How AI is Actually Getting Used Inside the Pentagon
Federal procurement officials weren't using large language models two years ago. Now they are. That shift - from zero exposure to active deployment - tells you something about the pace of change inside government.
Nick LaRovere spent years at Palantir before co-founding Pryzm, an AI platform built specifically for companies selling to the federal government. His vantage point spans both sides: he's seen how government buyers operate, and he's building tools to help vendors reach them more effectively.
The basic problem Pryzm solves is straightforward. By the time a contract appears on SAM.gov, the deal is already decided. Real opportunity happens in the months before public posting, when relationships matter and context wins deals. Pryzm pulls data from CRMs, email, Slack, and procurement sources to help companies identify which officials have budget, what they're buying, and how to reach them.
The Defense Budget Opened a Window
Three forces converged recently. First, Congress and the administration prioritized acquisition reform after decades of talk. Second, geopolitical pressure made speed essential. Third, AI made new workflows possible.
The defense budget hit $1.5 trillion. That expanded the addressable market significantly. But more important was the shift toward "commercial first" technology - the idea that off-the-shelf solutions often beat custom-built systems at lower cost.
Companies like Palantir and Anduril had to sue the federal government to get a fair look. That groundwork opened doors for others. Government officials began accepting that innovation could come from outside the traditional defense contractors.
"Colors of Money" and Flexibility
Federal budget categories - called "colors of money" - restrict how agencies can spend funds. A dollar allocated for research can't be spent on operations. The logic is oversight: Congress wants to ensure tax dollars go where intended.
But the system creates friction. Agencies can't reallocate resources when priorities shift or vendors underperform. New flexibility measures, still emerging, would let certain officials control broader budget categories with more discretion, provided they report how they spent it.
This shift matters for vendors. It means government buyers can move faster and take calculated risks on new solutions.
The Procurement Workflow Problem
Large defense contractors operate five to ten different CRM systems across business units. Government agencies do the same. When a federal official meets with vendors, they're hit by multiple uncoordinated pitches from different parts of the same company.
Government agencies customizing commercial CRM software for federal workflows spend hundreds of thousands to millions of dollars. The result is brittle systems that break when updated and can't adapt to new needs.
Pryzm was built for government business development from the start. It understands appropriations processes, the congressional calendar, and how federal buyers actually work. No customization required.
What Palantir Taught About Building
LaRovere credits Palantir's "forward deployed engineering" model with shaping his approach. The idea: get engineers close to customers so product decisions have proper context.
During COVID, he left holiday break to work directly with CDC officials on pandemic response. Daily calls with the customer. Rapid iteration. The friction of moving fast with a real client beat the comfort of building in isolation.
That experience taught him something about hard problems. If a problem is genuinely difficult, solving it creates a large market, high willingness to pay, and a defensible advantage. Those are the problems worth tackling.
The Hiring Reality
As Pryzm scaled, LaRovere said the biggest surprise was how much the job became about hiring. Keeping a high talent bar, retaining people, and keeping teams aligned is harder than it sounds.
The best engineers are vastly more valuable than the second-best. That gap compounds across the organization. Getting the word out to talented people outside traditional defense circles - engineers who want a mission, not just a job - requires sustained effort.
That's branding work. It's getting founders in front of audiences. It's showing people why the problem matters.
What's Changing on the Government Side
Pentagon officials are now using generative AI and LLM tools. genAI.mil is being rolled out across the Department of Defense. Procurement workflows that ran on spreadsheets and 20-year-old systems are getting AI assistance.
Government agencies are using Pryzm's tools to identify which vendors match their needs. Instead of manually reviewing VC websites to find companies with specific capabilities, they can now run an intelligent workflow that surfaces the right matches.
This is AI for Government in practice - not replacing human judgment, but giving decision-makers better context and faster answers.
The Broader Shift
The administration's focus on getting acquisition right, combined with real security pressures, created space for change. Venture capital is now investing in defense tech at scale. Hardware companies are getting traction. But they all face the same barrier: knowing how to sell to government.
LaRovere sees Pryzm as an enabler for that broader ecosystem. His stated goal: every company in America becomes part of the defense industrial base. That requires tools that make government selling accessible to non-traditional vendors.
In World War II, automobile manufacturers built airplanes. The infrastructure didn't exist to make that easy, but the need was urgent and the will was there. LaRovere thinks we're in a similar moment now.
Your membership also unlocks:
