TLDR
- Salesforce is expanding its work with the U.S. Department of Transportation to add AI agents that automate routine tasks while keeping humans in the loop.
- The Transportation Department plans to use AI agents for citizen support, data analysis, and grant processing as part of a broader modernization effort.
- Salesforce is positioning itself as an AI orchestration layer for government, allowing agencies to choose among multiple large language models under strict security controls.
Salesforce is deepening its long-running technology relationship with the U.S. Department of Transportation to modernize its public services, automate routine work and improve citizen engagement.
The expanded initiative centers on unifying major USDOT systems on the Salesforce platform and adding Salesforce’s Agentforce AI agents to handle tasks ranging from consumer complaints to data analysis and grant administration, according to the software giant.
Paul Tatum, executive vice president of global public sector solutions at Salesforce, said the current administration is acting more quickly than prior administrations to adopt technology that can make government more efficient.
He pointed to President Trump’s AI Action Plan that directed all departments to mobilize on AI. “You have to get it into your departments because it brings efficiency, speed, and, frankly, credibility to a technology that can be transformational for a country,” Tatum said in an interview with The AI Innovator. The administration’s focus is “very encouraging.”
The Department of Transportation agreement builds on years of foundational work focused on data consolidation, workflow automation, and security compliance. Salesforce said it has already helped USDOT streamline core functions across the agency, including citizen complaint portals, interstate data sharing, and grant management. The addition of AI agents is intended to augment those systems rather than replace human workers.
AI agents to assist civil servants
At the center of the effort is Agentforce, Salesforce’s AI agent platform, which Tatum described as a way to let an AI workforce assist civil servants. “It’s not just about the AI or LLMs,” he said. “It’s about the agent layer and ‘digital employees’ that can work alongside us or government civil servants to augment and take action in the context of the work that they’re doing.”
Security and compliance are a major reason agencies have been cautious about deploying generative AI. Tatum said Salesforce invested heavily to obtain FedRAMP High authorization for its platform, including Agentforce, to meet federal security and privacy standards. “That gives our government and specifically federal customers the assurance and understanding and confidence that we are delivering rock-solid security measures,” he said.
According to Salesforce, USDOT plans to deploy AI agents across three primary areas: citizen support, data analysis, and grant administration. AI agents will be used to provide around-the-clock assistance for routine inquiries and complaints, analyze large datasets such as traffic and weather data to generate alerts, and review grant applications to flag missing information and compliance issues for human review.
Tatum emphasized that the AI deployments remain a “vision” as agencies move carefully to prepare the groundwork. “The next step for them is to activate the AI in Salesforce to be able to augment the work inside of those three use cases,” he said. DOT CIO Pavan Pidugu has “really leaned into making the vision for the department a reality through technology.”
Internally at the DOT, AI agents could do things like analyze data flowing in from all 50 states to help identify infrastructure issues more quickly. For grants, agents would not decide who gets funding, but would determine whether applications are complete. “The goal in any government organization is, ‘let’s let the AI do the mundane stuff so our overworked civil servants can do more valuable higher-order stuff’,” Tatum said.
Grounded on government data
Salesforce provides the orchestration layer that lets DOT employees choose among large language models to fit their needs. Tatum said workers can choose among models from providers such as OpenAI, Anthropic, or Google through a drop-down menu. “The government uses them all.”
That flexibility matters in a fast-moving AI market, Tatum said, where models “are leapfrogging each other weekly.” Salesforce also has agreements with model providers not to train their models on customer data without permission, a point Tatum said is critical for government use cases.
Concerns about hallucinations remain a major obstacle for generative AI in regulated environments. Tatum acknowledged that hallucinations originate at the model level but said Salesforce adds multiple layers of control, including testing, citations, data masking, and policy grounding. “No government is going to use a general-purpose LLM trained on the internet to do anything,” he said. Instead, models are trained on agency-specific policies and documentation to serve as grounding.
We needed to make a strategic shift to deliver on our mission: adding AI at the core of everything we’re doing. – DOT CIO Pavan Pidugu
Salesforce also uses its own AI internally as a proving ground. Tatum pointed to the company’s customer support site, which is now largely handled by AI agents. He said the “CSAT (customer satisfaction) scores from customers are really high.”
Success metrics, whether at the DOT, Salesforce or other organization, include frequency of use, accuracy, speed, latency, CSAT and other KPIs. Tatum said Salesforce itself is transforming itself into an agentic enterprise.
Tatum declined to give a timeline for when USDOT’s AI agents would be publicly visible, saying that decision rests with the agency. “I don’t want to speak for them on their schedule,” he said. “They’re actively building out the platforms that I just talked about.”
In a statement, DOT CIO Pidugu said, “To deliver on our mission of building a safe, efficient, and modern transportation system, we have to start with making technology our biggest asset. We needed to make a strategic shift to deliver on our mission: adding AI at the core of everything we’re doing. Now, we’re not just deploying new technology, we’re empowering our agency to work more effectively, and most importantly, ensuring the safety and future of our country’s transportation system.”
