The previous two posts in this series ended on a closing thought and a held breath. The Shrinking Layer described three phases of displacement. What the Work Was For argued that the secular discourse has no framework for human purpose independent of productive output, and that the oldest framework we have in the Western tradition does.
Neither post told you what to build.
This one does. Not as a moral treatise. As a product spec.
If human worth is intrinsic and the four verbs from last week are create, relate, steward, and rest, then each of those is a category of work that becomes more valuable, not less, as autonomous agents absorb the rest. The market does not know it yet. The builders who move first in these categories are setting the defaults for the next decade of how humans and agents share a world.
I am not going to argue the theology here. I already did. What follows is the engineering shape of the argument, with numbers, names, and the product decisions that embody the thesis.
The frame: four verbs, four categories
Think of the four verbs as four categories of consumer and commercial demand that are structurally under-served right now and will be over-served eventually, in that order. Early movers define the shape. Late movers build thin wrappers over defaults somebody else set.
Each of the four has:
- A real, measurable shift in demand that is already visible in 2026 data.
- A set of incumbents whose product assumptions predate agent displacement.
- A design principle that disqualifies the obvious approach.
- A handful of concrete product shapes that would embody it.
I will walk through each.
Create
The Cognizant January 2026 report puts 93% of US jobs in AI's exposure zone, six years ahead of the 2024 forecast, with the low-exposure share of the workforce collapsing from 31% to 7% in two years. The average exposure score is 39%, 30% higher than what was originally forecast for 2032. The non-automatable share of tasks has fallen from 57% to 32%. What remains is disproportionately the work that cannot be separated from a human author.
At the same time, the creator economy has crossed $200 billion globally, with credible projections to $480 billion by 2027 and $1.35 trillion by 2035. Goldman's forecast is the less bullish one. Over 200 million people worldwide now identify as content creators. That is not a niche. That is a structural reallocation of economic activity away from jobs-of-output toward jobs-of-signature.
The design principle: signature beats scale. When agents can produce infinite competent output, the scarce asset is a specific human's taste applied to a specific problem, with provenance that can be verified.
Provenance infrastructure. C2PA and the Content Authenticity Initiative already have 6,000+ institutional members including Adobe, Microsoft, Google, and Meta. EU AI Act Article 50 enforcement begins August 2026. Samsung and Google sign photos natively. The category is being built. Most of the plumbing is not.
Human-as-conductor tooling. Not "agent makes the thing" and not "human makes the thing alone." Human directs a fleet of agents, signs the output, owns the result. Every creative tool on the market assumes one or the other. The seat between them is empty.
Bespoke markets. Etsy assumed "find a small batch of the thing you want." The post-agent assumption: "the thing you want does not exist yet because nobody has bothered to make your specific version." The product gap is the marketplace that trades on maker identity as a first-class field.
Craft transmission. Six in ten Gen Z say they will pursue trade work. Electricians under 30 are making $240,000 on data-center builds. The DOL announced $145 million in apprenticeship grants. The product gap is the platform that matches tacit-knowledge holders to apprentices at scale, because the ten thousand crafts that are one funeral away from being lost cannot wait for the existing training ecosystem to catch up.
Matthew Crawford's "It is by having hands that man is the most intelligent of animals" (via Anaxagoras) was right in 2009 when he wrote it and is more right now. "The satisfactions of manifesting oneself concretely through manual competence seem to relieve him of the felt need to offer chattering interpretations of himself to vindicate his worth. He can simply point: the building stands, the car now runs, the lights are on." When chattering interpretations are free and infinite (agents produce them on demand), the pointing becomes the value.
Relate
The US Surgeon General's 2023 advisory documented that half of American adults experience loneliness; 8% have no close friends; hours spent socializing in person have fallen 20 per month since 2003, while hours spent alone have risen 24 per month. The health cost alone is $6.7 billion in excess Medicare spending from elder social isolation. Poor social relationships raise heart disease risk 29%, stroke risk 32%, and dementia risk among the elderly by roughly 50%.
Meanwhile, the coaching industry is at $5.34 billion globally, up 62% since 2019, with 122,974 professional coaches worldwide (up 54% in six years). The experience economy is projected at $2.58 trillion by 2035. US consumers are directing a near-record 25% of their budget at services tied to experiences. Music festivals, sports, restaurants, gatherings, coaching, therapy, and live events are eating discretionary spend.
Put it together: people have less relational infrastructure than at any point in measured American history, and they are paying a record share of their income to get some of it back. That is not a crisis of desire. That is a crisis of product shape.
The design principle: presence is the product. Not content about presence. Not digital substitutes. The actual fact of a room with other humans in it, at a specific time. Any product that claims to deliver this via a feed is not delivering it.
Gathering infrastructure. The toolkit for hosting a dinner, a salon, a circle is fractured across ten apps and none of them treat the host as the first-class user. The product gap is the operating system for the person hosting, measured by how reliably it turns vague relational intent into a recurring embodied reality.
Chapter-based communities. 67% of community members say they stay for shared identity, not algorithmic relevance. The product opportunity is the infrastructure that lets mid-size organizations (YPO, alumni networks, churches, guilds, trade unions) run their local chapters well, with embodied presence as the primary metric and digital as scaffolding. Putnam's Bowling Alone data showed the 1965-2000 collapse of civic associations. The digital decade did not revive them. The post-agent decade might, if the product exists.
Care professions. 6.1 million home-care openings projected by 2034. 800,000 seniors on waitlists today. Aide turnover at 70-80% in the first 100 days. Every hour an aide spends charting is an hour she is not present. The product gap is the agent-assisted care platform where agents do the paperwork and the human does the work that cannot be automated: sitting with a dying person, remembering the name of the dog in the photograph. The dignity is in the presence. The agent is the administrative tax.
Trust networks. When anyone can generate a résumé, a portfolio, a reference letter, the value of a handful of real humans saying "I vouch for her" rises exponentially. The hiring, lending, housing, and dating markets all run on vouching already. None of them have proper infrastructure for it.
Sherry Turkle's line was "we expect more from technology and less from each other." The next decade's winning products for this category will expect less from technology and more from each other, and they will be the ones that do the unsexy scaffolding work (calendars, payments, logistics) so that presence becomes the default rather than the heroic exception.
Steward
The global AI governance and compliance market was $2.2 billion in 2025, projected to $2.55 billion in 2026, and one forecast from market.us puts it at $68.2 billion by 2035 at a 39.4% CAGR. Gartner calls it a billion-dollar category by 2030. Sixty percent of Fortune 500 companies used AI auditing solutions in 2024. The EU AI Act, California SB 942, sectoral requirements in finance, healthcare, and hiring, and emerging US federal frameworks are all forcing the market open. Credo AI, Monitaur, 2021.AI, Fairly AI are early entrants. This is the layer-two vocation Arendt would recognize: not the work itself but the human accountability for it.
Meanwhile, data-center environmental cost last year was $25 billion in health and environmental externalities, with $3.7 billion tied directly to AI. AI water footprint is projected at 312 to 765 billion liters in 2025. Each 100-word AI prompt consumes roughly a water bottle's worth. The environmental bill for the agent economy is being written now, and there is no market-side mechanism yet to price it properly.
The design principle: accountability needs to be technical, not ceremonial. An ethics board that meets quarterly is a performance. A tool that logs every agent decision and puts a name next to every consequential action is stewardship.
Agent evaluation. The missing piece in most deployments is not the agent. It is the evidence the agent did what you told it to do. Evals-as-a-service, drift detection, continuous alignment monitoring. The productized, compliance-grade, regulator-ready surface is still early.
Decision provenance. What data trained the agent? What decisions has it made? Which humans are accountable for which classes of output? C2PA handles content provenance at the artifact layer. The agent-decision layer needs something equivalent.
HITL orchestration. Not "human clicks approve." Human receives a compressed summary of what the agent wants to do, with alternatives and the cost of being wrong, at the moment when the decision is still reversible. This is the dashboard every regulated vertical will need.
Environmental accounting. The kWh, the liters, the carbon, priced per agent invocation. Nobody wants to build this because it makes the product look expensive. Someone will, first as a moat for responsible operators, then as a compliance requirement.
Tacit knowledge preservation. Every skilled worker who retires without transmitting what they know is a library burning. The product shape is the tool that lets a master capture knowledge as a byproduct of doing the work, with agents handling translation and indexing so the master does not have to stop being a master in order to teach.
Wendell Berry's distinction was that the exploiter wants to earn as much as possible by as little work as possible; the nurturer expects a decent living from his work, but his characteristic wish is to work as well as possible. The steward builds the tool that measures and rewards the second posture, not the first.
Rest
US smartphone users check their phones 205 times a day. Thirty-five percent of UK adults are actively reducing screen time. The dumbphone market is over $10 billion and grew 25% in 2025. The Light Phone III ships at $699 with a 5-10 year repairable lifespan. Mental-health and contemplative-practice app markets are in the double-digit billions. Adolescent depression screening hit 19.2% positive in 2025, the highest ever recorded.
In a productivity-first economy, rest is theft. In a post-agent economy, where productivity is free, rest is what separates humans who are still humans from humans who have outsourced their selfhood. Josef Pieper saw this coming in 1948. "Leisure has been, and always will be, the first foundation of any culture." "When we really let our minds rest contemplatively on a rose in bud, on a child at play, on a divine mystery, we are rested and quickened as though by a dreamless sleep." The market is finally catching up to a claim that was strange and unfashionable for most of the 20th century and is now a $10 billion category by accident.
The design principle: intentional friction is the feature. Products that succeed in this category refuse to do things other software does: refuse to notify, refuse to load fast, refuse to show a feed, refuse to be present during specific hours. The friction is the entire reason the product exists.
Reduced devices. The Light Phone, the Brick, the Boox readers, the Freewrite. The Kindle at its best is this; the Kindle after the ad layer is not. The job-to-be-done is absence, not capability.
Time and place ritual. The infrastructure for someone to genuinely be unreachable between Friday sundown and Saturday sundown, for a family to eat dinner without a device in the room, is shockingly thin. Monastic Rule-style products that help a group agree on and enforce a time discipline, with real lock-outs, are a category that does not exist yet and should.
Hobby discovery. Board games are a $22 billion industry growing faster than video games. The plumbing for finding the league, the chess night, the knitting circle is terrible because the advertising-driven internet has no incentive to reduce your time online. The gap is the discovery layer paid for by hobby operators who want more participants.
Retreats and camps. Fifty thousand operators in the US and Europe, each running on Airtable, Mailchimp, and handwritten intake forms. Somebody will build the Shopify for retreats.
Recovery from agent overuse. The market I am the least sure about and the most sure will emerge. It will look like a cross between a gym and a rehab. It will be a real category by 2030.
The Monday-morning test
Here is the question founders should ask themselves first thing Monday, regardless of which of the four they are aiming at:
What does my product do with the hour it saves the user?
If the answer is "it gives them back to the user for rest, relationship, creation, or stewardship," that is a product that works with the four verbs. If the answer is "it lets them generate more output," that is a product that is working against them and will eventually lose, because output is the thing the agents are taking over anyway.
The corollary:
What does my product do when the user is wrong?
If the agent just does what it is told, the product is commodity. If the product pushes back, asks, surfaces the cost of being wrong, refuses on alignment or ethical grounds, or escalates to a human who is accountable, the product is doing stewardship work that commodity agents cannot. The $1,000 human-in-the-loop threshold is a crude first version. The sophisticated version is the product that knows when the user is about to damage a relationship, burn out a community, or violate a regulatory line, and treats that moment as a product surface, not a bug report.
What the market will not tell you
Every category in this essay is underserved because the measurement systems that capital allocators use cannot easily see them. Venture capital is optimized for revenue growth curves; most of the work described here is high-retention, slow-compounding, emotionally-load-bearing, and geographically fragmented. LTV is enormous and CAC is irreducible. Growth is slow and durable rather than viral and fragile.
This is not a coincidence. The reason these categories are underbuilt is exactly the reason they will matter: they are the categories where the work cannot be abstracted away from the person doing it. The agents cannot take them. The scaling laws do not collapse them. The hedonic treadmill does not consume them. And the people who need them are every human being alive, at some point, and most of them right now.
When autonomous agents saturate every category where human output was the product, the categories where human presence is the product become the entire economy above a certain line of necessity. We are building the infrastructure for that economy now, one product decision at a time.
"He who has a why to live for can bear almost any how." The how is changing faster than any of us expected. The why has not moved at all. The product builders who align their Monday mornings with the why are building the economic form of the world that is already arriving.