InnovateUS 
What we learned from 250,000 public workers about AI adoption
reprinted from an article in Fast Company, October 26, 2026
Our research reveals what workers need to use AI well. The next challenge is understanding whether that learning translates into better government services.
As generative AI sweeps across industries, some workers using it are reporting increases in productivity. But inside government, where those benefits could translate into better public services, many employees are still unsure how the technology applies to their jobs, how to use it, or whether they are allowed to.
Learning together online also gives us an opportunity to listen systematically to people’s experiences with AI.
That’s one takeaway from our research at InnovateUS, a nonprofit learning initiative that has trained more than 250,000 government workers in AI, digital technology, and innovative skills. They work in all 50 states and the District of Columbia, in jobs ranging from public safety and benefits administration to legal services and executive leadership. Our aim is to give people working in state and local government accessible opportunities to develop skills that help them serve the public, and ultimately deliver services worthy of residents’ trust.
After training, highly confident respondents planned to apply AI to an average of 3.8 tasks, compared with 1.5 among those with low confidence.
During no-cost skill-building workshops across the country and on the web, public servants trade experiences about what works and what doesn’t. Participants have explored using artificial intelligence to improve benefits delivery with New York City’s health commissioner and learned from current and former state-government leaders how to introduce workers to using and even building AI. Recently, one state leader in Colorado described how he is using generative AI to improve performance reviews; he reported better feedback, greater worker satisfaction, reduced turnover, and time savings. These exchanges help public professionals learn from one another’s experiments and consider what they could adapt to their own agencies.
Learning together online also gives us an opportunity to listen systematically to people’s experiences with AI. That listening underpins the first report from our Observatory of Public Sector AI, drawing on 106,300 precourse survey responses and 67,002 postcourse responses from participants in our introductory generative AI course, collected between June 2024 and May 2026. The findings reveal a substantial gap between the headlines around generative AI and everyday working life in government.
What keeps public workers from using AI
Among public-sector respondents entering our foundational generative AI course, 51% had never tried the technology. Another 14% had experimented with it personally but weren’t yet using it at work. While technology companies race to build autonomous agents, many workers are still trying to understand what AI has to do with their jobs, why they are being asked to use it, and how to use it well. They’re also not sure if they’re allowed to use it at all.
The findings complicate the stereotype of government employees as resistant to change. Among respondents who weren’t using generative AI, 28% said they didn’t think it applied to their work, 23% didn’t know how to use it, and 19% didn’t know whether they were allowed to use it at work.
Managers face similar uncertainty. Among managers who weren’t using AI, 24% didn’t know how, and 22% didn’t know whether it was permitted. Employees look to supervisors for guidance, but those supervisors may need help themselves.
Listening to the workforce gives leaders something concrete to address. Providing access to a tool or publishing a policy leaves substantial work undone. Employees need relevant examples, opportunities to practice, and managers prepared to support responsible use.
At a time when many people are skeptical of AI and uncertain about how it will benefit them, government has an opportunity to demonstrate what the technology can accomplish in the public interest. AI should not be a technology whose benefits accrue primarily to the companies building it or the people who can afford it.
To turn AI use into public value, we need to help employees connect new skills to specific service problems. This might look like reducing wait times when people call about health or unemployment insurance, using AI to make it cheaper and easier to engage the public in policymaking, or even knowing how to regulate and buy AI to reduce dependence on any one vendor. And we need to give them the support to change how work gets done. Systematic workforce training creates an opportunity to build that capability to listen, learn, and deliver, and to understand whether it makes a difference.
Beginners report the largest gains
One of our most encouraging findings concerns people starting with the least familiarity. In the matched responses, learners who entered with no reported AI knowledge showed the largest gains in reported knowledge. Those who began with no confidence also experienced particularly substantial improvements in confidence.
This parallels research in the private sector. A study of private sector customer-support workers by academics at Stanford and MIT found that less experienced and lower-skilled workers gained the most from an AI assistant. That study measured workplace performance; ours measures reported learning. Together, these studies suggest that organizations should pay attention to people who are just getting started. Concentrating support on enthusiastic early adopters risks overlooking workers who could benefit considerably from an introduction.
We also found a striking relationship between confidence and the possibilities people could envision for generative AI tools. After training, highly confident respondents planned to apply AI to an average of 3.8 tasks, compared with 1.5 among those with low confidence. That association does not establish that confidence causes broader use. It suggests that learning involves more than acquiring facts about a technology. People also need to feel capable of identifying and trying appropriate applications.
Confidence alone, however, is an incomplete measure of success. Someone can feel more capable without becoming more accurate. Training must develop the judgment to question an answer, verify a source, protect sensitive information, and recognize when AI is inappropriate. We need to assess those capabilities alongside confidence.
From better tasks to better services
When asked how they hoped to use AI, participants readily described activities such as drafting reports, creating presentations, and summarizing documents. Fewer responses connected those activities to outcomes such as making services more accessible or improving public participation.
Our questions partly explain that pattern. We asked about specific tasks, and brief open-ended responses offered limited insight into what people hoped to accomplish through them. We have begun collecting voice-based feedback and are adding questions that probe whether learning is helping employees improve services.
Private-sector research underscores why the distinction between individual learning and organizational outcomes matters. In McKinsey’s 2026 global survey, 80% of respondents said AI improved their own productivity, yet only 37% reported a positive impact on their organization’s earnings. Helping individuals work faster does not automatically improve an organization’s performance.
For democratic institutions, the corresponding test is whether time saved translates into better public services, more engagement with residents, and more effective problem solving. Does faster document processing reduce the wait for benefits? Does clearer communication help more people complete an application? Does using AI to analyze public feedback lead to a better decision? Ultimately, do people trust government more? Training needs to help employees make those connections, and evaluation needs to tell us whether they hold up in practice.
What we still need to learn
Our first report captures an early stage in a much longer process. Its surveys are voluntary, and participants are a large but not representative sample of the government workforce. The results measure reported knowledge, confidence, and intended use immediately before and after a course. They do not establish whether people retain what they learn, apply it consistently, make fewer mistakes, or deliver better services. Without a comparison group, we also cannot isolate the effect of training from other influences. The scale of the dataset helps us examine patterns across professions and organizations, though it does not remove those limitations.
Our opportunities to learn more are accelerating. As our learner population grows and we expand the range of people teaching and learning from one another, we can investigate which approaches help different workers and what support they need after an introductory course.
Places such as Washington, D.C., which has adopted a whole-of-workforce training approach, offer an especially valuable opportunity. Studying learning across an organization can help us understand how individual skills interact with managerial support, access to tools, and agency rules.
The most exciting next step is connecting evidence about learning with administrative measures of service performance and other data on trust in government. With appropriate privacy protections, we could examine whether training is associated with shorter wait times, more accurate case resolution, fewer application errors, easier access to services, and more outreach to residents.
Those studies will require careful design. A shorter queue could reflect staffing changes, new software, or lower demand. Following teams over time and comparing their experiences could help us understand whether, how, and under what conditions learning contributes to more effective, efficient and engaged institutions.
We have encouraging evidence that public professionals can gain knowledge and confidence around AI, including those starting with little familiarity. Now we need to follow what happens next as we work with governments to roll out new training that teaches people how to use AI in their specific role and to work in new ways.
For government leaders, that means treating workforce learning as an investment whose results deserve sustained attention. This would include things like offering practical training, preparing managers to support it, giving employees opportunities to apply what they learn, and measuring what changes for the people they serve.
At a time when many people are skeptical of AI and uncertain about how it will benefit them, government has an opportunity to demonstrate what the technology can accomplish in the public interest. AI should not be a technology whose benefits accrue primarily to the companies building it or the people who can afford it.
By equipping public servants with the skills to use AI responsibly and creatively, we can put powerful new capabilities in the hands of people whose job is to serve others. The measure of success will not be how many government employees use AI or how much time they save, but whether people experience better services, have more opportunities to participate in decisions, and see government solving problems that matter to them. Making those kinds of differences in people’s lives could make AI worthy of their trust and, more importantly, help rebuild their trust in government.