U.S. Army’s Advice for Industry About AI-Related Solutions

The “easy” way of picking one prime contractor to handle everything artificial intelligence (AI) for the U.S. Army is one way to go forward. However, it will not be the method the service uses to secure large language, deep reinforcement learning and natural language models; development of algorithms; or testing and securing of AI tools, leaders say.

The Army has slated funds for AI to go into the service’s programs of records in about a year under the fiscal year 2026 program objective memorandum (POM) cycle. Acquisition leaders in the Office of the Assistant Secretary of the Army (Acquisition, Logistics and Technology), or ASA (ALT), have specified that they will be turning to a number of companies across a wide AI solution landscape to fill their needs.

“The ‘easy button’ would be to go to one prime—and they all say they will solve all your AI problems, that they’ve got it all figured out, with a turnkey solution,” said Col. Chris Anderson, USA, project manager, Intelligence Systems and Analytics, Program Executive Office – Intelligence, Electronic Warfare and Sensors.

“That would really be the easy button, but it would be a disservice to really the platform and the soldier that needs it because it is evolving so rapidly,” he said. “Nobody has the code cracked on every bit of AI. It is really going to take kind of a mix of government, industry, academia, large and small business.”

Col. Anderson manages the Army’s Project Linchpin effort to bring AI into Army operations. He spoke on a panel Wednesday with Jennifer Swanson, deputy assistant secretary of the Army for Data, Engineering and Software, ASA (ALT), and Dan Snyder, senior director and analyst at Gartner, at AFCEA International’s TechNet Augusta Conference in Augusta, Georgia.

“There has been a lot of buzz, certainly everywhere with regard to AI ... but nobody has solved all of it. That's 100% true,” Swanson said.

In fact, over the last two years of starting to pull in more AI capabilities, the “vast majority” of industry responses to AI-related solicitations, white papers or requests for information have come from small businesses or nontraditional defense companies.

“That was definitely a lesson learned,” Col. Anderson noted. “Just the [sheer] broadness of the AI ecosystem.”

In turn, this is pushing industry AI solutions to be interoperable from the start, Snyder observed.

“I think the beauty of Linchpin is that it forces them to be interoperable,” he said. “On the commercial side, a lot of vendors that we're speaking with in AI tend to want to boil things down to ‘a stack,’ ... the AI infrastructure, the models and some of the applications. And it's tough to get them out of that mindset when we are developing the solutions that they are enthusiastic are going to serve the needs of military most significantly. The beauty of Linchpin is kind of breaking that mold and forcing everything to quite explicitly be integrated and interoperable.”

The leaders advised industry to stop claiming they can do it all and to clearly identify their particular AI-niche.

“I would say 18 months ago, we were getting a lot from the big primes, you probably know who they are, and they were saying, ‘Hey, we have your turnkey solution, Army,'” Col. Anderson stressed. “But I think the message has gotten out there: It is going to be a whole bunch of industry partners. We are looking for best of breed and easily replaceable. So, figure out what your niche is, what is your passion project, where do you fit in this. Nobody is going to be doing it all, but if you are really into testing, AI, I think there's a place for you at the table. If you're into data curation or data labeling, there's a space for you to do that. And then industry, academia, whomever is actually developing the AI models and algorithms—that's another component.”

The ASA (ALT) leaders are now pursuing a 500-day plan for AI, to see what is possible from industry, prepare for the POM cycle and develop requirements to field of actual AI solutions to soldiers. They are building on their so-called 100-day sprint plan completed in June.

The 100-day sprint resulted in a “playbook” for AI acquisition, the identification of Army and Department of Defense Science and Technology investments that could be aligned to the service’s AI strategy, the creation of a generative AI policy and #DefendAI, an AI-layered defense framework, as well as AI training for the ASA (ALT) acquisition staff.

“We know that there are some significant gaps that still need to be solved,” Swanson stated. “For example, how do we test it? If you ask ChatGPT the same thing repeatedly, you don't always get the same answer because it's not a deterministic model like with software, so how do you test that to make sure that when you put something like that in a soldier's hands, that it is going to give the right answer? That is one of the things that I know you all will really help us to work through as part of the 500-day plan that we've just kicked off.”

Because the POM funds are not in place until fiscal year 2026, the Army will continue to pursue AI-related contracting through Small Business Innovation Research and other alternative contract vehicles, Col. Anderson said.

“We don't have POM money until FY26 [fiscal year 2026],” Swanson added. “However, there has been a tremendous amount of work done. You guys have been able to leverage some dollars from other areas to do data labeling work and to really award a few small contracts to start exploring what this would look like.”

Nevertheless, the service’s AI effort will hinge on industry capabilities, and the leaders plan to continue the great pace of meeting AI-related companies.

“We have done over 500 one-on-one engagements with industry, academia and other government agencies, just trying to understand the Army and the industry AI ecosystem, and then trying to figure out, how does that apply to a PM [program manager] shop?” Col. Anderson said.

The colonel encouraged industry to respond to its current request for information on AI that is still open for comment.

“I always have to give a shout out to the Linchpin team, which started out as two people,” Col. Anderson shared. “We are up to 20 people now, but it is a very small team moving fast. Over the last two years, we've done multiple RFIs [requests for information], evaluated multiple white papers and responses from industry that have actually shaped our next RFI. We're taking your feedback and using that to fine tune our efforts.”