CBP and Commerce Put AI to Work Inside Federal Systems

U.S. Customs and Border Protection is using artificial intelligence to rewrite parts of its legacy software and speed changes to the federal system that manages commercial operations. The agency says the tools have shortened modernization timelines for some components by 40%, giving officials new ways to process data and expand the Automated Commercial Environment.

The State Rebuilds Its Machinery

Thomas Mills, executive director of the Cargo Systems Program Directorate at CBP, said the agency has used AI for facial recognition and entity resolution for years. Now it applies the technology to legacy code, including stored procedures, databases and software that parses inbound messages. Some of that code was written in C and C++.

At a recent panel at the ATARC AI Summit, Mills said CBP had seen “right about a 40% acceleration in in our ability to be able to move legacy components into modernized components.” The agency also uses AI for automated data extraction and data governance. Mills cautioned that “the whole garbage in, garbage out principle is still relevant,” warning that large language models produce bad results when they work from bad data.

CBP developed several underlying components of the Automated Commercial Environment in the early 2000s. The agency recently moved off and retired its mainframe from the 1980s, Mills said. CBP had estimated modernization would take five to seven years. “The components that we thought would take a long time, we've seen a 40% compression in the timeline, so code generation is real,” he said.

That acceleration doesn't mean generated code runs unchecked. Mills said staff with the necessary expertise review the output instead of trusting it blindly. People still assess the work the tools produce.

Small Sprints, Big Systems

CBP uses two-week development sprints for many software projects. Mills said dividing work into small, iterative stages helps the agency estimate costs, explain results and seek funding. He described the method as “small chunks, fail fast and then find out ways to scale.”

The approach makes modernization legible to the institutions that approve and fund it: teams break work into stages and use results to make the case for the next step. That’s the machinery of government modernization as its own officials describe it—software, cost estimates, funding requests and decisions about what to scale.

Commerce Connects Old Data to New Models

At the Commerce Department, bureaus and offices apply AI to modernization, mission work and back-office functions. Matt Passos, acting chief technology officer for Commerce, said improving data is the shared goal across the department’s use cases.

Commerce works with partners and the General Services Administration's USAi platform. Application programming interfaces give older data systems access to AI and help clean and standardize their data, Passos said. Census, the National Institute of Standards and Technology, and NOAA use systems that access data and run queries through USAi.

Commerce also uses AI to analyze capital-planning data, estimate total cost of ownership and inform future investment decisions. Passos said agencies face uncertainty about AI pricing and the challenge of quantifying its value. The tools may make data easier to use, but agencies still have to decide what those systems cost and which investments to pursue.

Ryan Palmer, USAi architecture lead at GSA, said agencies need to use new models as they become available and integrate USAi with multiple cloud providers. The platform evaluates models from leading American AI companies and makes them available through FedRAMP-authorized platforms. More than 20 agencies use USAi, Palmer said. He described AI as a way to help people unlock data and expand legacy-system capabilities without requiring a full system rewrite.