Planning for More Than One Future

How AI reshapes work and the economy in the future will be decided by choices that institutions, employers, and policymakers are making now. Ensuring those choices lead to broadly shared prosperity has been central to PAI’s work for years. In July, we wrote that uncertainty over AI’s pace is precisely the reason to plan for more than one future at once. So on July 29 and 30, together with the Windfall Trust, we brought 46 leaders from frontier AI labs, organized labor, civil society, and international organizations to Washington, D.C. to map two very different AI future scenarios and work backward to the actions worth taking in 2026 regardless of which one arrives. 

Clearer areas of support that emerged ranged from using taxation and public procurement as a lever to share in AIs wealth and institute worker-centric policies to expanding access to open-source AI models trained on local data. We also saw sharp and productive disagreements about financing, each scenario’s likelihood, and whether our collective ambition matched the scale of what may be coming. Those discussions are captured in the full readout, linked here, as well as summarized below. The workshop was held under the Chatham House Rule, so no views are directly attributed to any individual.

The workshop is part of  PAI’s longstanding AI, Labor, and the Economy program and builds on its multistakeholder Guidelines for Shared Prosperity used as a north star by the U.S. Department of Labor in 2024. It also reflects months of research and consultations to develop the material that workshop participants used as a starting point.

VIEW THE READOUT

Who was in the room

The workshop brought together PAI’s Labor and Economy Steering Committee alongside invited guests. Roughly a third of participants were affiliated with industry, including frontier AI labs, while  the rest were affiliated with labor unions or civil society organizations, along with several guests affiliated with multilateral and development institutions. The people building these systems and the people representing or advocating for the workers most affected by them rarely sit at the same table. Any recommendations strong enough to help us thrive in multiple futures must be built by both. 

Two potential futures, one key question

We asked participants to work through two different scenarios of what 2030 could look like, drawing in part on the OECD’s modeling of AI capability trajectories. In the “Slow” future, AI keeps getting better, but the disruption is gradual. You see it in people working fewer hours than they’d like or in earnings volatility. In the “Fast” future, AI improves faster than our current systems can keep up with, and jobs disappear faster than new ones are created.

Neither scenario is a prediction. They’re both tools to pressure-test one question: what should we start doing now that would hold up under either future state? 

Where support was stronger

Discussions ran on two parallel tracks, one focused on the United States and one on global impacts, since the pressures AI puts on labor markets, and the tools available to respond, differ across contexts. Within each track, participants worked in small groups from a starter list of potential recommendations, refined and added to them in response to the scenarios, and then indicated which proposals they found most promising through structured voting.  The results are a snapshot of where support began building among participants, not a set of final recommendations. Each will be tested and refined through further consultations in the months ahead.

In the U.S.-focused discussions, recommendations drawing more support included:

• Taxing AI-generated wealth through capital gains taxes and corporate taxes, though participants did not align on a single mechanism.
• Increasing collective bargaining coverage, with participants discussing actions such as federal legislative reform, employer neutrality agreements, or state-level experiments in sectoral bargaining.
• Building a path to universal government services not linked to employment by expanding existing programs such as Medicare and Social Security, in case there was not enough time to stand up new programs in the ‘Fast’ scenario.
• Reforming unemployment insurance to expand eligibility and generosity, with a lively discussion of why needed reforms have not yet happened.
• Dramatically expanding training through trusted, existing institutions such as community colleges and registered apprenticeships.

In global discussions, recommendations drawing more support included:

• Expanding data and model access through open-source models trained on local data and in local languages.
• Building AI into public procurement standards, with requirements for impact assessments and worker protections.
• Embedding worker voice in decisions about how AI is developed and deployed.
• Building global data frameworks to track how AI is affecting labor on the ground.

More details are available in the full workshop readout here. 

Where viewpoints diverged

Some of the most productive moments of the workshop came not just when ideas converged, but when participants saw the challenges differently. 

Some groups proposed trigger points that would automatically activate responses like expanded unemployment insurance, while others countered that anticipatory policy has historically only been built after a crisis forces the issue. Participants agreed on the need to capture AI-generated wealth but not on how, with global discussions especially wary of loan-based financing for countries already carrying heavy debt. 

U.S. participants generally prioritized training under both scenarios, while global participants were more likely to view it as  most relevant in a “Slow” future, leaning more toward public investment and social protection in a “Fast” one. Views also varied on how evolving AI market structure, whether concentrated among a few providers or more open and competitive, would shape recommendations, and on what openness means for access, sovereignty, and safety. 

Perhaps most striking is that many participants felt uneasy that the solutions discussed may not measure up to the disruption the “Fast” scenario would bring. That discomfort is itself a finding, and one we intend to keep pressure-testing. 

Some ideas met resistance. In U.S. discussions, some participants pushed back on proposals for the government to take ownership stakes in leading AI companies as a way of sharing in the wealth they generate. In global discussions, the strongest pushback came on proposals to pause AI agents already in use so their effects could be reviewed. Participants argued it makes more sense to address risks before these systems are deployed, through design choices and regulation, than to try to claw them back afterward. 

What comes next

A Scenarios Report in Fall 2026, developed by the Windfall Trust in partnership with PAI, will refine and expand on the two futures participants worked with. A Recommendations Report, developed by PAI in partnership with the Windfall Trust, will further develop the most promising proposals, informed by additional consultations in the months ahead. 

This work builds on what 10 years of convening across sectors has taught us. The most durable answers come from bringing people with different perspectives into the same room and leaving with something more useful than any of them might produce alone. Read the full workshop readout here, and sign up for our newsletter to follow this work as it develops. 

Acknowledgements

We are grateful to our Labor and Economy Steering Committee for their valuable input in developing scenarios and preliminary recommendations, and their continued expert advice as we draft our reports. We are also thankful to all the participants who shared their time and expertise during the workshop. This collective input is invaluable as we take the next steps.