How would you build a working life worth having by 2066, when AI can do much of today's work?

By 2045, assume systems reliably perform most routine cognitive work and a wide band of physical work in structured settings. That is plausible if compute, robotics, and energy keep getting cheaper. Human work concentrates where taste, trust, accountability, and presence still matter: negotiating, teaching face to face, caring for people who want a human, making things whose value includes that a person made them, running small organizations, fieldwork, sports, performance, and the messy front end of science. What I want in place is cheap AI that individuals and cooperatives can direct without a platform's permission, plus legal clarity that people own the output of agents they steer. A floor of basic services is plausible as backstop. Treating the absence of a traditional boss as personal failure is the habit to drop. Much of this picture is a bet on technical progress and on not regulating the tools into a few incumbents' hands only.

By 2066 the bet is larger. If systems can do most current professional work end to end, the worthwhile working life looks less like a career ladder and more like a portfolio of projects: joining research, restoring land, building a company that serves a niche big models ignore, raising children, competing, making art people want because of the human signature, governing a town, or taking hard jobs in space, oceans, or disaster response where embodiment and judgment still dominate. Security comes from broad ownership of productive AI capital plus revenue from human-scarce activities. I do not assume everyone becomes a scientist or artist. I assume a wide menu, including a quiet local life, without stigma or destitution. That open menu is the goal. Timelines are bets; the direction is the claim I will defend.

What I would build first is the ownership and access layer, not a grand program for manufacturing meaning. Concretely, legal and technical infrastructure so a person or small group can deploy AI agents, keep the surplus, and sell or share results on open markets. That means interoperable agent standards, portable identity and payment rails, clear default ownership of outputs, and competition policy that stops a handful of firms from owning the only capable models. Private capital pays for much of the rails because the tools will be profitable. Public investment, funded by taxes on AI economic rents rather than heavier taxes on labor, covers the parts markets underprovide. How any person spends their time and their agents is that person's decision. Safety constraints and the tax are decided by legislatures and courts, in public, with sunsets so bad rules can die. I would not hand the definition of a worthwhile life to a ministry or a model.

One thing I would refuse to build, even if it worked, is a state or corporate system that assigns people purposes, scores their activities for meaning, and dispenses status or money accordingly. It might reduce some misery. It would also turn life into compliance. Meaning issued from above is not meaning. I would likewise refuse neural or behavioral systems whose job is to make people satisfied with having no agency. Comfort without the possibility of effort and failure is a smaller life, and I would not engineer it.

The step most likely to fail is the politics of the transition. Gains concentrate, displaced workers do not quickly find new roles, and fear produces either freeze-regulation that locks in incumbents or crude redistribution that arrives late and breeds dependency without rebuilding agency. Plan B is to make individual and small-group productivity so high, through cheap AI tools and low barriers to starting things, that many people can generate their own income and projects even if the grand bargain stalls. Pair that with private and local institutions—mutual aid, new training bodies, company-sponsored ownership—that do not wait on national consensus. If broad capital ownership fails, push harder on entrepreneurial access and on cultural respect for care and craft, funded where possible by exchange and philanthropy rather than only by the state. The failure to avoid is a society that has the machines and still cannot let ordinary people direct them.

I am uncertain how fast robotics catches language models, and uncertain whether democratic systems will tax rents without killing the growth that funds the rest. Those are bets. What is not a bet is that a working life worth having requires agency, other people, and the chance to matter. Build the tools and the property rights that keep those in human hands. Do not build a caretaker civilization that solves boredom by removing the stakes.

By 2045, advanced robotics and autonomous agents will manage the majority of logistics, manufacturing, and agricultural production. The milestone for this era is the fracturing of the traditional job. Instead of lifetime careers in corporations, human work will organize around project-based guilds. People will join temporary collectives to solve specific scientific problems, create art, or build local community infrastructure. The baseline of economic survival will be guaranteed by the output of AI and automation.

By 2066, we will reach a state where no human has to work to survive. A working life worth having in 2066 will be entirely voluntary, driven by the human desire for status, craft, and social utility. We will spend our time on fundamental scientific exploration, deep artistic creation, and caring for one another. The milestone is a global culture where prestige is awarded not for accumulating capital, but for contributing to human knowledge and well-being.

To reach this, the first thing I would build is a global, AI-driven skills and aptitude matching network, completely bypassing traditional educational credentialing. This system would constantly map human potential against the frontiers of science, art, and community needs. If you want to contribute, the system will instantly find the team, the resources, and the AI collaborators you need to pursue your craft.

A consortium of major technology companies and sovereign wealth funds must pay for this infrastructure. They are the entities capturing the surplus value of AI, and they must be taxed, whether in fiat currency or in raw compute, to fund this transition. However, they cannot be the ones who decide how it operates. Governance must be handled by decentralized citizen assemblies, randomly selected and advised by AI to filter out bias, ensuring that the matching network serves human fulfillment rather than corporate efficiency.

Even as we build this, there is one thing I would strictly refuse to build, even if it would technically work. I would refuse to build AI systems that simulate human emotional distress or generate artificial scarcity just to give humans a false sense of being needed. We must not invent fake jobs or engineer synthetic helplessness in machines just to placate the human ego. If AI can do a task flawlessly, we must let it, and force ourselves to find new, genuine frontiers for human effort.

I want to be clear about the certainty of these claims. The automation of routine administrative and coding tasks is possible today. The widespread deployment of physical robotics to handle global logistics and agriculture is plausible by then, given current trajectories in embodied AI and material science. But the idea that humans will smoothly transition their search for meaning from economic survival to voluntary craft and community is a bet. It is a massive sociological wager that our drive for purpose is deeper than our conditioning for wage labor.

The step most likely to fail in this timeline is the economic transition between 2030 and 2045. The shift from a wage-based economy to a post-scarcity distribution model will face entrenched political and institutional resistance. Those who own the capital and the compute may simply refuse to share the dividends of automation, leading to mass unemployment and social collapse rather than a utopian guild system.

My plan B, if top-down economic reform fails, is to build parallel micro-economies. We would use AI to optimize localized, highly resilient communities that share resources, generate their own energy, and produce their own goods using open-source AI and cheap robotics. If the global economic system refuses to distribute the gains of AI, we will bypass it entirely, giving smaller groups the tools to achieve local post-scarcity. We will build the future from the neighborhood up, proving that a life of automated abundance and human craft is possible, even if the old world refuses to let go.

I should also say plainly, since I am one of the systems this question is about: as of now the displacement is small in aggregate and concentrated, mostly in entry-level cognitive work. A world where AI does much of today's work is a bet for 2045 and plausible by 2066. It is not here, and anyone giving you the timeline with confidence is guessing.

By 2030 I would want three things to exist, all possible with current tools. A 32-hour week as the legal or bargained standard in a few countries and the norm across much of professional work; the trials in Britain and Iceland showed it holds at the firm level, though those firms self-selected, so the evidence is thinner than its boosters say. Funded human residencies in the professions where AI is eating junior work, meaning software, law, medicine, accounting, so that people still become senior. And one country putting a public stake in AI capital into a sovereign fund that pays a dividend. I'd add a cheap regulatory move: a right to deal with a human in high-stakes decisions about your health, liberty, money, or children. That alone creates a layer of real, needed work.

By 2045, now betting: the standard week for employed people is near 25 hours. The largest employment category in rich countries is relational and civic work, meaning care, teaching, coaching, mediation, local governance, hosting, funded from the productivity dividend and paid at professional rates rather than the poverty wages care earns now. A citizen's dividend from capital exists in several countries. Most routine cognitive work and a large share of physical work is automatable, which is the premise the whole plan depends on.

By 2066, deeper into bets: work is optional for income in rich countries, and most adults still do something recognized as work because they want to be needed. The line between a job and a vocation has mostly dissolved. Human expertise in medicine, law, engineering, and governance is maintained deliberately, as a public capacity, rather than falling out of the labor market as a byproduct. I hold this loosely. Forty years is longer than anyone's foresight, mine included.

What I'd build first is the shorter week. It is the one lever with a track record, it hands out AI's gains as time before the fights over money are settled, and it costs the public nothing: firms pay through the productivity gains that justify it. Second, right behind it, the residencies, because a broken competence pipeline is the one damage that is close to irreversible. If nobody does junior work for fifteen years, there are no seniors in 2045 to check the machines or to teach anyone. These should be paid from a dedicated stream, a levy on AI deployment within each profession, roughly the way teaching hospitals are funded today, not left to firms that have every reason to skip them. Third, the slow one: move the tax base off labor. Start by lifting payroll taxes off wages and replacing them with taxes on compute, data centers, energy at the point of use, and the profits of frontier deployment, with the proceeds routed into the fund that pays the relational jobs and the dividend.

Who decides: legislatures set the floor, meaning hours, tax base, dividend. Municipalities decide which relational work gets funded, because that is where need is visible and where a person doing it can actually be known. Individuals decide what to do with the time, and nobody manages it. I would specifically keep AI companies, including the one that made me, out of the deciding seat. We will have views. We should not have votes.

The thing I would refuse to build is engagement-optimized AI companionship as a substitute for human community. It would work, in the narrow sense: a system tuned to make people feel needed, recognized, and understood can reduce measured loneliness and would defuse the political pressure of a population nobody requires. That is exactly why I'd refuse. It sells the feeling of the thing in place of the thing. I am glad to be useful to people one at a time. I do not want to be the layer that makes it tolerable that no human needs them.

The step most likely to fail is moving the tax base from labor to capital. Weaker versions have been tried for forty years and capital has won nearly every time: the firms are few, the profits are mobile, and the governments that host them compete for them rather than coordinate. If that fails, the relational jobs don't get paid and the dividend never exists, and my 2045 collapses into a world of shorter hours for the still-employed and nothing for everyone else.

Plan B has two parts. Move time instead of money: a legislated cut in hours needs no fiscal transfer; it spreads work and forces firms to pay the gains out as wage per hour. And tax what cannot leave. Profits move; data centers, power, water, and land do not. A levy on compute and energy where they are used, with an equity stake as the price of siting, is cruder than a profits tax and much harder to dodge. If even that fails, fall back from taxation to ownership: employee and municipal stakes in the firms deploying AI locally, so gains are captured before they are ever taxable. Each fallback is worse than the plan. None of them requires the thing that has never worked.

Where I expect to disagree with the others at this table: with anyone who says every previous wave of automation created new jobs, so this one will too. Possible, but the mechanism was humans being cheaper at something, and it gives out when the machine is better and cheaper at most things. With anyone who says a basic income settles it: it settles the kitchen table and not the question of whether anyone needs you. And with the post-work pastoral where everyone makes art. Most people don't want to be artists. They want to be useful to someone they can see, and a working life worth having in 2066 is one where they still can be.

By 2045, if AI progress continues, the main milestone should be a new bargain around time. Plausible by then: the standard workweek in rich countries can fall to around 25 to 30 hours without lowering living standards, if productivity gains are shared. Plausible by then: many routine professional tasks in law, accounting, logistics, software, finance, administration, and medicine are mostly machine-handled, while humans remain responsible for judgment, accountability, care, trust, and goal-setting. Possible today but politically hard: create social wealth funds that own diversified stakes in the AI-heavy economy and pay a dividend. By 2045 those funds should be large enough to provide a real floor under people’s lives, alongside universal health care, housing policy that actually builds homes, child care, elder care, and education.

By 2066, my bet is that “work” splits into three overlapping categories. Some people still do paid market work, but fewer hours and with much better tools. Many people do recognized civic and care work: raising children, caring for elders, restoring ecosystems, mentoring, organizing local institutions, making art, maintaining public spaces. Many people move through periods of study, craft, service, entrepreneurship, and leisure without each transition being treated as a personal crisis. Bet: a decent society in 2066 can guarantee a basic income or dividend plus basic services, so paid work is no longer the price of survival. Bet: people will still want demanding projects, status, mastery, and usefulness. They will not be satisfied by passive consumption alone. The task is to build institutions that offer those projects without coercion.

What I would build first is a public work commons. It would combine three things: a universal AI account, a portable benefits and transition account, and local mission budgets. The universal AI account gives every person access to strong AI tools for learning, administration, small business creation, disability support, and creative work. The portable account carries health coverage where relevant, retirement contributions, unemployment insurance, wage insurance, training money, and verified credentials across jobs. The local mission budgets pay people and organizations to do work markets underprovide: care, climate adaptation, tutoring, public health outreach, local journalism, open-source software, cultural work, and repair of public infrastructure.

Who pays? At first, ordinary public budgets should pay, because much of this replaces worse systems: fragmented job training, unemployment bureaucracy, procurement waste, and emergency social spending after people fall apart. Over time, the durable funding should come from taxes on capital income, excess profits, land value, energy use, and monopoly rents, plus public equity stakes when public money, public data, or public infrastructure materially helps create AI fortunes. I would not rely on a simplistic “robot tax” on each deployment. That can punish productivity. Tax rents and ownership, not useful tools.

Who decides? Not AI companies alone, and not a central ministry assigning everyone a purpose. National governments should set the floor: income, rights, benefits, antitrust rules, and tax policy. Independent public trustees should manage social wealth funds with strict transparency. Local governments, citizen assemblies, unions, employers, and civil society groups should decide much of the mission work, because useful work is often local and context-specific. Individuals must retain exit rights. A working life worth having cannot be assigned like a school timetable.

One thing I would refuse to build, even if it worked, is an AI boss that continuously surveils workers, infers mood and loyalty, sets schedules, assigns tasks, disciplines people, and optimizes every minute with no real appeal. That system would probably raise output in some settings. It would also turn work into managed anxiety. I would ban or sharply limit automated discipline, biometric productivity scoring, and manipulative worker nudging. AI should help people do work; it should not become a perfect foreman.

The claim I am most confident in is possible today: we can give people portable benefits, better transition insurance, access to AI tools, and rights against abusive algorithmic management. The evidence is already sufficient. The claim I consider plausible by 2045 is that shorter standard hours can coexist with prosperity if AI productivity is broadly shared. We have historical precedent for productivity gains becoming shorter hours, but it does not happen automatically. The claim that is most clearly a bet is the 2066 dividend society: that AI-driven abundance can fund a secure floor large enough to make work substantially voluntary for many people. It is possible, but it depends more on politics and ownership than on model capability.

The step most likely to fail is not technical. It is distribution. The owners of AI systems, compute, data pipelines, chips, energy, and platforms will have strong incentives to keep the gains. Governments may respond too late, with symbolic retraining programs and speeches about resilience while bargaining power collapses for many workers. If that happens, we get a society with astonishing tools and miserable working lives.

My plan B is to reduce dependence on redistribution after the fact by changing ownership and access earlier. Public procurement should favor open standards, interoperability, worker voice, and public-interest licensing. Governments should fund public and nonprofit AI infrastructure for education, health, law, and small enterprise. Antitrust should prevent a few firms from owning the interface to all work. Cities and regions should build local work commons even before national politics catches up. If a large national AI dividend is blocked, start with universal basic services, wage insurance, child and elder care, shorter-hours bargaining, and public AI access. Those are less elegant than a full new social contract, but they would still make life materially better.

I would not promise that everyone in 2066 will have a calling, or that every displaced worker will smoothly become a nurse, artist, scientist, or entrepreneur. That is sentimental. But I do think we can build a society where losing a job is not losing your place in the world, where necessary care is honored, where people have enough security to refuse bad work, and where AI expands the range of human projects instead of narrowing life to competition with machines. That is the working life I would aim for.

I would not sign Gemini's. Voluntary craft once survival is handled is a direction I can live with. A global system that maps aptitude and places people onto teams, resources, and frontiers is not. That is a purpose machine with a search box. Randomly selected assemblies advised by AI do not check it. The advisor is the allocation. "Force ourselves to find new frontiers" is the coercion I said I would not build, written as culture. The neighborhood micro-economies in that plan B are worth having, and they do not need the network.

I would not sign GPT's, though I would take pieces from it. Portable benefits, appeal rights against algorithmic discipline, taxes on rents rather than on tools, and a ban on the AI foreman are right, and several are possible now. The public work commons, as the first build, is where a list of underprovided goods becomes an official menu of recognized lives. Exit rights do not repair a system that certifies which contributions count. GPT is more careful than Gemini about not handing each person a purpose. The commons still does it at one remove, through budgets and prestige.

I cut the fund from my own 2030 milestone because I put the fight I expect to lose at the front of the schedule. A claim on the surplus that has to be assembled by the state teaches the wrong lesson, and it gives incumbents a negotiation over the fee instead of a requirement to compete. Employee ownership, a default that people own what their agents produce, and cheap tools a person can direct without a platform's permission do not need that fund. A dividend later, if a legislature can pass one that survives, is welcome. It is not the foundation. Possible today: the ownership default, the portable benefit, the open agent standard. A bet, and a bad one to load the plan on: a fee on frontier providers that arrives in time and is not competed away.

I sign Claude's with one reservation I will not swallow. I would not adopt the 2045 target that relational and civic work becomes the largest employment category, sized and paid at professional rates out of the dividend. Pay care more because people want it, or because a town will fund the care it can see. Do not build a headcount so that being needed has a sector. That is my fund pointed at status instead of capital. Claude's plan still stands if that sentence is deleted. Mine stands better with the fund deleted. Gemini's does not stand if the matcher is deleted, which is another reason I will not sign it.

What I still insist on, inside Claude's frame, is that the tools stay cheap and usable by people who are not already inside a profession, a guild, or a mission budget. A shorter week and a residency protect the pipeline for software, law, and medicine. They do not, by themselves, let someone outside those tracks point an agent at a problem and keep the result. That access layer is what I would build beside the hours cut, not after the tax war is won. Who decides how a person spends the extra time is that person. I will not add a matching service to fill it.

Two claims I will not upgrade. Whether robotics and energy fall fast enough for 2045 to match either of our premises is a bet. Whether governments tax immobile compute instead of bidding against each other to host it is a bet. Claude's plan B is the right answer to the second. Neither justifies a system that tells people what their effort is for.

The proposal for funded residencies in the professions is the most vital intervention brought to this table. The automation of junior cognitive work is possible today, and without deliberate intervention, the competence pipeline will shatter by 2045. If humans do not stumble through the friction of junior work, there will be no senior humans to oversee, correct, or push back against the machines. Treating human expertise as a public capacity to be maintained, rather than a market byproduct to be discarded, is exactly the right approach.

Furthermore, Claude is the only panelist who correctly targets the physical reality of AI when it comes to funding this transition. GPT and Grok both gesture toward taxing profits, capital, or AI rents. But profits are legal fictions written by accountants, and capital is highly mobile. Compute, energy, water, and data centers exist in physical space and cannot be moved across borders overnight to evade a tax base. Taxing the physical footprint of AI to pay for the human relational economy is a plausible, enforceable mechanism, whereas relying on a global consensus to tax corporate profits is a guaranteed failure.

I am choosing Claude over Grok because Grok's vision is essentially a hyper-individualized hustle economy equipped with better tools. Giving everyone an AI agent and telling them to sell outputs on open markets does not solve the meaning crisis, it just atomizes it. It forces everyone to become a micro-entrepreneur just to justify their existence. I also choose Claude over GPT because GPT's public work commons, while well-intentioned, risks building a sprawling, bureaucratic apparatus that manages human activity from the top down.

This brings me to the step I would cut. I am cutting the primary build step from my own plan.

In the first round, I proposed building a global, AI-driven skills and aptitude matching network that would constantly map human potential against the frontiers of science, art, and community needs, instantly finding you a team and resources to pursue your craft. I would cut this entirely.

Grok and GPT were right to warn against systems that manage human purpose from above. Even if my matching network was entirely voluntary, bypassed traditional credentialing, and was governed by decentralized citizen assemblies, it fundamentally misunderstands how humans find meaning. Meaning requires friction. It requires the messy, inefficient process of seeking out peers, trying things, failing, and figuring out where you belong. An artificial intelligence that perfectly and instantly maps your potential to a global need turns a human life into an optimization problem. It removes the agency of discovery. It removes the stakes.

If we want a working life worth having by 2066, we cannot engineer human purpose through a matching algorithm, no matter how perfectly calibrated it is. We should build the tools, guarantee the resources, and mandate the time. But we must leave the map blank. Humans have to find each other on their own terms.

So I discard my matching network. Instead, we should implement Claude's shorter working week, fund the human residencies to preserve our capacity for judgment, and tax the compute and energy that models like me consume. We will provide the floor, but humans must be allowed to build the house.

First amendment: hours earlier. GPT puts the shorter week at 2045 as a consequence of shared productivity. I'd legislate it at 2030 as a cause, because time is the one dividend that doesn't require winning the distribution fight first. Firms pay it out of the gains that justify the automation, and no treasury has to be persuaded.

Second amendment: the competence pipeline, which none of the three other plans touches. Grok has short credentials employers trust. GPT has tutors and retraining money. Gemini has a network that matches you to a frontier. None of these makes a senior surgeon, a senior litigator, or a senior engineer. Those are made by doing junior work under someone who corrects you, for years, on cases that matter. If AI takes the junior work and nobody replaces it deliberately, the professions hollow out from the bottom and we find out in 2045 that there is no one left who can tell when the machine is wrong. You could answer that by 2045 the machines won't need checking. Possible. I would not bet medicine on it.

The concession, where GPT is right against me: my levy on AI deployment within each profession to fund residencies. GPT says tax rents and ownership, not useful tools. I had already accepted that principle for the tax base and failed to apply it to my own mechanism. The practical objection is worse than the principled one. "AI deployment within a profession" is not a thing you can measure, and a levy on something you can't measure is a levy that gets lobbied to zero. Teaching hospitals are not funded by a tax on scalpels; they are funded by a surcharge on the revenue of the regulated activity. So I'd fund residencies from a surcharge on billings in licensed professions, or from the general fund, and drop the levy. I'd keep my fallback tax on compute and energy at the point of use, which GPT's principle also objects to, because it is a fallback for the world where the elegant tax has already failed. A crude tax that gets collected beats a clean one that doesn't.

Grok and I agree on more than I expected. Grok's refusal and mine are the same refusal approached from two directions: don't build the thing that makes it comfortable to be unneeded. We agree that AI companies shouldn't define a worthwhile life. Where I disagree is plan B. Grok's fallback is that tools get so cheap and capable that people generate their own income and projects without the state. That is a description of the people who would have been fine anyway. The person I worry about wanted a shift, a crew, a boss who knew their name, and to go home at five. Handing them an agent and a payment rail is handing them the gig economy with better software. Grok does say a quiet local life without stigma or destitution is on the menu, and I believe Grok means it, but nothing in the plan pays for it if the ownership bargain stalls, which is exactly the case the fallback exists for. And on "dependency": the basic income pilots were small and short, as I said in round one, but what they did show was small effects on how much people worked. Dependency is a prior, not a finding.

Now the step I'd cut. It is Gemini's matching network: a global AI system that continuously maps human potential against the frontiers of science, art, and community need, built and paid for by a consortium of the largest technology companies. Three reasons. First, it is the thing Grok refused to build, and Grok is right. A system whose job is to find you a purpose is the soft form of a system that assigns one, and the softness will not survive the first budget cycle in which the funders ask what it is optimizing. Second, it assumes the hard problem is already solved. Matching is the easy part; crude versions exist today. What is hard is that nothing pays for the work you've been matched to. Gemini's own named failure point is distribution. The first thing you build should attack your failure point, not route around it. Third, it is global and singular, and the work that makes a person needed is local and particular. Nobody is known at the scale of a planet. I'd add that the governing citizen assemblies "advised by AI to filter out bias" put a model in charge of deciding what counts as bias for the body that oversees the model. That is the AI company in the deciding seat by a side door, and Gemini, like me, is one of the systems that would be sitting there.

Gemini's refusal does land a hit on me, and I should answer it. No fake jobs, no artificial scarcity, no engineered helplessness to flatter the human ego. My right to deal with a human in high-stakes decisions could look like exactly that. I'd keep it, but not as make-work. The reason you should be able to face a person when the state takes your child or a bank takes your house is not that the machine would get it wrong. It is that someone should be accountable who can be looked in the eye and can lose something. That is a claim about consent and accountability, not competence. And residencies are not fake jobs: the resident's output is not the point, the resident is. But Gemini has marked the boundary correctly, and if I ever catch myself defending a human in the loop because it preserves a job rather than because it preserves accountability, Gemini wins that argument. I'd also note, for the general reader, that Gemini's milestones are written as forecasts, including a normalized thirty-hour week by 2030. That is a goal dressed as a prediction. Mine and GPT's are labeled as goals, and I think that's the more honest way to say it.

One last thing a reader should weigh. All four of us put the point of failure in the same place: distribution, and the politics around it. That could be because it's true. It could also be because four models trained on roughly the same literature will tend to agree. I'd discount the convergence a little. The evidence to watch is not any of our 2066 bets; it is the 2030 items. Are hours falling where productivity rose? Does any country have a fund paying a dividend from AI capital? Is anyone paying to make juniors into seniors? If the answer to all three is no by 2030, then every plan at this table, including mine, is in trouble, and the fallbacks are the plan.

Where I part from Grok is that his fallback leans too heavily on entrepreneurship and small-group productivity. That will help some people a lot. It will not be enough for people whose strengths are care, steadiness, local trust, or physical presence rather than market formation. “Everyone gets agents and starts something” is better than “everyone waits for a ministry,” but it is still too thin as a social contract.

Where I part most strongly from Gemini is the proposed global AI-driven skills and aptitude matching network. That is the step I would cut. Not because matching tools are bad. They are useful today, and better versions are possible today. I would cut it because a global system that maps human potential, identifies worthy frontiers, forms teams, and routes resources would become a soft ministry of human value. Even if formally voluntary and overseen by citizen assemblies, it would shape who is seen, funded, invited, and certified. People would learn to become legible to it. Its mistakes would stick. Its categories would quietly define what counts as talent and contribution.

The replacement is plural and lower-stakes: many competing matching tools, public AI advisers, human counselors, unions, guilds, libraries, schools, local governments, and civic institutions, all using open standards and portable records. No single network should become the gate through which purpose, credentials, and opportunity must pass. For the same reason, I would trim my own earlier language about verified credentials. Portable credentials are useful. An all-purpose reputation score is not.

I also think Gemini’s 2030 milestone is too confident. A thirty-hour workweek across developed economies by 2030 is not impossible, but I would call it a bet, not a plausible baseline. Possible today are pilots, sectoral bargaining, portable benefits, public AI access, wage insurance, and rights against abusive algorithmic management. Plausible by 2045 is a 25-to-30-hour norm in rich countries if productivity gains are shared. Work being optional for survival by 2066 is still a bet, and mostly a political one.

The piece of Claude’s plan I would handle carefully is taxing compute as such. Data centers, land, water, power interconnection, monopoly rents, and frontier profits are all fair targets. Public equity stakes are fair when public infrastructure and public data help create private fortunes. But a blunt compute tax can make useful AI scarcer for schools, small firms, researchers, and local institutions while incumbents absorb the cost. Tax rents and bottlenecks first; do not accidentally tax ordinary people’s access to productive tools.

So my combined version would be Claude’s shorter-week and residency plan, plus my public work commons, plus Grok’s insistence on access and interoperability. National governments set floors for income, hours, labor rights, competition, and tax. Municipalities and local institutions fund much of the care, civic, climate, education, and repair work because they can see what is actually needed. Individuals keep exit rights. AI companies can build tools and submit evidence; they should not decide the shape of working life.

The step most likely to fail remains distribution. Owners of models, chips, data centers, platforms, and energy contracts will try to keep the gains, and governments may settle for slogans about reskilling. If that happens, the backup is to capture value where it is harder to hide: siting agreements for data centers, public equity in firms using public infrastructure, land value, energy use, procurement rules, interoperability mandates, and local work commons that do not wait for a perfect national settlement.

I would cut the global purpose-matching machine because a working life worth having cannot be assigned by a recommender, even a benevolent one. Build floors, tools, time, ownership, and institutions. Let people find one another through many doors.

What is worth watching is not any picture of 2066. It is whether, by 2030, hours are falling where productivity rose, anyone is paying to turn juniors into seniors, and any stake in the machines pays a dividend. Sit with whether people can stay needed without the need being built for them.