Roles for a Future Civilization

Every civilization keeps two lists. One is the list of roles it celebrates — the founders, the visionaries, the names on the building. The other is the list of roles it actually runs on — the people who keep the water clean, the institutions talking to each other, the rules enforced and the trust intact. A healthy civilization assumes these are the same list. They never have been, and the gap between them is the most under-read signal in strategy.

Strategic intelligence is, at bottom, the discipline of telling those two lists apart — reading what a system depends on rather than what it advertises. It is the difference between the marble façade and the load-bearing wall behind it. And by that measure the modern world is operating on a broken instrument: it reads its own civilizational health off a prestige gauge that is wired backwards.

Here is the claim in its strong form: the prestige gradient is inverted relative to the dependency gradient, and it is inverted by mechanism, not by accident. Prestige flows to the roles whose per-hour value is collapsing; dependency concentrates in the roles physics refuses to make cheaper. Two forces ride every productivity surge — Jevons, where we spend more on what gets cheaper, and Baumol, where we spend more on what stubbornly does not — and they guarantee the divergence. The roles a civilization can automate get cheap and celebrated; the roles it cannot become the relative luxury, and get filed under “unskilled” precisely because no machine can replace them.

The mechanism has a name worth keeping in front of you: the gauge reads velocity and calls it value. Falling cost looks like progress, and progress looks like merit, so esteem chases the roles whose cost is in freefall and mistakes cheap-and-abundant for scarce-and-valuable. This is why the occupational-prestige tables and the pay tables barely correlate. Prestige is a separate market, priced on narrative, and it is mispriced in a known direction.

Now add the accelerant. AI is collapsing the celebrated generation work to commodity, and when a machine does 99% of a task, the last one percent becomes the whole price — the human residual that verifies, integrates, and maintains bottlenecks everything and commands the wage. The binding constraint is migrating, in real time, off the roles we honor and onto the roles we ignore. AI does not threaten the overlooked roles. It manufactures them.

And the roles it is manufacturing demand into share one disqualifying property: their work cannot be attributed to them. The connector’s bridge, the enabler’s leverage, the maintainer’s prevented failure — all of it shows up as the absence of a problem somewhere else, invisible to the ledger that funds it. The same invisibility that makes these roles load-bearing is why the market underproduces them. The decomposers go missing first — the lowest-prestige guilds vanish before the collapse, and the collapse looks like abundance right up until it doesn’t.

This is not a morality essay. It is a strategic-intelligence failure with a body count. Rome’s aqueducts failed as maintenance, not engineering. Modern infrastructure scores its best-ever grade atop a multi-trillion-dollar shortfall. Millions of trades positions go unfilled while the workers who hold them age out. A civilization that cannot tell its prestige list from its dependency list is a civilization flying blind into its own maintenance crisis — and calling the descent a meritocracy.

What follows is the breakdown — twelve roles every society runs on, each dissected the same way across six aspects, ordered from the most over-honored to the most starved. Read it as an intelligence product: a map of which roles are celebrated, which are faked, and which are being starved at the exact moment they become load-bearing — and therefore where the next century is actually going to be built.

Summary

The twelve roles, ordered from the most over-honored to the most starved relative to how much a society actually depends on them:

• The Thinker — frames reality. Honored ≈ depended-on. Overproduced at the prestige end, undersupplied at the rigorous end; degrades into thought-leadership theater.

• The Generator / Doer — makes new things. Over-honored. The celebrated role; its output is exactly what AI is commoditizing fastest. Cheap and abundant, mistaken for scarce.

• The Manager — allocates and directs. Over-honored at the top, mis-practiced in the middle. Captures rents (CEO pay +1,094% since 1978) while real coordination work goes undone.

• The Allocator / Risk-Bearer — points capital at the future. Over-honored; uniquely able to capture rents. Builds when it bears real downside; extracts when it doesn’t.

• The Translator / Sensemaker — makes reality legible. Mixed and dangerous. Holds the dial on the prestige gauge itself; the loud are over-honored, the careful under-honored.

• The Connector — bridges disconnected groups. Under-honored. Highest network multiplier, lowest individual value-capture; non-attributable by construction, so chronically underproduced.

• The Enabler — multiplies everyone else. Invisible to the ledger. The single largest driver of org performance that only 5% of executives rank in their top three.

• The Critic / Verifier — catches the error, holds the standard. Under-honored, repricing hard. As generation goes free, verification becomes the whole price — the clearest AI-era growth role.

• The Integrator — makes systems cohere. The violent reprice. A $1.8T market the prestige gauge ignored; comp and demand now snapping upward 700%+ as AI exposes integration as the real constraint.

• The Cultivator — develops the humans the stack depends on. Esteemed, starved. Recognition granted, resources withheld; the productivity-resistant role we underpay while demand surges.

• The Maintainer — keeps what exists alive. Severely starved. Where a society defers it, the collapse is always a maintenance signature wearing the mask of an external shock.

• The Steward / Guardian — protects the long horizon and the commons. The most starved of all. Pure slow-feedback work, so the prestige gauge can’t see the slow death until it arrives.

The roles a flourishing society is actually bottlenecked on — Connector, Enabler, Critic, Integrator, Cultivator, Maintainer, Steward — share one disqualifying property: their work cannot be attributed to them. It shows up as the absence of a problem somewhere else, invisible to the ledger that funds it. The same invisibility that makes them load-bearing is why the market underproduces them. The fix is not information (you’ve just read the data; it won’t move a single career). The fix is title arbitrage — re-pointing status onto the load-bearing role before the prestige gauge catches up. Each role below is broken down across six aspects: what it actually does · how it works well · where it sits on the prestige-vs-dependency gradient · its failure modes · its supply and demand · how to do it or allocate into it.

The breakdown: seven roles a civilization runs on

Three forces recur across every role, so name them once. The gauge reads velocity and calls it value — esteem chases falling-cost roles and mistakes cheap-and-abundant for valuable. The last one percent becomes the whole price — when a machine does 99% of a task, the human residual that’s left bottlenecks everything and commands the wage. The decomposers go missing first — the lowest-prestige maintenance guilds vanish before the collapse, and the collapse looks like abundance right up until it doesn’t. The roles are ordered from most over-honored to most starved.

1. The Thinker — frames the reality everyone else operates inside

What it actually does. The Thinker produces the knowledge artifacts — theories, frameworks, models, standards — through which a civilization understands itself. This is the base layer of the stack: it lets intelligence compound across generations instead of resetting each time. The Thinker’s output is not “ideas” but the lens other roles can’t see without. Done at the highest level it is integration of knowledge across domains — the generalist who synthesizes where specialists can’t (Epstein, Range, 2019). Value increasingly migrates here as execution gets automated: information is only an input, and the judgment layer is where the worth concentrates (Autor, NBER w32140).

How it works well.

• Strips a problem to first principles, then rebuilds forward — physics, not analogy.

• Synthesizes across domains; the breakthrough usually arrives from the unrelated field.

• Produces reusable frames (a stack, a gradient, a constraint) others can operate, not one-off takes.

• Is willing to be wrong loudly and update — variance preserved, not optimized away.

Where it sits on the gradient. Roughly honored in proportion to dependency — the one role where the prestige gauge is mostly calibrated. Genuine thinkers are scarce and esteemed. The distortion is at the edges: the appearance of thinking is over-rewarded while the slow, unglamorous work of actually framing a problem is not.

Failure modes.

• Thought-leadership theater — producing the affect of insight (the talk, the thread) without the load-bearing frame underneath.

• Over-attribution — society credits the visible “visionary” and discounts the system that produced the idea, the “romance of leadership” at civilizational scale (Meindl et al.).

• Capture — the thinker who frames reality for whoever pays, laundering interest as analysis.

Supply & demand. Over-supplied at the prestige end, under-supplied at the rigorous end. Elite talent floods toward visible idea-work — ~50% of Harvard and Stanford graduates pour into finance, consulting, and tech (Binder et al.) — but the patient, low-status work of building durable frameworks is undersupplied. AI raises the premium on real thinking (taste, judgment, what’s worth making) while flooding the zone with synthetic plausibility.

How to do it / allocate into it. Earn the frame before you sell it; build instruments others can reuse; hold variance — a civilization needs its slow learners to keep options alive (March, 1991).

2. The Generator / Doer — makes the new thing that didn’t exist yesterday

What it actually does. The Generator builds — writes the code, founds the company, ships the artifact, makes the move. It defines the ceiling of what a system can become; a founder assembles a structure from nothing (founder vs operator). This is the most visible, most narratable role, and therefore the one prestige worships. It is also the role whose unit cost AI is collapsing fastest.

How it works well.

• Converts a frame into a working artifact under real constraints.

• Owns outcomes, not motion — ships the thing, then owns what it does in the world.

• Knows when to stop generating and hand off to the roles that integrate and maintain.

Where it sits on the gradient. The most over-honored role in the modern economy. By Jevons, generation is exactly what productivity makes cheap and abundant — and the gauge reads that abundance as merit. Falling cost looks like progress; progress looks like the genius who shipped. The prestige is real; the scarcity that would justify it is evaporating.

Failure modes.

• Founder worship — treating the generator as the whole system, when “a founder with a decent idea and a great operator beats a founder with a great idea and no operator almost every time” (founder vs operator).

• Generation without integration — shipping artifacts the organization can’t absorb; the all-doers monoculture.

• Liability of newness — novel structures fail at higher rates; generation is fragile until maintained (organizational ecology).

• Debt at scale — AI lets the doer ship more and more rot: >15% of AI-assisted commits introduce a defect, ~24% surviving uncleaned (arXiv, 2026); experienced developers were 19% slower with AI while believing they were 20% faster (METR, 2025).

Supply & demand. Structurally over-supplied and getting more so — generation is the layer being commoditized in real time (Sonar: 42% of committed code now AI-generated, heading for two-thirds by 2027). The marginal generator adds the least; the binding constraint has already moved downstream.

How to do it / allocate into it. Generate where the frame is genuinely new and the integration path exists — not where you’re adding the millionth cheap artifact. The durable value was never the generation; it’s owning the workflow the generation plugs into (a16z).

3. The Manager — allocates attention, people, and capital

What it actually does. The Manager directs scarce resources to where they produce the most — the coordination function that turns a pile of doers into a system. Done right it is pure leverage on the highest-cost asset (other people’s attention and effort). Done as titled, it is the role most prone to capturing reward without producing the coordination it’s paid for.

How it works well.

• Subordinates everything to the system’s actual constraint, not the loudest department (Theory of Constraints).

• Balances exploration and exploitation — refuses the local optimum that exploitation alone converges to (March, 1991).

• Makes others more effective and measures itself by their output, not its own visibility.

Where it sits on the gradient. Over-honored at the top, mis-practiced in the middle. Executive prestige and pay have decoupled from contribution: CEO pay rose 1,094% from 1978–2024 against 26% for the typical worker — read by the data as “rents,” not “a rising value of skills” (EPI). The coordination work that actually matters is mostly done lower down and credited higher up.

Failure modes.

• Rent capture — extracting reward via position rather than producing coordination; market power rising from 21% to 61% markups since 1980 is the macro signature (De Loecker et al.).

• Productivity theater — managing the appearance of work; 83% of employees admit to performative busywork, rationally, because visibility is rewarded (Visier).

• Over-attribution — claiming system outcomes as personal leadership (romance of leadership).

Supply & demand. Over-supplied at the extractive end, under-supplied at the genuinely-coordinative end. There is no shortage of people who hold the title; there is a chronic shortage of managers who actually subordinate themselves to the constraint.

How to do it / allocate into it. Manage the constraint, not the org chart. Treat your contribution as the freed capacity of everyone you coordinate. If you can’t point to whose throughput you raised, you’re capturing, not managing.

4. The Connector — bridges the groups that would otherwise never talk

What it actually does. The Connector spans the structural holes between disconnected clusters — carrying information, opportunity, and trust across boundaries that would otherwise stay sealed. This is the coordination-token layer of the stack: it’s how trust and information travel across contexts without renegotiation. The Connector’s ideas are judged more valuable precisely because they arbitrage what each isolated group can’t see (Burt, 2004).

How it works well.

• Sits deliberately at the edge of multiple worlds and translates between their vocabularies.

• Moves information and opportunity before anyone prices the bridge.

• Maintains ties actively — brokerage is a behavior, not just a position (brokerage review).

Where it sits on the gradient. Badly under-honored. This is the first of the three bottleneck roles, and it carries the defining curse: the highest network multiplier, the lowest individual value-capture. The value is real and ephemeral — nine of ten bridging ties vanish year to year, and the premium dissolves the instant anyone else bridges the gap (Burt, 2000). Non-attributable by construction.

Failure modes.

• Burnout — spanning holes imposes relentless translation load; brokering raises burnout and even abusive behavior (Org Science, 2023).

• Invisibility tax — the work evaporates before it can be measured, so it’s never credited and never funded.

• Gatekeeping — the connector who hoards the hole instead of bridging it, extracting rent from a gap they should close.

Supply & demand. Chronically under-produced, because the role is rationally avoided: high cost to the person, low capturable reward. Demand is rising as systems fragment and AI multiplies the number of things that must be reconciled.

How to do it / allocate into it. Make the bridge legible — log who you connected and what flowed, so the value survives long enough to be seen. Bridge holes; don’t camp on them.

5. The Enabler — multiplies everyone else’s output

What it actually does. The Enabler builds the tooling, platforms, and support systems that raise the productivity of every other role — the leverage layer. Its ROI lives entirely in other people’s freed capacity, which is exactly why the ledger can’t see it. Best-in-class internal tooling is the single largest driver of business performance, and only 5% of executives rank it among their top three enablers (McKinsey).

How it works well.

• Builds the platform once so a hundred people stop solving the same problem.

• Measures itself in others’ throughput, not its own activity — and makes that legible (revenue-per-employee is the cleanest proxy for whether the enabler layer works) (SaaS Capital).

• Treats productivity as multidimensional, resisting the single-metric trap that hides enabler value (SPACE framework).

Where it sits on the gradient. Invisible to the system that funds it. Enabler value shows up as the absence of friction elsewhere, so accounting files it as a cost center and cuts it first in a downturn (McKinsey G&A). Second of the three bottleneck roles.

Failure modes.

• Illegible leverage — up to 70% of internal “platform” teams fail to deliver measurable impact and nearly 30% never even define success (The New Stack). An enabler who can’t make the leverage visible genuinely deserves the axe — invisibility is a real tax here, not only an injustice.

• Build-for-its-own-sake — tooling nobody adopts; leverage that exists only on the slide.

• First against the wall — cut in the downturn precisely because the value was never made legible.

Supply & demand. Under-funded relative to impact, because the ledger structurally can’t price it. Demand rises as systems grow more complex and the cost of not having leverage compounds.

How to do it / allocate into it. Instrument your own impact — name the capacity you freed in someone else’s numbers. Leverage that isn’t measured will be cut; leverage that is measured gets funded.

6. The Integrator — makes the pieces actually work as a whole

What it actually does. The Integrator stitches separate systems, teams, and artifacts into something that functions end to end — the embedded engineer who makes the deployment work, the chief of staff who connects siloed work streams, the glue that turns components into a coherent stack. In machine-learning systems, at most ~5% of the code is the model; ≥95% is the glue that integrates it (Sculley et al., 2015). Failure lives at the seams, not in the components.

How it works well.

• Owns the end-to-end workflow, not a single excellent feature — that’s where durable value sits (a16z).

• Embeds inside the real system long enough to make it cohere, absorbing the messy last mile.

• Translates between the people who built the parts and the people who must use the whole.

Where it sits on the gradient. The role the market is repricing most violently right now — proof the prior cheapness was suppressed abundance. Integration is a $1.8 trillion market the prestige gauge ignored, “the largest market in enterprise software the AI shift hasn’t touched” (a16z). Postings for embedded integrators rose ~729% in a year; comp now reaches $1.2M at frontier labs (BigGo; Perspective AI). Third of the three bottleneck roles — and the one AI is manufacturing fastest, because the last 1% (verify, integrate) becomes the whole price.

Failure modes.

• The glue-work penalty — do the highest-impact integration work and get told “that wasn’t a technical contribution,” so the rational person stops (Being Glue); formalized as “non-promotable tasks,” valuable to the org and useless to the career (Babcock et al., 2017).

• The services trap — copying the integrator aesthetic without owning the platform turns you into expensive humans wearing software’s clothes, no compounding moat (the Palantirization risk).

• 95% of enterprise AI pilots fail — almost always at integration, not model quality (MIT/Fortune).

Supply & demand. Chronically lacking and now the most actively-hunted role in the economy — the clearest live example of the dependency gradient finally clearing its backlog.

How to do it / allocate into it. Own the workflow, not the component. Take the integration residual AI keeps manufacturing — it’s the most defensible, least-contested position available. This is where title arbitrage already worked once (Palantir’s rename of the low-status “integration engineer” built a hiring moat) and where it will work again.

7. The Maintainer — keeps alive what everyone else already built

What it actually does. The Maintainer repairs, upkeeps, cares for, and decomposes — keeping existing systems, bodies, and institutions functioning so the rest of the stack has something to stand on. In ecological terms it is the decomposer guild: invisible, near-zero-prestige, and non-substitutable — remove it and the system suffocates on its own undigested output (ecosystem collapse). In software, ~70% of investment goes to maintenance, not new creation (Russell & Vinsel).

How it works well.

• Treats upkeep as scheduled, professionalized work — Rome cleaned its aqueducts on a 1-to-5-year cycle with a standing 700-person corps (Frontinus).

• Inverts the prestige gradient deliberately where failure is fast: the lowest-ranking sailor can halt carrier flight ops and is commended for a correct stop (HRO research); “Forceful Backup” makes challenging the senior a duty (Rickover).

• Acts on the math: a dollar of preventive maintenance saves six to ten.

Where it sits on the gradient. The most starved relative to dependency — and starvation tracks one variable: how fast failure kills you. Where failure is instant and lethal, the maintainer is sovereign (SUBSAFE took submarine losses from one every three years to zero in 57 by making the welder and the checklist supreme — SUBSAFE; a surgical checklist cut deaths 1.5%→0.8% across eight countries — NEJM, 2009). Where failure is slow and diffuse, the maintainer is abandoned even when the arithmetic is overwhelming.

Failure modes.

• Deferred-maintenance death spiral — U.S. infrastructure scores its highest grade ever, a C, atop a $3.7T shortfall, with deficiencies costing each household $9/day against a $5.48 fix (ASCE, 2025); deferred road-and-bridge maintenance has run net-negative almost every year since 2004 (Pew, 2025).

• Normalization of deviance — even elite institutions learn to stop seeing the warning once feedback lengthens (Vaughan, Challenger).

• The maintenance-signature collapse — Rome’s aqueducts failed as upkeep, not engineering; Tainter’s law says collapse is what happens when the compounding maintenance bill quietly exceeds the will to pay it (Tainter).

Supply & demand. Catastrophically under-supplied and aging out: across the developed world, millions of skilled-trades positions go unfilled by 2030 — a multi-hundred-billion exposure, roughly four jobs posted per new entrant (JLL via Fortune). Care work is paid $13.51/hr against a $27.31 average and still can’t find takers (EPI); home-health aide is already the largest occupation and adds the most jobs of any role this decade (BLS). The decomposers are going missing first.

How to do it / allocate into it. Maintain where the failure-feedback is slow — that’s where the math most favors you and the competition least bothers. And shorten your own feedback loops: the only durable cure for a deferred-maintenance culture is to make the cost of neglect visible this quarter, not in thirty years.

8. The Allocator / Risk-Bearer — points capital and downside at the future

What it actually does. The Allocator directs capital and bears risk — deciding which parts of the system get fed and standing in front of the downside when they fail. This is the coordination-token layer made active: money and prices are how a civilization places bets across contexts without renegotiating each one. Done right it is the role that funds the fragile new thing before anyone can prove it works (liability of newness). Done wrong it is the purest engine of rent extraction a society has.

How it works well.

• Allocates toward the constraint, not the celebrated non-constraint — funds the bottleneck, not the bubble.

• Bears real downside; the risk is the job, not a fee skimmed off other people’s risk.

• Backs the load-bearing-but-illegible role before the prestige gauge reprices it — title arbitrage is an allocation strategy.

Where it sits on the gradient. Over-honored, and uniquely able to capture rents. Finance and executive allocation pay have decoupled from contribution — CEO pay +1,094% since 1978, read as “rents” not skill (EPI); markups rose from 21% to 61% above cost since 1980 (De Loecker et al.); ~50% of elite graduates flow straight into finance, consulting, and tech (Binder et al.).

Failure modes.

• Rent-seeking over building — the canonical finding: more rent-seekers slows national growth, more builders speeds it (Murphy, Shleifer & Vishny, 1991).

• Value destruction dressed as allocation — by social-return accounting, bankers can destroy ~£7 of value per £1 they collect and tax accountants ~£47, while childcare creates £7–£9.50 (NEF, A Bit Rich).

• Feeding the non-constraint — pouring capital into the celebrated generator layer that was never the bottleneck.

Supply & demand. Over-supplied at the extractive end, under-supplied at the patient, long-horizon, risk-bearing end — the unglamorous bets on maintenance, integration, and care that the prestige gauge can’t see.

How to do it / allocate into it. Bear real downside; allocate to the constraint; treat re-pointing capital at the starved load-bearing roles as the highest-return arbitrage available — because the market has mispriced them for you.

9. The Translator / Sensemaker — makes complex reality legible to everyone else

What it actually does. The Translator converts between domains and compresses complexity into something a non-expert can act on — the communicator, the explainer, the narrator who turns a frame into public comprehension. This is the legitimacy-and-culture layer: it decides what a society understands itself to be doing. Distinct from the Connector, who bridges networks; the Translator bridges meaning. It is also the role that holds the prestige gauge’s dial — because prestige moves on narrative, not productivity (occupational-prestige data: teacher esteem +25 points since the 1980s while pay lagged).

How it works well.

• Compresses without distorting — the espresso shot, not the watered-down mush.

• Synthesizes across domains; the clearest explanation usually comes from someone who can see two fields at once (Epstein, Range).

• Earns trust as the scarce asset — when generation is free, “confidence becomes the differentiator” (Sonar).

Where it sits on the gradient. Mixed and dangerous. The loud sensemaker is over-honored; the careful one is under-honored. And because the Translator controls how a civilization narrates its own dependencies, a captured Translator is how the prestige gauge gets miscalibrated in the first place.

Failure modes.

• Narrative capture — sensemaking sold to whoever pays, laundering interest as explanation.

• The romance of leadership — over-attributing system outcomes to a visible hero is a sensemaking failure at civilizational scale (Meindl et al.).

• Manufactured legitimacy — the affect of insight without the load-bearing frame; thought-leadership theater.

Supply & demand. Over-supplied at the viral end, under-supplied at the rigorous-explainer end — and AI is flooding the zone with synthetic plausibility, which raises the premium on a trusted human sensemaker.

How to do it / allocate into it. Translate without distorting; guard trust as the moat; remember the gauge you control points the next generation’s careers — aim it at the dependency gradient, not the prestige one.

10. The Critic / Verifier — catches the error and holds the standard

What it actually does. The Verifier says no — auditing, editing, reviewing, red-teaming, refusing what doesn’t meet the standard. It is the quality function the whole stack leans on, and in the AI era it is the role value is migrating into fastest: when a machine generates 99% of the work, the last 1% — verification — becomes the whole price (Karpathy).

How it works well.

• Treats challenging the senior as a duty, not impertinence — “Forceful Backup” trained as an obligation (Rickover).

• Refuses to “live with” deficiencies; verbatim standards, independent skepticism.

• Is structurally independent of the thing it checks — a captured auditor is no auditor.

Where it sits on the gradient. Historically under-honored, now repricing hard. The “no” role is resented in slow-feedback systems and revered in fast-feedback ones. AI exposed the deficit: 96% of developers distrust AI-generated code’s correctness, yet only 48% always verify it (Sonar) — and the randomized evidence shows the value already relocated to the reviewer, even as the hype pointed the other way (METR, 2025: developers 19% slower with AI, the worth moving to oversight).

Failure modes.

• Verification theater — the rubber stamp; the 48% who never actually check.

• Normalization of deviance — the verifier who slowly learns to stop seeing the warning once nothing has blown up yet (Vaughan, Challenger).

• Over-correction — blocking everything; the verifier who forgets the answer is a ratio, not a veto on all motion.

Supply & demand. Structurally under-produced — low status, high friction — and now the single clearest AI-era growth role, because as generation goes free, confidence in deploying it is the only thing left that’s scarce.

How to do it / allocate into it. Make the catch legible; treat the “no” as load-bearing infrastructure; where failure-feedback is slow, manufacture the check before reality does it for you.

11. The Cultivator — develops the humans the whole stack depends on

What it actually does. The Cultivator raises and develops human capability across generations — teacher, mentor, coach, parent, trainer. It reproduces the very capacity every other role draws on; without it the stack has no next generation of Thinkers, Doers, or Maintainers. Distinct from the Maintainer, who keeps systems alive; the Cultivator maintains people. This is the knowledge-transmission layer, and it is productivity-resistant by nature — you cannot speed up the formation of judgment.

How it works well.

• Develops judgment, not just skill — the integrator-of-knowledge who thrives where rules are unclear (Epstein, Range).

• Works on a horizon nobody else measures; the payoff arrives decades out.

• Builds developed habits (the trained “Forceful Backup,” the checklist discipline) that outlast any single lesson.

Where it sits on the gradient. The recognition-without-resources split at its most extreme. Society openly esteems teachers — their prestige rose +25 points since the 1980s (prestige data) — while underpaying them and watching education spend double against flat measured output, the textbook signature of Baumol cost disease (Mercatus). Esteem granted, resources withheld.

Failure modes.

• Credential theater — signaling over actual development.

• AI deskilling — automating the complex parts and leaving humans the routine, so capability quietly hollows out (Anthropic Economic Index: net deskilling for most roles).

• Measuring activity, not formed capability — optimizing test scores or contact hours instead of judgment.

Supply & demand. Chronically undervalued and under-supplied: care and development workers earn $13.51–$13.81/hr against a $27.31 average, 1 in 6 below the poverty line, 88–94% women (EPI) — even as demand surges with an aging population (nurse practitioner +40% this decade, BLS).

How to do it / allocate into it. Cultivate judgment, not throughput; accept the slow feedback as the moat (it’s exactly why the role is undersupplied); and treat re-prestiging development — Europe’s apprenticeship traditions included — as title arbitrage on the longest-horizon asset a civilization owns.

12. The Steward / Guardian — protects the long horizon and the commons

What it actually does. The Steward guards long-horizon and shared assets against depletion — the fiduciary, the trustee, the custodian of institutions and commons no single generation owns. Distinct from the Maintainer, who fixes the existing system; the Steward defends the future of the stack, intergenerationally. Its entire job is the slowest-feedback work there is, which is precisely why it is starved.

How it works well.

• Schedules upkeep against a horizon nobody is measuring — Rome’s standing 700-strong familia aquaria cleaned the aqueducts on a 1-to-5-year cycle for centuries (Frontinus).

• Acts on the long math: a dollar of prevention saves six to ten, and the deficiency costs a household $9/day against a $5.48 fix (ASCE).

• Manufactures feedback where reality won’t provide it in time — shortening the loop so neglect has a visible cost now.

Where it sits on the gradient. The most starved role of all, because stewardship is pure slow-feedback work and the prestige gauge can’t see a slow death. Deferred road-and-bridge maintenance has run net-negative almost every year since 2004 (Pew); infrastructure scores its best-ever grade, a C, atop a $3.7T gap (ASCE).

Failure modes.

• Deferral — nobody dies this quarter, so the bill rolls forward until it can’t.

• Normalization of deviance — the slow erasure of the warning signal (Vaughan).

• The maintenance-signature collapse — Rome’s aqueducts failed as stewardship, not engineering; Tainter’s law says collapse is the unpaid upkeep bill of accumulated complexity coming due (Tainter).

Supply & demand. Catastrophically under-supplied; the aging-out trades shortage — millions of positions unfilled by 2030, a multi-hundred-billion exposure (JLL via Fortune) — is the steward shortage in physical form. Intergenerational by definition, so the market never prices it correctly.

How to do it / allocate into it. Shorten the feedback loop so neglect costs something today; steward where the horizon is longest and the competition thinnest; this is contributive justice with a balance sheet attached.

The synthesis: a flourishing society is a ratio, not a winner

No role wins. The error the prestige gauge encodes is believing the question is which role is most important — when the real question is what is the mix. Pour a civilization’s most talented people into the celebrated Generator and Manager layers — as elite schools do — and by the iron logic of constraints, throughput cannot rise, because you’re feeding the part that was never the bottleneck; growth actually slows (Murphy, Shleifer & Vishny, 1991). Ecology has run this experiment for a billion years: remove the keystone and diversity collapses (keystone concept); remove the “lazy” reserve and the loss isn’t compensated, because slack is the buffer against shock (PLOS ONE, 2017). But refuse the over-correction too — maximize the Maintainer or Integrator and you get the services trap. The all-founders monoculture and the all-maintainers monoculture are the same mistake in opposite jerseys.

So treat the gradient as a gauge, not a fact of nature — and recalibrate it. You can’t do that with information; status-seeking is hardwired and follows cues, not arguments (Anderson, 2015). The only working lever is title arbitrage: find the load-bearing-but-illegible role, give it a legible high-status wrapper, and capture the talent the market mispriced. Palantir did it to the integrator; Google did it to the sysadmin (the rebrand to “Site Reliability Engineer” reframed maintenance as engineering and the same work now pays $205K–$768K — Seeking SRE). A society — or an institution, or a person allocating one finite life — that re-points status onto its Connectors, Enablers, Critics, Integrators, Cultivators, Maintainers, and Stewards first is the one that gets to keep flourishing. Everyone else is buying the celebrated role at the top of the bubble and wondering why the lights keep going out. Find the role nobody is bidding on yet. That is where the next century is actually built.