclude suitable candidates, and obscure the role’s actual composition.
Do not retire a skill simply because a tool performs part of the task. It may still be needed to verify, recover, or decide. Review frequency, the consequences of an error, and the depth of understanding required for oversight. The relationship between person and capability may change before the capability stops mattering.
By the end of the example, the catalog does not need a section for every new development. It needs to reflect that the role now frames better questions, checks evidence, and takes responsibility for exceptions. That update is less eye-catching than adding a fashionable label and much more useful for assessment, development, and work allocation.
The architecture will never catch up with the work through editorial speed alone. It can do so through attentiveness: detecting early signals, retaining them without overreacting, and deciding when a pattern has emerged that changes the contribution.
Before turning a signal into a catalog change, also reconstruct where the decision, information, or coordination breaks down. The work’s path shows whether the problem calls for a new skill or a different interface.





