An interview with Dave Cole, chief innovation officer for the State of New Jersey
This is the fourth installment of Borrow and Steal, Vital City’s monthly interview series profiling evidence-backed policy ideas from around the world and the people translating them into reality.
Our guest is Dave Cole, chief innovation officer for the State of New Jersey. Dave’s team — the New Jersey Innovation Authority — was codified into state law in January 2026, turning what since 2018 had been a temporary grant-funded office into a permanent state agency. The team takes on tasks like building resident-facing digital services, identifying AI use cases and setting AI policy and running data-driven outreach to connect New Jerseyans to public programs.
New Jersey is an interesting case study in digital service delivery in part because it was the first state to make a generative AI tool — something known as the NJ AI Assistant — available to all its employees. Cole’s office led the rollout and has also helped build a range of more advanced AI applications. For example, his office used machine learning to clean data from various state programs to identify students eligible for federally funded summer meals. It also built a computer vision tool for the state Economic Development Authority to help process PDF applications, cutting review time by 20 days.
Nationwide, cities and states have struggled to reckon with how to harness the benefits of AI while minimizing risks to workers and residents. In New York, AI emerged as a focal point in the contentious NY-12 congressional primary. Gov. Kathy Hochul recently signed a moratorium on new data center construction, while simultaneously praising the technology’s utility in helping the state review and clean up antiquated laws. Meanwhile, in the city, Mayor Zohran Mamdani quickly shuttered a botched AI chatbot launched during the Adams administration — and announced a new push to deploy Public Interest Technology crews to fix digital services in select agencies.
In our interview, Cole walks through how the state built its robust digital service and AI infrastructure, how it decides which use cases to pursue and which to avoid and what he’d tell other government agencies earlier in the journey.
The following interview has been edited and condensed for clarity.
Cara Eckholm: Can you explain what a chief innovation officer does, and how the New Jersey Innovation Authority differs from a CTO’s office?
Dave Cole: The Innovation Authority encompasses the state’s digital service operation, a policy team focused on innovation guidance to agencies and the governor’s office, and a communications and engagement team that uses data-driven marketing to ensure residents — particularly underserved ones — are aware of and accessing government services. It’s a multidisciplinary team focused on using technology to improve outcomes.
Cara Eckholm: And does New Jersey have a separate CTO’s office? What’s the relationship between innovation and IT?
Dave Cole: We do, and I think the New Jersey model is fairly unique and effective. My role sits in the governor’s cabinet, which gives us direct access to the governor’s team — particularly important with a new governor (Mikie Sherrill), who came in January with a strong focus on operational excellence. The CTO leads the Office of Information Technology and is a close colleague. The rough division is that they start with core IT infrastructure — networks, servers, cloud providers — while we start with resident needs and program outcomes. We meet in the middle. Under our authority structure, the CTO also sits on our board and my role is written into law to advise and support his office.
Cara Eckholm: So you’re doing the user research, defining agency needs, developing the strategy and then collaborating with the CTO’s office on the build?
Dave Cole: We do a lot of the development and building ourselves, either with our internal team or with partners. To me, innovation is about using available tools to get effective outcomes — very often that means doing something differently. We’re not focused on R&D for its own sake; we’re focused on solving core problems facing New Jersey residents. The CTO’s office might provide cloud platforms like AWS or Azure; we might use those platforms to develop a resident-facing web service.
Cara Eckholm: Can you walk us through the history of the Innovation Authority? My understanding is the office was philanthropy-funded for a period before being codified into law.
Dave Cole: The predecessor, the Office of Innovation, has operated since 2018 on a state budget grant — essentially an appropriation that funded our team. As our scope grew, particularly through the COVID pandemic, it became increasingly clear we needed a dedicated statute that gave us a clear legal basis for the work we do with agencies. That bill passed and was signed in January. The authority structure also lets us continue receiving state appropriations and, where helpful, private grants to scale experimental work or areas where we lack state resources.
Cara Eckholm: What does the team look like today?
Dave Cole: About 80 full-time, plus contractors. The core disciplines are product management, human-centered design and research, content design and software engineering. We also have data analysts, communications and engagement specialists and operations staff who make it all run.
Cara Eckholm: Your office has taken the lead on AI policy for the state and on identifying productive use cases. How do you set policy in such a rapidly changing environment, and how do you determine when AI is actually the right tool?
Dave Cole: We started with a whole-of-government strategy. The previous governor established an AI Task Force that I think was effective at setting principles: take advantage of generative AI to improve service delivery, but with a clear focus on equity — don’t let the technology exacerbate existing inequalities. We bring technical expertise to the evaluation so we know what we’re getting into, and ensure guardrails are in place where necessary.
We built training available to the entire state workforce covering what generative AI is, common use cases and key considerations — including that AI models may reflect biases from their training data, and that any AI-generated work requires human review and accountability, not just a cursory sign-off. We also knew that prohibiting the technology outright rarely works: People are already using it in their personal lives. So we paired the training with practical tools.
Cara Eckholm: Tell me about the NJ AI Assistant. Who has access, and how widely is it used?
Dave Cole: The full state employee list has access through single sign-on. We pair it with the training — employees can test prompts in real time as they go through the training, then continue using the tool afterward. About 20% of the state workforce has logged in, and we’ve now seen over 1.2 million prompts generated through the platform. Usage ranges from people who log in once during training to employees who use it daily as a core productivity tool.
Cara Eckholm: How would you describe the tool’s functionality? Is it essentially a ChatGPT-style interface, or have you built more advanced capabilities into it?
Dave Cole: Think of it as a chat interface — you send requests, upload documents, get analysis — but hosted on state infrastructure with appropriate data and security controls. We use an open-source framework so we can add features over time, though we’re deliberate about keeping it accessible. For more advanced use cases, development teams across the state use AI for code generation — tools like Claude Code or open-source alternatives — backed by models procured through state cloud platforms.
Cara Eckholm: How did you decide to build your own tool rather than give employees enterprise access directly from one of the foundation model companies?
Dave Cole: Partly, we already had something in development for our own team. When we partnered with Innovate US to build out the training, building on state infrastructure gave us the most control over IT security. We had access to OpenAI models through Azure and Anthropic models through AWS, so we could offer the latest commercial models hosted on our own infrastructure with our own security regime in place.
Cara Eckholm: What was your timeline relative to the 2023 moment when ChatGPT changed everything?
Dave Cole: We released the assistant about six to nine months after ChatGPT’s release. Our earliest generative AI use cases were just using ChatGPT directly with public information — once we saw those early use cases land, we invested in building an application connected to the models via API rather than through the public chat interface.
Cara Eckholm: How do you choose between models as the landscape evolves?
Dave Cole: We have a general-purpose model for the main chatbot and then for custom applications we’ll score a variety of models against the specific use case. Cost is a real factor — some inference tasks can use more efficient models, others need higher reasoning capability.
Cara Eckholm: What use cases are most popular among state employees?
Dave Cole: Within the AI Assistant itself, pretty traditional generative AI: drafting and reviewing documents, brainstorming, data analysis from uploaded spreadsheets. The tool includes built-in reminders about human review and checking results. When we work with specific agencies, we don’t start by assuming AI is the answer — we identify the user need first. But there are recurring patterns. Document uploads come up constantly: tax clearance certificates, eligibility documents for benefits programs. The underlying problem is consistent enough that we’ve built reusable tools around it.
Cara Eckholm: Your website mentions a tool you built for the Economic Development Authority. Can you walk me through that use case?
Dave Cole: For small business economic development programs, uploaded documents were sitting in review queues for up to three weeks before someone could verify accuracy. We built a tool that uses an AI model to cross-check uploaded documents against a template and a prompt listing what to look for — validating that required fields are present, dates are current and that the document is the right one. The reviewer is now just checking AI outputs against source material rather than doing the review from scratch, which dramatically compresses cycle time. We also surfaced that validation to applicants at upload: If there’s a potential issue, we flag it and let them fix it before submitting. That check alone can save another three-week cure cycle.
Cara Eckholm: Did the Economic Development Authority come to you with the problem, or did you surface it?
Dave Cole: They came to us. They were concerned about customer experience and processing time, and asked whether tools existed to speed things up without sacrificing quality. That led to a pilot, which produced good results, and we’re now scaling the document review tool across human services, labor, education and other programs through funding from a Public Benefit Innovation Fund grant we received last year. The goal is a general-purpose platform that multiple agencies can access independently.
Cara Eckholm: Do you have outcome data on processing time reductions?
Dave Cole: Yes. In the EDA case, the baseline review cycle was around 21 days, and intricate documents could take an hour for a human to check manually. The AI model essentially eliminates that review cycle if it’s available at the point of upload. For memo generation from large source files, we’ve heard of processes that previously took a full week of staff time compressed to 15 to 30 minutes of AI-assisted drafting, followed by an hour or two of review.
There are real inefficiencies in government, but there’s also a heightened standard. We have to earn resident trust, which means considering factors private companies might not.
Cara Eckholm: What other big wins stand out?
Dave Cole: The AI training and assistant framework itself, as a model for responsible adoption — not blocking the technology but clearly communicating trade-offs and constraints so employees can proceed with the right tools. Beyond that, two examples stand out.
In partnership with the Department of Labor, we used AI to rewrite email and print notification templates in plain language. A designer on our team used AI prompts to automate much of that generation and then reviewed the outputs. The result was roughly 30% faster response rates from residents receiving notices — people took the required action more quickly because the language was clearer. We’ve now done over 100 template updates in just one program (unemployment insurance). Those Figma files and the research brief are available open source so other states can build on it.
The other is the Summer EBT program. Students enrolled in SNAP, TANF or Medicaid are eligible for summer meal benefits, but deduplicating enrollment lists across those programs is complex. We built a matching tool using traditional machine learning — not generative AI — to cross-reference datasets and eliminate duplicates. The result was cleaner program integrity and access to roughly $20 million in additional federal funding flowing to New Jersey families for summer food assistance.
Cara Eckholm: Government is often cited as a sector with lots of low-hanging AI fruit. Are there use cases you’ve tried that didn’t pan out, or areas where you think the hype outpaces the reality?
Dave Cole: Honestly, there are more use cases we’ve deliberately avoided than ones we’ve tried and abandoned. We try to be very careful upfront about whether AI is actually the most efficient and appropriate tool for a given problem. There are real inefficiencies in government, but there’s also a heightened standard. We have to earn resident trust, which means considering factors private companies might not.
One concrete area we’ve avoided is public-facing chatbots. If content is hard to find on a government website, the answer is usually better content design and information architecture, not a chatbot layered on top of a broken site. We’ve found chatbots more effective as internal tools where staff have enough domain knowledge to recognize when outputs aren’t quite right.
Cara Eckholm: So a chatbot isn’t going to fix a bad website.
Dave Cole: That would be my position.
Cara Eckholm: How are you thinking about AI’s labor implications for the state workforce?
Dave Cole: Both governors I’ve served under have been clear: AI is not a job replacement tool in New Jersey. We do see cases where work is repetitive and could benefit from automation to reduce human error. So we’re looking at applying AI where it can improve quality of life for employees or expand capacity in areas where we haven’t been able to hire. We’ve had open conversations on our own team about this. Viewpoints range from skepticism to cautious optimism, and we value all of them. In the private sector it’s a different story, and I wouldn’t presume to speak for those decisions, but I’ll note that there’s a lot of capital moving around and AI often gets used as the narrative for layoffs that have other drivers.
Cara Eckholm: What’s your advice for states earlier in their AI journey? And relatedly, how replicable is the Innovation Authority model?
Dave Cole: They’re intrinsically linked. The reason we were able to move quickly on the AI Assistant is that we had years of investment building a team that understood the technology and the state’s IT ecosystem. That capacity is the foundation, before any specific AI initiative.
Many states are doing this now; we’re not alone. But we benefited from being early, and the team we’ve been able to attract is what makes it possible. For any state without that investment yet, the team is where to start. There are flexible models — we began as a budget grant backed by a nonprofit for staffing support, and evolved into a formal authority. I am always happy to talk to anyone in the public sector trying to figure out how to get started.
The other prerequisite is leadership alignment. Gov. Sherrill came in with a clear agenda around government efficiency — not just cost-cutting, but measuring and improving outcomes. When that direction permeates cabinet and agency leadership, it becomes much easier to do this work. People understand it’s okay to try things differently and that there will be support for it. AI fits within that foundation rather than having to justify itself against institutional inertia.
Cara Eckholm: You’ve used both “innovation authority” and “digital services” — are those synonymous in your mind?
Dave Cole: Generally, yes, though we think of the Innovation Authority a bit more broadly. We have four core functions: a resident experience team (digital services for residents), a business experience team (same for businesses), a communications and engagement team focused on performance-based marketing for public benefit programs, and a data and policy team. The C&E work is something I don’t think many states have: It’s essentially data-driven marketing applied to government programs — making sure people know about available benefits, and that outreach reaches communities equitably rather than just those with existing access to information. Any state can start with digital services as the specific focus and the broader structure can follow.
Cara Eckholm: So you’re actually optimizing ad spend for public benefit programs.
Dave Cole: That’s right. It’s a core part of the C&E work.
Cara Eckholm: Are there other states at a comparable level of maturity that people might not have heard about — especially outside the coasts?
Dave Cole: The Beeck Center at Georgetown maintains a digital state network with a comprehensive index of programs across the country. On specific states: California has a range of organizations and can teach us a lot about how to scale digital services. Pennsylvania has strong emerging talent under its new governor. Maryland and New York — at both the state and city levels — have active efforts. Colorado has done significant work on benefits access. Virginia has recently revitalized its digital services work. There’s a good support network for practitioners doing this work.
Cara Eckholm: Last question: What is your relationship to New Jersey’s cities?
Dave Cole: Our primary mandate and funding is tied to state programs, but we engage with local governments wherever those programs intersect. I’ve spoken at conferences of mayors and League of Municipalities events, and we’ve worked with individual municipal leaders, particularly around permitting reform, which sits at the state-local boundary. Our resident-first orientation helps here. If someone needs a building permit, they need it resolved quickly and correctly. Whether the relevant authority is the Department of Environmental Protection, Department of Transportation, Department of Community Affairs or a local agency is our problem to bridge, not theirs. Gov. Sherrill’s permitting initiative follows that model: convene all parties, share information and work collaboratively across jurisdictional lines.
Cara Eckholm: Thanks so much, Dave. This was a great conversation, I learned a ton.
Dave Cole: Thank you — and thanks for raising awareness of our work.
