Government agencies are moving artificial intelligence (AI) pilots into production at a pace that could make the public sector the largest adopter of agentic AI, according to Paul Tatum, executive vice president of global public sector solutions at Salesforce.
As agencies face mounting budget pressures and citizen expectations for digital-first experiences, Tatum said AI agents are already proving their value in real-world government operations.
“Governments will be the largest user of agentic technologies of any industry,” Tatum said during an interview with MeriTalk. “The need is great, right? Backlog, large customer base, millions, if not billions, of people … and most of the public-facing data that you need is out there.”
“I think we will see the ramp and acceleration of agentic technology in a significant way in the next 12 months,” he said.
The momentum comes as federal leaders increasingly view AI as a tool to improve citizen services, increase workforce productivity, and modernize aging digital experiences.
Tatum pointed to new IDC research sponsored by Salesforce showing that 85% of government leaders believe AI is significantly improving productivity within their organizations. At the same time, advances in AI technology and growing confidence in its capabilities have accelerated adoption across agencies.
From experimentation to execution
What has changed most over the past year, Tatum said, is the shift from experimentation to operational deployment.
For example, Tatum highlighted an Agentforce agent called Case Advocacy Librarian, or CAL, that Salesforce recently deployed at a government agency.
“That agent is now active. It’s been tested, it’s been scaled, it has access to all of the documentation to help that department with tax advocacy services,” he said.
That deployment is one of several government AI initiatives Salesforce is supporting. Tatum pointed to work with the Department of Veterans Affairs, where the agency is using Slack and Salesforce AI capabilities to improve services for veterans, as well as efforts supporting military recruiting and other Defense Department-related missions.
He also cited deployments spanning city services, postal operations, and local law enforcement.
“We think of these agents as almost part of our digital team or our organization in government, actively working on behalf of citizens or the employee to bring productivity,” Tatum said.
Salesforce launched Agentforce Public Sector in August 2025 to augment government workers with digital labor. Through the platform, government agencies can use AI agents to assist employees and constituents.
Building enterprise-grade AI
While enthusiasm for AI continues to grow, Tatum said one of the market’s biggest misconceptions is that organizations can quickly build enterprise-grade AI solutions without the governance, security, and workflow infrastructure required for government environments.
“You’re not going to build an enterprise application that needs security, scale, permissions, workflows, and context to the data in your enterprise … with an AI coding app,” he said. “There’s a different level of responsibility the government has for the trust of the data, the veracity of the answer, and then the overall access to citizens at scale. That’s what we do at Salesforce.”
According to Tatum, that focus on enterprise-scale deployment is where Salesforce stands apart in a crowded AI market. Rather than concentrating solely on large language models (LLMs), Tatum said Salesforce differentiates itself through a complete agentic platform.
“The crowded part of the market right now is talking about LLMs, but our differentiation is that we offer a full agentic platform … with all the capabilities government needs to get going on day one, right out of the box,” he said.
Tatum highlighted Salesforce’s recently announced Headless 360 capability as another differentiator.
The offering allows organizations to pair AI coding tools with Salesforce’s underlying platform to rapidly build applications while maintaining enterprise-grade security. He said the approach could dramatically shorten traditional government software development timelines while reducing risk.
“That’s a massive new opportunity for government, because typically writing software, testing software, scaling software is a multiyear, multibillion-million cycle – and a lot of them fail,” Tatum explained. “I think that’s going to completely change with agents and coding tools when they’re pointed and connected to powerful platforms like Salesforce.”
Tatum argued that one of the most promising opportunities for government AI lies in improving how citizens interact with agencies online. Many government websites remain difficult to navigate, he said, with complex processes spread across thousands of pages, documents, and links.
Agentic AI can simplify those interactions by serving as a conversational front door to government services. Salesforce has already seen success with that model through its own help portal, help.salesforce.com, where AI agents handle millions of customer interactions.
As agencies evaluate the return on their AI investments, Tatum said the most important step is measuring outcomes against existing processes. Metrics such as case resolution times, accuracy, escalation rates, and service costs provide a clear picture of mission impact.
“We are seeing that the average cost of a phone call into a government call center is $5 to $6,” Tatum said. By comparison, he pointed to a deployment with the Thames Valley Police in the United Kingdom, where an AI agent named “Bobbi” reduced engagement costs to approximately 26 cents, while maintaining service quality.
“The cost savings are significant without compromising the quality of the response, the personalization, or the accuracy of it,” he said.
Preparing for an agentic future
Looking ahead, Tatum expects the next wave of adoption to focus on citizen-facing services, particularly multilingual AI agents capable of handling increasingly complex interactions. He predicts AI-enabled government experiences will become as commonplace as AI-powered search has become for consumers.
Still, he emphasized that government agencies must balance speed with responsibility by choosing trusted partners and building for long-term sustainability rather than short-term experimentation.
“Think about the scale and long-term durability of your agentic enterprise, and if you have the partnerships in the industry to really drive [it],” Tatum advised.
“You’re going to have thousands of agents across an enterprise. How do you manage them? How do you observe them? How do you connect them? How do you make sure they’re doing the right thing at the right time with the right question, the right data? That’s what we’re offering, and it’s super important for government,” he said.