California has decided that understanding whether artificial intelligence is truly destroying jobs requires data, not opinions. In recent weeks, it activated the first state-level tool in the United States designed to monitor AI’s impact on employment. The initial results, far from alarming, paint a complex picture: no mass layoffs yet, but some indicators are already flashing in the San Francisco Bay Area and among workers with a college degree.

Hard data, at last

Until now, the debate on AI and work has been fueled by conflicting forecasts. Some imagine entire job categories wiped out by generative AI, while others bet on a new balance between automation and human creativity. Almost everyone, however, argued without public and verifiable data. The algorithms that power Large Language Models are often developed and deployed in opaque cloud environments, making it difficult for regulators and companies themselves to quantify the real effects on employment. The California tool – whose technical details have not yet been disclosed – represents a first attempt to build an empirical observatory. It cross-references labor market information with AI tool adoption, trying to isolate causal links.

First signals: Bay Area and degree holders under watch

Preliminary data point to a still-stable job market, but with potential tension exactly where the concentration of technological skills is highest. The Bay Area, heart of innovation, shows the first shifts, as does the segment of workers with university degrees. This is not surprising: language models can now automate cognitive tasks, from report writing to code generation, that until a few years ago seemed the exclusive preserve of skilled professionals. This is not yet a wave of unemployment, but a warning sign not to let our guard down.

AI in the enterprise: why on-premise helps maintain control

For companies evaluating the integration of LLMs into their processes – and that for privacy, security, or Total Cost of Ownership reasons choose on-premise deployment – the California experience shines a spotlight on an often overlooked aspect: the ability to monitor automation’s impact in-house. Keeping inference infrastructure within one’s own boundaries allows tracking which tasks are delegated to models, how productivity changes, and whether imbalances emerge that require skill reallocation measures. It is no coincidence that, for those weighing trade-offs between cloud and on-premise, analytical frameworks – such as those offered by AI-RADAR at /llm-onpremise – help evaluate these factors. Internal transparency becomes a competitive advantage, because it allows AI to be adopted consciously, without passively suffering the consequences of ungoverned deployments.

A lesson for the future of work

The California observatory is a wake-up call for all organizations: measuring is not optional. Whether it is a public administration or a company running its own LLMs on local servers, the ability to collect data on human-machine interaction is the foundation for active labor policies. Data sovereignty, in this sense, is not just a GDPR compliance issue, but a tool for social governance. If AI is to be an ally and not a threat, the path runs through public and private observability tools that make visible what is still too opaque today.