For the Innovators — Moral Restraint Moves Us From Greed to Generosity

Executive Summary

A Positive Model for Technology Stewardship

Technology will continue to amplify human capability. The question is whether the application-layer operators who deploy that capability optimize for engagement extraction — the greed model — or for genuine helpfulness — the generosity model. This paper argues for the second, sketches the mechanics, and describes the kind of builder the moment calls for.

• The application layer is a worldview layer. As capability rises, the app layer no longer merely extracts economic value — it embodies and directs an opinionated worldview.
• Technology has no duty of care. Medicine, law, finance, and engineering all owe an enforceable duty to the people they affect. Consumer software owes none.
• The result is innovation gone wrong. Platforms that don't uphold their own user's incentive — social media, job sites, dating apps — produce the opposite of what they promise.
• A few builders with a common profile shape the world. The inventors of every general-purpose technology share three traits: technical curiosity, a positive worldview motivation, and a distribution engine.
• The choice is greed or generosity. The same technology, tuned for engagement extraction or for genuine helpfulness, produces opposite outcomes — and the difference is the operator's, not the science's.

Section 1

The Importance of the Application Layer

The technology value chain has flipped. For most of computing history, advantage came from controlling the foundational layer — proprietary processors, proprietary operating systems, proprietary protocols. Today nearly all foundational developments are open source or available at very low cost. The hard parts at the bottom of the stack have been commodified.

Monetization no longer happens at the foundational layer. It happens at the application layer — where the operator chooses what to build, who to build it for, and what to tune for.

For years that was mostly an economic observation: the application layer was where value got captured, where attention was converted into revenue. That is still true, and it is no longer sufficient. As the underlying capability grows more powerful, the application layer has quietly taken on a second role. It is no longer merely an extractive economic layer. It is increasingly where a technology's worldview is chosen.

By worldview I mean something specific: a working set of assumptions about what a good outcome is, whom the product is for, and which human ends it should serve. A search box, a feed, a hiring filter, a chat assistant — each embeds a view of what the user is trying to do and what counts as helping them do it. When the underlying capability was thin, those assumptions stayed mostly hidden, expressed only in defaults and ranking weights. As the capability deepens, they surface, and the application layer can direct them with growing force.

The relationship is close to proportional. The more powerful the foundational capability, the more explicitly the application layer can embody and direct an opinionated worldview — for better or for worse. A recommendation engine can only nudge; a capable assistant can advise, decline, reframe, and persuade. That is why the application-layer choice, once mostly a question of business model, is becoming a question of formation: what kind of person, and what kind of public, the technology is quietly shaping us toward.

This is the most important property of the moment for anyone considering building. The advantage now sits with operators who understand how to deploy already-existing capability into already-existing markets — not with researchers who can produce a new foundational breakthrough. That makes the design questions, the user-empathy questions, and the incentive-design questions the binding constraints, not the science.

Every major technological leap of the last six centuries has the same five-layer structure beneath it: dormant components already exist; an inventor sees across silos and composes them; the resulting innovation only matters once an ecosystem ships with it; and the impacts on human life run downstream of all of the above. The shape is consistent enough to function as a forward-looking guide.

Section 2

A Missing Duty of Care

Every mature profession that can harm the people it serves has, over time, been made to answer for that harm. Consumer technology is the conspicuous exception. Those who build the software billions of people now live inside owe no enforceable duty to the people it shapes.

A duty of care is a simple idea with a long history: if your work can injure someone who relies on it, you are obligated to take reasonable care not to, and you can be held to account when you fail. It is not a mission statement or a code of ethics posted on a wall. It is a binding, external standard — one that outlives the good intentions of any individual practitioner.

The professions that carry real weight all have some version of it. Medicine begins with primum non nocere — first, do no harm — and enforces it through licensure and malpractice liability. Law binds attorneys to a fiduciary duty to their clients, policed by the bar. Finance holds advisers and brokers to fiduciary and suitability standards under regulators like the SEC and FINRA. Civil engineering requires licensure, attaches personal liability, and asks a named engineer to stamp the drawings — to put a signature on the claim that the bridge will hold.

The common thread is not virtue. It is accountability that does not depend on virtue. A physician who means well and harms a patient is still liable; an engineer whose bridge fails cannot answer that his heart was in the right place. The duty attaches to the role, not to the character of the person filling it.

Consumer technology has no equivalent. A team can ship a product to a hundred million people — many of them children — and owe them nothing of the kind a physician owes a patient or an engineer owes the public that crosses the bridge. There is no license to lose, no drawing to stamp, no standard of care a court can hold the product against. The result is a domain of enormous influence over human attention, relationships, and formation, operating almost entirely on the goodwill of whoever happens to run it.

This is the gap. The next section walks through what tends to grow in it — the specific ways technology has failed to uphold the interests of the very users it was built to serve.

Section 3

Innovation Gone Wrong — When Technology Doesn't Uphold the Incentive of Its End User

With no duty of care to hold the line, a product's incentives are free to wander from the people it serves — and across the platform decade, they did. Several of the most-used technology platforms of the past fifteen years have produced the opposite of their stated promise. The pattern is the same across categories: platform incentives drift from user welfare, the platform optimizes for what it can measure (engagement, paid postings, swipes), and the user gets the unintended consequence.

Social Media — Was Supposed to Make Us More Social

The U.S. Surgeon General's 2023 advisory on social media and youth mental health documents the opposite. Adolescents using social media more than three hours per day face roughly double the risk of poor mental-health outcomes including depression and anxiety. Adolescent loneliness has climbed alongside, not fallen against, the rise in connectedness the platforms were supposed to provide.

Job Sites — Were Supposed to Make Hiring Easier

Harvard Business School's 2021 study Hidden Workers: Untapped Talent documents the opposite. The largest job platforms screen out millions of qualified candidates through inflexible filters, drive employers into near-oligopoly contracts (over $2,000 per year per posting, often on three-year terms), and derive over 95% of revenue from companies paying to post roles. Job seekers respond by adopting marketplace-abusing "auto-apply" tools that degrade quality for everyone.

Dating Apps — Were Supposed to Make Finding a Partner Easier

Pew Research's 2020 survey documents the opposite. 47% of Americans say dating is harder today than it was ten years ago. The engagement-optimization model that works for ad businesses works against the user's stated goal in the dating context — every successful match removes two paying users from the platform.

The mechanism in each case is the same. Platform incentives are not aligned with user welfare. Platforms capture attention, convert it to engagement, convert engagement to revenue. The user experience is a means to that end, not the end itself.

None of this was inevitable, and none of it is a mystery of the technology itself. To understand how these loops get built — and how they might be built differently — it helps to look first at the small number of people who build them, and at what they have historically had in common.

Section 4

The World Is Uniquely Shaped by a Small Number of Innovators With a Common Background

The next chapter of technological progress will be shaped by who chooses to build. Looking back at the people who drove past general-purpose technologies — magnetic compass, printing press, steam engine, electricity, internal combustion, television, personal computers, smartphones, AI — the inventors themselves, spread across continents and centuries, share a surprisingly consistent profile.

• Modest Family Backgrounds. Few were born into wealth or power.
• Non-Traditional Education — often in areas adjacent to their eventual contribution, or no formal education at all.
• Early-Childhood Fascinations with their eventual area of interest. Edison had a lab at age ten. Gutenberg was obsessed with precision mechanics as a young man.
• Intense Hardships or Uncommon Scenarios in formative years.
• A Trifecta of Motivations that combined to produce sustained, focused work.

The trifecta of motivations is the part most worth attending to. It shows up consistently across the inventors of every general-purpose technology in the modern era.

The pattern becomes most visible in the people themselves. Four short profiles, spanning six hundred years.

The unifying theme is the same in each case: dormant components, an inventor with the trifecta of motivations, a commercial wedge that funds the early ecosystem, and downstream impacts that exceed anything the original inventor could have specified. The pattern is not just historical. It is the playbook for what builders in the next decade can do — if they bring the values layer. The third motivation can no longer be left to default to engagement-extraction. The values layer has to do more work than it did last time.

Section 5

A New Model — Moving From Greed to Generosity

The mechanism by which a technology platform causes externalities can be drawn as a simple loop with three nodes — Foundational Tech, the App, and the User. The same foundational tech can be wired into two very different loops depending on what the operator tunes for.

The difference between the two models is not the foundational technology — that's the same in both. Transformer architectures, recommender systems, and large-scale compute are accessible to either kind of operator. The difference is the tuning objective at the application layer. An engagement-tuned platform produces the greed model's outcomes; a helpfulness-tuned platform produces the generosity model's outcomes.

The greed model's failure mode is the one most of the platform decade has demonstrated. The app sits between the foundational layer and the user, and tunes its dials toward whatever metric correlates with revenue. Engagement is the default because engagement is the cleanest signal: more time on platform, more sessions, more notifications opened. The path from there to recursive addiction is short, because the model the platform has built of each user becomes more accurate every minute the user is on it. The externalities — pricing power, wealth concentration, mental-health damage — are not anomalies. They are the loop running as designed.

The generosity model breaks the loop at one specific point: what the app tunes for. The dials point at helpfulness — did the user accomplish what they came to do; did the product save them time; did it reduce a problem instead of producing a new one. Profit is downstream of that, not upstream. A helpfulness-tuned platform can still be highly profitable. It is simply optimizing for a different signal, and the difference compounds.

This is a choice the operator makes. It is constrained by capital markets, by competitive dynamics, and by the cost structure of running platforms at scale — but it is a choice. Several existing companies sit closer to the generosity model: Wikipedia, Signal, certain B Corps, some open-source maintainers, parts of the developer-tool ecosystem. The argument of this paper is that more should — and the next section is the playbook for how.

Section 6

What to Build — The Generosity Operator's Playbook

The argument so far is that the application-layer operator has a choice the foundational technology does not. Here is what that choice looks like in practice — the moves a generosity-model builder actually makes, and the moves they refuse.

Choose a Wedge Where Profit and User Welfare Align

The greed model's structural problem is that revenue grows when the user's stated goal recedes. Engagement extraction is the cleanest example: each successful match removes a dating-app user, each closed tab removes a social-media user, so the platform tunes against its own users' purposes. Generosity-model wedges run the other direction. Subscriptions a user renews because the product worked. Transaction fees that scale with the user's success (Stripe, Shopify, marketplace cuts paid only when the seller sells). Enterprise SaaS where retention is the metric. Paid software where you own what you bought. These are not exotic business models — they are the older models the engagement economy displaced. Choosing one of them is the first and most consequential decision a generosity operator makes.

Measure What the User Is Trying to Do

A generosity operator builds reporting infrastructure that tracks the user's actual outcome — minutes saved, problems solved, jobs landed, partners found, sleep recovered — and shows it back to the user. The greed model tracks engagement and hides everything else; the generosity model treats user outcomes as the product's reason for being and makes them visible. This sounds soft. It is not. It is the variable that changes what the team decides to ship next quarter.

Audit Your Externalities

Every product creates costs the user does not see on the invoice — time displaced, sleep lost, attention fragmented, anxiety produced, money spent against value received, friction generated for the people the user lives with. The greed model treats these as off-balance-sheet, since they do not show up in the platform's revenue. The generosity operator inverts this. Externalities are measured, named, and reported back to the user. A social platform reports the share of users whose nightly sleep was displaced by the product. A subscription service reports the share of users who paid for a month they did not use. A marketplace reports failed transactions, not only completed ones. This is the surest test of whether the incentives are actually user-aligned: when the operator cannot in good conscience publish the externality numbers, the loop is not generosity.

Treat Profit as Oxygen, Not the Goal

The generosity model does not need less profit. It needs profit that is sufficient to sustain the firm and its mission, not maximized at the user's expense. Practical versions: caps on shareholder distribution, reinvestment thresholds tied to product quality rather than headcount growth, public-benefit corporation structures with teeth, steward-ownership trusts (Patagonia, Bosch, Zeiss). Each of these pre-commits the firm to a model the next CEO cannot quietly walk back when public-market pressure builds.

Pick the Capital Structure That Protects the Model

The most common failure mode for a well-intentioned founder is not bad intent — it is good intent overwhelmed by capital structure. A firm that has accepted growth-stage venture funding cannot easily refuse the engagement-maximization playbook; the cap table has already voted. The generosity operator either takes slower capital from the start (revenue-financed, founder- and community-owned, mission-aligned LPs, family offices with long horizons) or chooses a legal form — public-benefit corporation, perpetual-purpose trust — that binds the company to the mission across ownership changes. Patagonia is the live example of this done well at scale.

Refuse the Dark Patterns by Default

A short and non-exhaustive list of features the generosity operator does not ship: autoplay, infinite scroll, push notifications optimized for return rather than for the user's stated need, dark UX on cancellation, hidden pricing, default opt-in to data sharing, gamification of variable rewards, "streak" mechanics on platforms that are not actually about practicing a skill. Each is a small choice that compounds. Generosity-model firms keep a written list of what they will not build and re-publish it as the team grows.

Build for Durability Over Scale

Not every good business should be a category killer. Several of the most generosity-aligned operators of the last three decades — Wikipedia, Signal, Basecamp, certain B Corps, parts of the developer-tool ecosystem — chose durability over hyperscale. They serve a constrained market well, sustain themselves on the revenue that market produces, and refuse the exit path that would compromise the model. The greed model treats this as a failure of ambition. The generosity model treats it as evidence the operator knew what business they were in.

Closing

The Question Is the Choice

The amplification will keep growing. Each foundational layer that opens — transformers in 2017, multimodality in 2023, agentic action in the years just ahead — drops the floor of what an application-layer operator can deploy. The question is not whether to build but who builds, and under whose incentive structure.

The greed model is the default. It is what falls out of unconstrained engagement optimization. It is what venture capital tends to fund, what public markets tend to reward, and what the major firms have visibly chosen. The generosity model is a deliberate departure from that default — operators willing to take less profit in exchange for less harm, capital allocators willing to fund them, regulators willing to constrain the most extractive applications, and a broader culture willing to recognize the difference.

Picture two twenty-five-year arcs. In the first, the application layer continues to optimize as it has, the externalities continue to accumulate, and the firms that own the loops continue to consolidate wealth, attention, and intelligence. In the second, a new class of operator emerges — fewer in number than the greed-model firms but commercially serious, capital-structured to last, and visibly building products users would defend in court if asked. The technology is the same in both arcs. The composition of who builds, and on what terms, is what determines which one we get.

The technology we are building can amplify either purpose. It is amplifying both right now. The next decade decides which amplification dominates the rest of the century.

Appendix A

References and Source Data

• Smith, A. (1776). An Inquiry into the Nature and Causes of the Wealth of Nations.
• Heritage Foundation. (2025). Index of Economic Freedom.
• Our World in Data. Life expectancy 1900–2021; infant mortality 1900–2020; illiteracy rates 1820–2020; extreme poverty 1820–2020; global democracy 1820–2020.
• U.S. Surgeon General. (2023). Social Media and Youth Mental Health: A Surgeon General's Advisory.
• Kwa, T., et al. (2025). Measuring AI Ability to Complete Long Tasks. Frontier AI time horizon doubling ~7 months.
• Hassabis, D. (2025). Interview in WIRED Magazine: 5-year time horizon for AGI.
• Harvard Business School. (2021). Hidden Workers: Untapped Talent.
• Pew Research Center. (2020). The Virtues and Downsides of Online Dating.
• Gerlich, M. et al. (2024). AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking.
• Gomez-Uribe, C. A., & Hunt, N. (2015). The Netflix Recommender System (Netflix Inc.): recommenders drive 80% of watch time.
• Priori Data. (2025). Global average social-media use: 2h23m/day.
• Washington Post. (2020). How the Coronavirus Pandemic Helped Floyd Protests Become the Biggest in U.S. History.
• S&P 500 market-cap and top-17-tech market-cap data: SEC filings, 2000–2024 (tickers: AAPL, AMAT, AMD, AMZN, AVGO, CRM, GOOG, IBM, KLAC, LRCX, META, MU, MSFT, NFLX, NVDA, ORCL, PYPL, QCOM, TSLA).
• Buyback data for software-tech sample (AAPL, EBAY, GOOG, META, MTCH, NFLX, ORCL): SEC filings, 2000–2024.

Companion Projects in the EconFaithAI Series

• TechConcentration — wealth concentration of the tech sector in U.S. equity markets, 1900–2025.
• TechReligiousProfile — religious composition of tech CEO leadership and workforce.
• AttentionAndIntelligence — recommender-driven attention concentration + supercomputer Rmax concentration.
• ChildhoodInfluence — share of formative influence on U.S. children, 1900–2020.
• MachineVsHumanIntelligence — frontier AI IQ trajectory and population scale.
• FruitOfSpiritTech — spiritual audit of 124 technologies across the nine fruits of the Spirit.
• PrayerAndFasting — two-millennium trajectory of Christianity's two foundational disciplines.