Arthur Matuszewski’s “pyramid of value” (administration, judgment, orchestration, white space) and where AI can bring the most value.

Most AI-in-HR plans are built from a list of tasks, activities, procedures. Someone writes down what the function does, ranks the tasks by hours spent, and points AI at the top of the list. That is how you end up with talent-acquisition bots screening candidates who used AI to write their applications. Slop on both ends, sadly, all positioned as efficiency and productivity to the purse holders.

There is a better starting question, and we picked it up from Arthur Matuszewski on the latest Workestration episode. Arthur runs Carrara, an operating firm that embeds inside companies to do the work across talent, finance, go-to-market, and ops. When Donna asked where AI was doing the most work for his own team, he answered with a map.

The map

Arthur describes a pyramid of value with four layers:

• Administration at the base: the coordination, plumbing, and process that keeps work moving.
• Specialized judgment in the middle: the analysis and craft a particular expert brings.
• Resourcing and orchestration above that: who does what, and how those decisions get made better.
• White space at the top: where you know there is opportunity but not yet what people can make of it.

His point is not that AI lives at one layer. It is showing up at all four. The work is figuring out what it does at each, because the answer, and the cost of getting it wrong, is different every time.

Layer by layer

Administration: leverage you can see

At the base, Arthur’s coordinators and operations people wire APIs and tools together into one shared way of operating. The obvious win is leverage: more output per person. The less obvious win is observability: a shared system lets you catch the errors in a workflow, fix them faster, and earn the right to serve clients at higher throughput or higher quality problem. This is the layer where the case is easy and clean. Nobody was doing this coordination especially well anyway, regardless of the tools one invested in.

Specialized judgment: compounding the good stuff

In the middle, Arthur says the core thing AI does is “compound the ability to share the goods.” His example: a good spreadsheet usually lives on one person’s drive, where a few colleagues can edit it. Drop the same logic into a shared skill the whole team can run, and people start standing on each other’s shoulders, refining the analysis vs rebuilding it. Judgment is no longer isolated to one expert's head but rather accrued to the overall group.

Resourcing and orchestration: knowing your own supply and demand

One layer up, the question is who should be on what. Arthur describes Carrara as a micro-economy: this person is great at one thing, that person at another, so how do you pull someone in for a quick consult instead of committing them to a whole project? AI here is less about doing the task and more about matching capability to need, the resourcing decision that most orgs make on gut feel, watercolor encounter or a hallway "run-into."

White space: pointing capability at the unknown

At the top, Arthur’s team builds its own harness so it can widen its aperture across clients, carrying what it learns in one domain into healthcare, freight, or manufacturing. The old version of this was a room of McKinsey partners holding context in their heads. The new version pairs a subject-matter expert with capable AI and points them at a problem nobody has scoped yet, establishing foundations that then compound. It is the highest-leverage layer and the hardest to delegate to a tool alone.

Why the layer is THE point

The reason most AI deployments stall is not the model but rather that teams point automation at work they don’t actually understand. As Arthur puts it, “it’s harder to automate the things that you don’t understand.” The first-principles move is to understand what should be done, automate that, and only then expect to zoom out.

That reframes the roadmapping exercise. Instead of asking what AI can do to a task, ask which layer the task lives on. Administration rewards automation directly. Specialized judgment rewards sharing and compounding (vs replacement.) Orchestration rewards better matching. White space rewards a human and a tool exploring / collaborating together. The wrong intervention at the wrong layer (automate judgment you haven’t articulated, or “orchestrate” work you can’t yet see) leads to motion without value.

What the map does to the org chart

Follow the logic and the shape of the company changes. When administration is leveraged, judgment is shared, and orchestration is sharper, you need fewer people at the base holding up the few at the top. Arthur’s framing: the pyramid inverts into a diamond. Fewer roles devoted to mass production, more points of concentrated leverage, and a rise in what he calls the exponential individual, the person whose judgment and tools let them do the work of many.

The risk in that shape is the one Arthur names as a “cascade of absence”: each layer hires the one below and steps back, until the actual work falls to whoever is left holding it. The map is the antidote. If you know which layer value is created on, you know where leaders still have to show up and do the work rather than delegate it away. Or, in his blunter version, you have to keep holding the spear whatever your title is.

What to act on

Three moves, all of which run through the map.

1. Sort your AI use cases by layer, not by hours saved. Take your top ten and label each one: administration, specialized judgment, orchestration, or white space. The label tells you what intervention fits best (automate, share, match, or explore) and flags the ones you’ve mislabeled.
2. Don’t automate what you can’t yet describe. For anything on the judgment layer, write down how the work is actually done before you point a tool at it. If you can’t, that is the project, not the automation.
3. Find where your leaders have gone absent. Trace one important workflow down the layers and see where the work has quietly cascaded to the lowest rung. That is where someone senior needs to pick the spear back up.

The flashy version of this conversation is which tasks AI will take. The useful version is Arthur’s: name the layer first, and the question of what to do about it mostly answers itself.

Listen to the full conversation at Workestration.ai or on Apple Podcasts, Spotify, YouTube.

References and further reading:

• Focus on the Work, Arthur Matuszewski, Carrara
• Carrara, Arthur’s operating firm
• The Mythical Man-Month, by Fred Brooks, on why adding people to late projects backfires, revisited in an age of higher output at lower headcount