I was a government official in the 1990s and watched the economy get turned upside-down. It’s happening again

Today’s economy is being shaped by a range of competing forces: technological disruption, a rapidly globalizing marketplace, and a workforce caught in between — largely without a safety net designed for the speed of change. But this isn’t the first time that’s happened. As someone who worked for the U.S. government throughout the 1990s, I’m seeing evidence that the ‘90s are back in ways that extend beyond FX’s Love Story and the resurgence of grunge.

My work at the U.S. Department of Labor in the ‘90s coincided with the passage of the North American Free Trade Agreement and the acceleration of trade with China. I saw the ways those dynamics created winners and losers: some industries flourished while others collapsed, and six million manufacturing jobs disappeared. Today, AI may be creating a similarly defining moment of labor-market uncertainty, and at an even larger scale. Some estimates suggest that 6% of American jobs could disappear due to AI, affecting 11 million American workers.  

At a time of major labor market disruption, the ‘90s offer both a playbook and a warning. Some of the responses I helped the government build supported workers in a time of economic turmoil, but others missed the mark. As we try to navigate a world increasingly dominated by AI, what can we learn from what was attempted, what worked, and what fell short thirty years ago?

Ideas Whose Time Has Come 

“The rising costs of postsecondary education are putting higher education out of reach for an increasing number of citizens.” 

“The skills of United States workers have not kept pace with the complexity of current job requirements.”  

These may sound like today’s concerns. But they come from the introductions to the 1993 act that created AmeriCorps and the Youth Apprenticeship Act of 1990. Those bills reflected a growing recognition that many previous career paths for high school graduates who did not attend college—especially in manufacturing—were no longer viable. Along with the School-to-Work Opportunities Act of 1994, they aimed to help young Americans access education and training experiences that would set them up for their future careers. 

But neither youth apprenticeship nor career-connected learning ever became central to the American education system. The lesson from the 1990s is not that these ideas were flawed. It’s that they were never fully embraced (The Apprenticeship Act never even came to a floor vote).

Today, as AI’s initial impacts hit younger workers hardest, these approaches are more relevant than ever.  Youth apprenticeship offers workers the opportunity to learn and earn simultaneously, gaining experience, skills, and professional networks.  Skilled service offers similar opportunities: Governors like Spencer Cox of Utah and Wes Moore of Maryland are looking for ways to create paths from service to careers. And modern career-connected learning efforts led by nonprofits like Britebound are doubling down on mentorship and real-world experience.

Ways We Missed the Mark

We were, perhaps, less prescient in our signature safety-net response to trade displacement, Trade Adjustment Assistance. It was intended to support workers negatively impacted by globalization by providing income support and, in theory, retraining. But the scale of the effort never matched the level of need, and results were mixed at best. Many workers never received support due to insufficient funds or bureaucratic barriers. Others saw benefits expire before the end of their retraining programs.   

Policymakers in the ‘90s also missed the mark in supporting the most-affected regions.  Displaced workers rarely moved to where jobs were. Most stayed put, often taking lower-paying, service-sector jobs or leaving the workforce entirely. Communities in the Rust Belt and the furniture and fabric-making regions of the Mid-Atlantic took a generation to recover. In short, programs like trade adjustment assistance cushioned individuals while doing little for the communities they came from.

Though AI’s impact will be more diffuse than the trade-related impacts of the ‘90s, it still will not affect every place equally. Some regions will be more resilient, particularly those with a wider range of occupations able to absorb displaced workers. Other regions may struggle to adapt. Ignoring that reality risks repeating the same mistakes. A regional lens allows public sector investment to focus on industries emerging within a particular area and the pathways into them.  Rather than expecting workers to move to opportunity, policy can bring opportunity—and preparation—to them.

For Congress, that should include expanding the Reemployment Services and Eligibility Assessments (RESEA) program, which provides individualized career counseling, job search assistance, labor market information, and resume support. RESEA is one of the most effective tools for reconnecting unemployed workers to jobs, with each dollar invested saving the government four dollars in avoided unemployment insurance costs. Those high-ROI programs are the ones the government needs to be investing in as we brace for AI’s continued impact.

History Doesn’t Repeat, It Rhymes

Of course, today’s moment differs from the ‘90s in important ways.  

The biggest difference is speed and scale. When manufacturing jobs disappeared in the ’90s, the pace of change — though devastating in concentrated communities — unfolded over many years. AI is moving faster and will impact many more industries. The workers in today’s disrupted roles may have a fraction of the adjustment time that displaced manufacturing workers had, and the systems meant to support them haven’t gotten any larger or better-connected than they were 30 years ago. If programs were underpowered to meet the scale of change in the ‘90s, they are woefully underprepared for the scale of change that may happen next. 

Today, we have a window of opportunity to respond differently. We can lean into the promising practices of the ‘90s while also upleveling our response, creating the foundation to support workers at the scale and speed of AI. If we can learn from, and improve upon, what we did 30 years ago, we’ll be better prepared to lead in the economy of 30 years from now.

The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.

This story was originally featured on Fortune.com