Written by: Zachary Hill, Chief Executive Officer

AI Is Here. What Is It Really Doing to Work?

When a new technology arrives, public conversation tends to split in two directions. One side promises a golden age. The other predicts the end of work as we know it. We heard versions of both during the Dot Com Boom, and again when smartphones transformed daily life and business operations in the 2000s. Today, artificial intelligence and robotic automation have revived that same anxiety, only faster and louder. But the strongest workforce evidence suggests a more grounded conclusion: AI is changing the composition of work more than eliminating work outright. The real question is not whether people, humans really, will have work to do. It is whether workers, schools, and communities can adapt quickly enough to the new mix of tasks, tools, and expectations now taking shape (Autor; Acemoglu and Restrepo).

That distinction matters. Economists have long argued that technology affects labor markets through two competing forces. One force displaces workers from tasks machines can do more cheaply or quickly. The other creates new tasks, new products, and new forms of demand that still require people. Daron Acemoglu and Pascual Restrepo describe this as the tension between displacement and reinstatement. Their point is not that technology is harmless. It is that labor market outcomes depend on whether new human work emerges fast enough to offset what gets automated (Acemoglu and Restrepo). David Autor makes a similar case in his review of technological change, tracing a path from early computer optimism to today’s uncertainty around AI. History does not tell us that every worker is safe. It does tell us that “technology arrives, jobs vanish forever” is too simple to be a serious analysis (Autor).

That historical perspective is useful because previous tech booms did not produce a total collapse in work. In fact, a recent National Bureau of Economic Research chapter on emerging digital technologies found an overall positive net impact on U.S. employment from 2012 to 2019, the period when smartphones, cloud tools, and platform-based digital systems became deeply embedded in the economy (Borgonovi et al.). The Dot Com Boom raised demand for digital, analytical, and technical work. The smartphone era reorganized communication, retail, logistics, customer engagement, and app-based services. Jobs changed. Some shrank. New ones appeared. The labor market absorbed the shock unevenly, but it did not disappear (Borgonovi et al.; Autor).

AI and robotics do appear to be different in one important way. Earlier waves of automation often hit routine production and lower-wage work first. Brookings finds that generative AI exposure is more concentrated in highly educated, higher-paying regions and occupations than earlier automation waves were (Muro, Methkupally, and Kinder). That matters for California, where the places most exposed to AI are not the places many people traditionally picture when they think of automation. According to Brookings, estimated workforce exposure to generative AI ranges from 42.8 percent in Santa Clara County to 26.7 percent in Mono County (Muro, Methkupally, and Kinder). In practical terms, California’s most AI-exposed regions are often its knowledge-intensive regions, not just its factory floors or warehouses.

National labor data points in the same direction. The U.S. Bureau of Labor Statistics (BLS) says AI is likely to have its clearest near-term effects in occupations whose core tasks are easiest to replicate with generative systems. BLS specifically highlights roles such as customer service representatives and medical transcriptionists as more exposed to decline, while pointing out that software developers may benefit from rising demand because firms still need people to build, maintain, govern, and integrate AI tools across organizations (Machovec, Rieley, and Rolen). This is a useful corrective to the popular storyline that only manual work is vulnerable to automation. Some of the most immediate pressure is showing up in office-based roles built around documentation, routine communication, transcription, or standardized information processing (Machovec, Rieley, and Rolen).

California’s own projections make that contrast even clearer. The California Employment Development Department projects that customer service representatives in California will decline from 199,900 workers in 2022 to 193,500 by 2032, a drop of 3.2 percent (Baldassare et al.). Word processors and typists are projected to fall much more sharply, from 9,100 to 5,600, a decline of 38.5 percent (Baldassare et al.). At the same time, many occupations that involve physical presence, complex judgment in changing environments, or direct patient interaction continue to grow. California projects physical therapists to grow 27.2 percent from 2022 to 2032, diagnostic medical sonographers to grow 24.1 percent, and medical assistants to grow 25.6 percent (Baldassare et al.). These jobs use technology, sometimes heavily, but they are not reducible to software alone. They require observation, communication, dexterity, trust, and real-time decision-making in human settings.

That does not mean robotics and automation are irrelevant outside offices and hospitals. Quite the opposite. Robotics continue to reshape logistics, advanced manufacturing, and maintenance work. But here too, the evidence is more nuanced than the phrase “robots are taking jobs” suggests. BLS notes that even highly publicized technologies such as autonomous trucking have not yet produced the dramatic labor market collapse many predicted. In fact, BLS reports that heavy and tractor-trailer truck driver employment rose from about 1.7 million in 2012 to more than 2.2 million in 2023 (Machovec, Rieley, and Rolen). The lesson is not that automation never matters. The lesson is that adoption is slower, messier, and more constrained by regulation, infrastructure, safety, and real-world operating conditions than public debate often assumes.

California’s public mood reflects both uncertainty and realism. In a December 2024 Public Policy Institute of California survey, 52 percent of Californians said they viewed AI unfavorably, and half of employed adults said AI would decrease the number of jobs in their industry (Baldassare et al.). At the same time, large majorities supported increased public funding for job training (Baldassare et al.). That combination is telling. Californians are not simply anti-technology. They are signaling that they believe transition costs are real, and that adaptation will require investment, not slogans.

So what should educators, school leaders, employers and community stakeholders take from all of this? First, the evidence does not support a blanket message of despair. Work is not ending. Second, the evidence does not support complacency either. Tasks are being redistributed inside occupations, and that will reward workers who can combine technical fluency with human judgment, communication, troubleshooting, and problem-solving (Broecke; Autor). Third, California’s challenge is statewide, but uneven. Regions built around knowledge work may feel AI pressure sooner, while other regions may see slower but still significant shifts through automation, scheduling software, diagnostics, and workflow tools (Muro, Methkupally, and Kinder). The practical response is to prepare students and workers for co-work with robotic automation and AI, not just tool use. Knowing how to use AI matters. Knowing when to question it, interpret it, and build around it may matter even more (Broecke).

That may be the most useful aspect of the Dot Com and smartphone eras to remember. New technology does reduce demand for some tasks. It also creates new expectations, new roles, and new combinations of skills. The future of work in California is unlikely to belong to people who can be replaced by a prompt or a machine. It is more likely to belong to people who can work with new tools while still doing what tools cannot do well on their own: build trust, solve messy problems, adapt in context, and make sound judgments when the script runs out (Autor; Broecke).