The most useful question about AI in a workplace is not whether the technology can do a task. It is whether the people closest to the work have a hand in deciding how it gets used. That distinction shapes every pilot we run, and it shapes how we think about this technology more broadly.
Our position: The same AI tool can be deployed to hinder a worker or to elevate their work. Which one happens is not determined by the technology. It is determined by the people choosing how to introduce it, who they consult, what training they offer, and whether workers can push back when something is not working.
We build pilots for our clients the same way: the people doing the work pick the task, shape the tool, and own the result.
The approach matters more than the technology
AI replaces the worker
Treats the work as a cost to eliminate. Cuts staff first, asks questions later. Treats the loss of institutional knowledge as a rounding error. Frequently misses the parts of the job that were never written down.
AI supports the worker
Treats the worker as the expert. Uses AI to remove drudgery, free up judgment, and let people work at the top of their skills. Measures success by whether the worker can do more meaningful work, not by headcount reduction.
The evidence supports both possibilities. The same body of research that warns about displacement also documents wage gains, role enhancement, and new openings for workers in AI-exposed sectors. Whether a given workplace lands closer to one outcome or the other depends on choices leaders make, not on the technology itself.
What the data shows
Two truths exist in the same data set. The same period that produces wage growth for workers using AI well also produces planned reductions in roles where it is being used to replace them. The choice is which one a given organization moves toward.
Three commitments that change the outcome
Augmentation over replacement is a leadership choice, not a technology destiny
The same generative AI model that can produce a memo can be used by a policy analyst to draft three times faster, or it can be used to replace the policy analyst. Both are technically possible. Which one happens depends on what the executive team decides to measure: throughput per worker, or workers per unit of throughput. Organizations that lead with the first measurement consistently report higher retention, better adoption, and fewer of the institutional-knowledge gaps that show up six months after a layoff. Organizations that lead with the second measurement frequently find that the institutional knowledge they cut was holding together more of the work than the org chart suggested.
Worker voice before deployment, not after
The federal rule on high-impact AI systems already requires consultation with affected communities, but only after the system is built. Brookings researchers note that post-hoc consultation is not the same as participatory design. The cheapest moment to identify what AI will get wrong is before a single workflow is changed. The people who do the work daily know things the procurement team does not.
Worker voice shows up two ways. Where workers are organized, it shows up through collective bargaining. The Writers Guild and Screen Actors Guild contracts of 2023 set early precedents on consent, compensation, and credit. The AFL-CIO Technology Institute, launched in 2024, has built ongoing work with Unite Here on algorithmic management in hospitality, with the Communications Workers of America on bargaining principles, and with the broader labor movement on participatory technology design. The California Labor Federation has published its own principles document, which states a position we share: there is nothing inevitable about how this technology gets deployed.
Where workers are not organized, the same principle holds and the same outcomes follow. Aspen Digital's Frontline AI guide documents that workplaces minimizing worker autonomy in technology rollouts consistently see higher attrition and slower adoption. Participatory design with non-represented workers (employee advisory groups, before-and-after listening sessions, identified feedback channels) is not a substitute for bargaining rights, but it can be a meaningful instrument when bargaining is not on the table.
"Workers are the experts in how technology impacts their work."Liz Shuler, President, AFL-CIO
Access to real skill-building, not just access to a tool
Handing a worker a chatbot login is not training. Several public programs are now built on the recognition that AI capability comes from guided practice on real work, not from a one-time webinar.
The federal AI Upskill Accelerator at the Economic Development Administration committed $25 million to industry-led workforce partnerships. The Department of Labor's Make America AI-Ready initiative launched to deliver literacy training over text message, designed for workers without laptops or reliable broadband. New York State's InnovateUS partnership trained over 200,000 public sector learners and expanded AI training to the full state workforce after a successful pilot. On the private side, Anthropic Academy offers a free AI Fluency for Nonprofits course co-developed with GivingTuesday, alongside free AI Fluency courses for general professionals, educators, and students.
None of this is comprehensive yet. The point is that training infrastructure designed specifically to make AI accessible to workers, not just to executives, is real and growing. It is a small set of options today; it will not stay small.
A short timeline of worker-centered AI work
Writers Guild and Screen Actors Guild contracts set early precedents on consent, compensation, and credit for AI use. California Labor Federation publishes its principles on AI in the workplace.
AFL-CIO Technology Institute launches. Microsoft signs a neutrality and worker-input agreement with the AFL-CIO. IBEW publishes its AI Data Center Principles. The U.S. Department of Labor releases AI Best Practices framing worker empowerment as a federal priority.
AFL-CIO Tech Institute and Unite Here publish joint work on algorithmic management in hospitality. Communications Workers of America extends bargaining principles into game development through ZeniMax.
The trajectory is cautiously headed towards support. Each year since 2023 has added bargaining language, public training infrastructure, and research on what worker-centered AI deployment actually requires. What the next entry on this timeline says will be shaped by the work being built now.
Pilots that start with the person doing the work
Every Passons AI pilot is fixed fee, approximately two weeks, and structured so that a frontline worker, not an executive, picks the task we automate. We sit with the people doing the work, watch what they actually do, and build a workflow they own afterward. The deliverable belongs to them and the organization. The skill belongs to them.
We do not run pilots to justify staff reductions. If an agency is looking for that, we are not the right partner. We are the right partner for leaders who believe their workforce is the asset, and that AI is most valuable when it lets that workforce do more of the work that actually matters.