California’s Department of Technology (CDT) has been aggressively assessing how AI technologies can benefit and provide the optimal value for the State departments to deliver more effective services to the public. The clearest example of this approach is the recent California-Anthropic partnership, which gives all state agencies, cities, and counties access to Claude at 50% off (per Office of Governor Gavin Newsom, June 29, 2026). Government adoption of AI is accelerating fast right now. The state has also made a significant, initial investment to build and soon deploy a statewide AI assistant, called Poppy. The goal of Poppy is as an in-office productivity assistant, mainly to help departments draft documents, analyze complex data, and research policy rules to improve services between agencies and to the public. In addition, the Department of General Services has recently begun adding AI-focused software providers to the CA Statewide License Program (SLP). California is pushing to ensure that all future IT spending takes full advantage of the value AI can bring the State.
With most technical adoption, including AI, preparation and planning are the key building blocks of success. Specifically in California, along with many other states, AI could bring innovation and streamline how the State replaces the legacy infrastructure that still exists. In recent years, multiple CA IT initiatives have showcased the importance of really understanding these legacy systems and the data that they contain. Taking a data-focused approach across the massive amount of legacy data in these systems is crucial. AI doesn't necessarily fix these data legacy issues and, in fact, is even more dependent on the back-end data than standard implementations. Because of this, implementation of AI requires data readiness as a core component of success and the cornerstone of an AI-driven project.
With the introduction of AI benefits to these modernization efforts, key building blocks include 3 specific items:
- Data standardization: Consistent formats, fields, and definitions need to exist across systems so you're feeding your tools information they can actually understand. AI will not understand your tribal knowledge. An example of inconsistent data that would need to be standardized is fields having different names across different systems. It may also mean taking inventory of your different systems and mapping them out. Standardization isn't a one-time project; it's an ongoing one, as systems and data sources will continue to change and grow. But as a first step toward AI readiness, taking inventory of all your systems and looking for inconsistent data fields is a great place to start.
- Data governance & management: Before implementation, there should be a clear conversation and clear written rules about what's AI-exposable versus restricted, before a tool ever touches it. That's what prevents accidents after the fact. Governance in this context isn't really a policy document. It's decisions and accountability. Governing the data is extremely important for agencies carrying legal exposure or working under federal regulations. Governance has to be a cross-functional exercise - legal, program leads, and privacy officers all need to be involved in creating that outline. And it's not really about blocking AI's access. It's about creating a tool that your staff can confidently use, because they're informed on how to use it.
- Data access & control: While governance focuses on what should be exposed and policy, controlling access is the rules set up to technically hold it in place. These are your enforced technical barriers, whether that's through audit logging, encryption in transit and at rest, or role-based permissions. AI tool security practices need to match the same level as existing systems before they're connected. Without this layer, governance is just a document with rules. Access control is what actually keeps your data safe and protected.
Leveraging and incorporating AI as a key asset to all state sponsored/approved IT projects going forward provides California the unique ability to lead the way in how Departments can prepare to maximize the benefits of AI by initially focusing on tackling the vast amount of legacy data as a first step to planning for what benefits can come from an AI-backed solution. Without this initial focus on backend data and structure, there will be a much harder, riskier and more expensive path to implementing an AI solution. California is setting the tone and is a trailblazer in AI and many other states are taking notice.
As California and other states work through their AI adoption, partners like VIP are helping government clients turn decades of legacy data into an asset rather than a hurdle that they can actually build AI on.
