The Right to Begin
Accessible intelligence widens the circle of people who can attempt consequential work. Judgment, trust, and the ability to finish determine what that freedom becomes.
Compare heat exchanger geometries in simulation. A promising design is an option to investigate.
Opening
An unfamiliar problem can stop a person before the work begins. The literature is inaccessible. The vocabulary belongs to another profession. The first calculation requires knowledge that would take months to acquire. Before an idea has encountered a serious objection, the cost of finding out whether it deserves attention has already defeated it.
That threshold is moving. A person can ask for an explanation, interrogate it, examine the underlying research, and develop a first design. A small team can investigate a question that would once have required a larger technical staff. Software agents can carry authorized parts of the inquiry forward. More people can reach the point at which the merits of their work become visible.
My outlook begins with this expansion of agency. People should be able to take an unfamiliar problem seriously before an institution has decided to take them seriously. The opportunity is larger than faster completion of familiar tasks. It includes work that previously would never have been attempted, by people who previously would never have considered themselves capable of attempting it.
The consequence is a change in what earns authority. Credentials, organizations, and reputation still matter, but more people can arrive with something concrete to examine: a calculation, a prototype, a reproducible result, a customer who needs the thing to work. As the cost of producing a plausible proposal falls, the ability to judge it and carry it through becomes more consequential.
The right to begin should become more widely available. The responsibility for the result should become more explicit.
A higher floor, earned competence
“Anyone can do anything” is a useful conviction when it encourages someone to learn the unfamiliar discipline, build the first prototype, and discover the limits through serious effort. Its force lies in refusing to mistake an inherited starting point for a permanent boundary.
Useful cognitive assistance can raise that starting point. It can make explanation, analysis, and technical support available when a person needs them. This is what I mean by a higher capability floor: a change in access to help and in the work someone can attempt, rather than a sudden biological change in population intelligence.
A 2025 study of AI assistance in customer support offers a bounded example. Among the workers studied, issues resolved per hour increased by 15% on average, with larger gains among less experienced and lower-skilled workers. The most experienced workers saw smaller speed gains and some declines in quality.1 Within that setting, useful practices became more accessible to people who had not yet acquired them through years of experience. The uneven results also show why access to the same tool need not produce the same benefit.
The invitation to begin carries an obligation to learn. A persuasive explanation can be wrong. A design can conceal a bad assumption. A newcomer needs feedback, practice, and enough contact with consequences to recognize where an expert must take responsibility. Education, language, time, money, and confidence continue to shape who can benefit from assistance.
Expanding agency therefore means building routes into competence. Someone should be able to move from understanding a question to conducting a useful investigation, then to earning progressively greater responsibility. The value of assistance grows when it strengthens that progression. A tool that helps a learner recognize uncertainty may contribute more than one that makes every answer sound equally assured.
This changes the starting question for an ambitious undertaking. Before assembling a substantial organization, ask what the work actually requires.
The organization has to earn its size
People seek groups for belonging, support, and security. Those human needs endure. The economic reasons for adding people to a project are more open to revision.
Some work requires complementary disciplines, continuous coverage, independent scrutiny, or many hands in the same place. Some organizations grew because information was difficult to retrieve and coordination was expensive. When tools change those costs, inherited headcount and reporting structures deserve another look.
In a field experiment at Procter & Gamble, individuals using generative AI matched the performance of teams without it on product-innovation proposals. Assistance also helped bridge participants' technical and commercial perspectives, while human judgment remained valuable in evaluating ideas.2 This does not establish that an individual can replace any team. It gives substance to a narrower and more useful question: which benefits of a larger group can now be supplied through a different arrangement?
A compact team can keep decisions close to the problem and make responsibility easier to locate. It must still contain, or obtain, enough expertise to recognize a failure and enough independent judgment to challenge a persuasive mistake. Reducing coordination is valuable only if the resulting organization can deliver dependable work.
The measurement has to include the difficult final steps. METR's early-2025 experiment found that the studied experienced developers took 19% longer with AI tools available. Its February 2026 update suggested improvement, but selection and measurement problems prevented a reliable estimate of the newer effect. A later survey examined workers' self-reported gains, a different kind of evidence.3 More drafts, code, or agents can bring more supervision. Completed work after review, correction, and integration is the meaningful unit.
These changes expose incumbents whose processes preserve yesterday's costs. An entrant can begin with a clean slate: fewer inherited handoffs, fewer departments to defend, and a workflow designed around present capabilities. Where the old process exists largely to manage expensive information, that freedom can be a substantial advantage.
The productivity J-curve research explains why access to technology alone is insufficient. General-purpose technologies require complementary investment in processes, skills, and other intangible assets before their benefits fully appear.4 The work must be redesigned around the capability. Simply placing a new tool inside an unchanged approval chain can leave the old constraint intact.
Incumbents also possess things that are hard to reproduce: customers, distribution, equipment, permissions, data, and operating knowledge. A procedure that looks cumbersome may encode a lesson learned through an expensive failure. The strongest established firm can combine those assets with a willingness to rebuild. The strongest entrant can distinguish obsolete assumptions from necessary discipline.
I expect opportunity wherever a capable person can see that distinction clearly. Organizational size should follow the requirements of useful output. It should never become a substitute for ambition.
Attention opens the door; evidence carries the promise
When more people can make something, more people must find a way to be discovered. A product needs a customer. A capable person needs others to recognize that capability. The cost of drafting, coding, and prototyping can fall while customer acquisition, testing, insurance, manufacturing, and institutional acceptance remain formidable barriers.
Distribution and personal branding consequently deserve serious attention. A recognizable body of work helps someone understand what a person knows, how that person reasons, and whether promises survive execution. Visibility can bring previously hidden competence into view. Fame can open a door that excellent private work never reaches.
The relationship then needs evidence. A consequential promise should be supported by something inspectable: a reproducible result, a delivered product, a reference with direct knowledge, or a record of decisions and corrections. As polished output becomes easier to produce, its appearance carries less information about the work behind it. Provenance and responsibility become more valuable.
The practical importance of reaching demand is already visible. In the Federal Reserve Banks' 2025 Small Business Credit Survey, published in 2026, reaching customers and growing sales was the most commonly reported operational challenge.5 The survey does not attribute the difficulty to AI. It shows why the ability to create and the ability to reach a buyer must develop together.
Consider how a new supplier might earn a first serious conversation. A polished claim announces a capability. A documented test explains the conditions under which it worked, the limits encountered, and the person who will stand behind the next delivery. The second gives a buyer something on which to base a decision. Over time, precise promises, visible corrections, and repeated delivery can support greater trust and greater responsibility.
That discipline becomes even more important when the first party examining an offer is software acting for a customer.
Serve people through the agents they authorize
An agent can search specifications, compare offers, run a permitted analysis, schedule work, or prepare a transaction. A person defines the purpose; software performs parts of the activity. An organization can authorize repeated tasks without placing an employee at every interaction.
Products and services should accommodate this form of use. Specifications need to be understandable, prices and conditions explicit, interfaces reliable, and transaction records inspectable. A useful offer should remain intelligible as it moves between a buyer, an agent, and a supplier.
Authority belongs in the design. NIST's proposed October 2026 tenets for agent identity emphasize distinct identities, traceable links to the people or organizations initiating work, bounded delegation, and revocable permissions.6 These principles help frame the essential questions: who authorized the action, what could the agent do, and how can that authority be withdrawn? People also need a way to revise instructions, recover from mistakes, and obtain direct human service.
The number of software processes can grow very differently from the human population. That creates scope for far more interactions, but every useful agent still draws on purposes, budgets, permissions, and computational resources. Starting another process creates no independent purchasing power.
Demography reinforces the need for precision. Many countries are aging or shrinking, while the UN's 2024 projection places the global population peak in the mid-2080s, with substantial differences between countries.7 Serving agents does not depend on a world already in population decline. The case follows from each person or organization being able to initiate more activity through software.
The consequential expansion is therefore in intelligence-users and intelligence in use: more humans with useful assistance, and more agents working on their behalf. What matters economically is what those users can now attempt.
More attempts must earn their next commitment
Imagine a small engineering team investigating a better heat exchanger. Previously, the cost of specialist time made a broad search impractical. With more affordable analysis, the team can examine the literature, generate designs and schematics, compare simulated geometries, and propose tests. It can investigate applications it would once have set aside.
This is a hypothetical example of the demand mechanism at the center of my outlook. Cheaper intelligence can expand participation in engineering, prototyping, research, science, and invention. It can also increase the ambition of each participant. More useful work may become worth attempting because its initial cost has fallen.
The stages matter. Generating a candidate consumes computation and electricity. Funding and validating an experiment adds laboratory capacity, samples, instruments, fabrication, and skilled attention. Deploying a successful result adds customers, finance, materials, supply chains, permissions, installation, and maintenance. Each larger commitment should require stronger evidence.
For the heat-exchanger team, good simulation could eliminate several unnecessary prototypes. It could also reveal a worthwhile application that would otherwise have remained unexplored. If a selected design survives a funded test and attracts a buyer, it creates work for fabricators, installers, and operators. Fewer resources per successful development and more developments worth pursuing can occur together.
This mechanism supports an opportunity thesis, not a universal prediction that resource consumption must rise. Tools can substitute for existing work, reduce material intensity, eliminate waste, or produce noise. Some savings will be retained. Many proposals should never receive a physical test. Agents cannot outrun their principals' budgets merely by generating options. The IEA's 2026 energy analysis similarly treats efficiency, uptake, and new applications as interacting forces in AI electricity demand.8
Selection is productive work. A good evaluation prevents an unpromising idea from consuming a laboratory's time. A well-designed experiment gives a promising idea a fair chance to earn evidence. A clear customer requirement keeps a technically interesting result connected to an actual use. The opportunity lies in improving the whole passage from possibility to commitment.
Take responsibility for what comes next
More proposals make capable operators more consequential. Someone has to identify the real dependency, assemble the necessary skills, define the test, decide what the evidence warrants, and remain accountable when the plan changes. Analysis can support this responsibility. It cannot assume it.
The intuition behind an “infinite money situation” concerns relative scarcity. Substantial pools of capital can coexist with too few credible people able to turn a promising idea into completed work. Capital itself remains finite, costly, and unevenly accessible. In the Federal Reserve Banks' survey, 42% of financing applicants obtained the full amount sought; the survey uses a weighted convenience sample.5 The presence of money somewhere in the economy does not make it available to every worthy project.
A capable operator makes the next decision intelligible. What need is being served? What has been demonstrated? What remains uncertain? Which milestone justifies further spending? This is how technical judgment, finance, and execution meet.
For an individual deciding where to direct effort, the discipline is practical. Find a concrete need. Learn enough to investigate it honestly. Reconsider the inherited process. Make the work visible and earn trust. Understand who will use it, including the software acting for them. Search existing knowledge for what has been missed. Choose a result worth taking responsibility for.
I want a world in which more people can reach this starting point, and in which their best work has a credible route into use. The freedom to attempt something consequential should not depend on first assembling the institution that would once have been necessary to attempt it.
Begin with the question you are now able to ask. Earn the evidence that permits the next step. Build the judgment and relationships needed to carry it through. A wider field of human agency becomes valuable when more people can finish something that matters.
Sources
Empirical references are carried forward from the source review dated 2 November 2026. They were not newly reverified for this edition. The outlook and hypothetical heat-exchanger example are interpretive; study results retain their stated scope.
- Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond, Generative AI at Work, The Quarterly Journal of Economics, 2025. Study of 5,172 customer-support workers; effects varied by experience and skill. ↩
- Fabrizio Dell'Acqua et al., The Cybernetic Teammate: A Field Experiment on Generative AI and Teamwork, Organization Science, published online 12 June 2026. Procter & Gamble product-innovation proposals, functional perspectives, and human evaluation. ↩
- METR, We are Changing our Developer Productivity Experiment Design, 24 February 2026; Measuring the Self-Reported Impact of Early-2026 AI on Technical Worker Productivity, 11 May 2026. Experimental estimates, later measurement limitations, and self-reported gains are distinct evidence. ↩
- Erik Brynjolfsson, Daniel Rock, and Chad Syverson, The Productivity J-Curve: How Intangibles Complement General Purpose Technologies, American Economic Journal: Macroeconomics, January 2021; authors' NBER working-paper version, revised January 2020. ↩
- Federal Reserve Banks, 2026 Report on Employer Firms: Findings from the 2025 Small Business Credit Survey, 3 March 2026. Customer acquisition and financing findings; financing percentages concern applicants. Weighted convenience sample. ↩1 ↩2
- NIST, Design Tenets for Agent Identity Management Systems, 27 October 2026. Proposed principles for identity, delegation, bounded access, and revocation. ↩
- United Nations DESA, World Population Prospects 2024 — Key Messages, July 2024. Projected global peak in the mid-2080s and differing national trajectories. ↩
- International Energy Agency, Key Questions on Energy and AI — Executive Summary, April 2026. Interaction of efficiency, uptake, and changing applications. ↩