Building (and Selling) AI for the Government

I have been building AI applications for government for about three years, first at a defense tech startup (Vannevar Labs) and then at xAI. Before that, I was an active-duty submarine officer. All together, in the last 12 years, I have spent only about 30 days not employed by the government or a contractor. Luckily, I’m in one of those periods of unemployment right now, and I’d like to use this opportunity to write candidly about our shadowy industry before I start my next job (spoiler alert, it’s going to be AI for the government).

Defense tech has suddenly become cool from the perspective of investors and startup founders, and many more people are interested in figuring out how they can make money off the government. As I’ll discuss, this is harder than it seems.

The primes

The current state of things is very bad.

When I worked on the crew of PCU New Jersey, a submarine under construction in Newport News Shipyard, I saw the current state of government contracting. As sailors, we were supposed to be the “demanding customer,” responsible for representing the taxpayer and holding the shipyard accountable. In reality, we were anything but. In between buying $5000 wrenches and $80 rolls of duct tape, we watched the ship’s estimated completion date slip three years, mostly due to poor planning and screwups. Despite their best efforts, the Navy has been utterly unable to make progress in shipbuilding, and the goal rate of two submarines a year seems more and more like a fantasy (the best we can do right now is 1.2, even though we pay for two).

Me and some of my former colleagues at USS New Jersey’s commissioning, three years late. By the time the ship was commissioned, I was a civilian!

Shipbuilding is not the only area in which US defense contractors are falling short. While I was on active duty, most Navy commands used a tool called NAVFIT98 to fill out performance reviews for sailors. NAVFIT98 was developed in 1998, hence the name. In 2019, the Navy awarded a $63 million contract to Deloitte1 for IT services in support of HR modernization. Specifically, these services include Performance Evaluation Technologies, enumerated in a follow-on contract. The new performance evaluation tool which was supposed to replace NAVFIT98 was called eNavFit.

eNavFit was a disaster. NAVFIT98 was scheduled for deprecation in 2022, but the Navy delayed this due to “latency issues” with eNavFit. Military Times reported on the delays, citing numerous embarrassing bugs like “entries entered into certain blocks… that later did not match what was on the submitted report.” In 2023, the Navy delayed the NAVFIT98 deprecation again, this time until 2025, and published a fact sheet explaining that they did not yet “have full confidence for Fleet wide implementation” of eNavFit. Finally, in 2025, the Navy fully discontinued eNavFit, and ordered all commands to switch back to NAVFIT98. This is especially embarrassing because eNavFit was a simple CRUD app! You could vibe code it in five minutes.

If you were a startup founder, you would probably see this situation and smell money. Prime contractors get away with hilariously bad performance and are often paid more, not less for it. The market is huge and the incumbents are old-fashioned. Most of all, the government is desperate for change, which is evident from the sweeping procurement modernization orders published by the Trump administration. This situation seems ripe for disruption; all you need to do is be a little better than the primes.

But would-be startup founders should approach the situation with a little humility and ask some obvious questions. If the primes are so bad, why is it rare for startups to supplant them? How have they retained their stranglehold on this market? Why have the contracting offices, so desperate for change, been unable to actually effect it? Could government contracting be harder than it seems?

Don’t get me wrong, I do not believe that the current primes will dominate forever. As a patriot and taxpayer, I want startups to replace them! But I also want these startups to be clear-eyed about the challenges they will face and allocate their effort the right way, so that a greater percentage of them succeed.

Winnerball

Disclaimer: I am not a salesperson.

Sales is the number one thing you need to do correctly to succeed in the government. The primes understand this and they allocate a huge amount of effort to sales. Defense tech companies tend to avoid the term “sales,” and talk euphemistically about “mission success,” “mission development,” “mission management,” “capture,” “GTM,” “BD,” etc., but I think it’s actually better to call sales by name, so that everyone involved understands the goal. I’m going to include customer success and lobbying in my definition of sales, because they primarily involve convincing people to give (or keep giving) you money.

It’s a common mistake for startups to neglect sales....