We need to preserve the humanity in government contracting in the age of AI everything

In 2007, Jo Hamilton pleaded guilty to 14 counts of false accounting and was required to pay £36,000 to recoup money she had apparently stolen in her time as sub-postmistress of the South Warnborough Post Office. Except here’s the problem: She didn’t steal anything.

Six years earlier, in a sleepy town in the south of England, Hamilton had purchased and taken over management of a small community co-op where the town’s post office was co-located. By 2003, the sub-postmaster had stepped down — so she filled that role too. She had almost no training, but the pace of business was slow enough that it didn't seem to matter. That same year, a new chip-and-pin machine was installed and the post office's accounting systems went digital.

By December, Hamilton found her first discrepancy when the Horizon software the system used showed a shortfall in the amount of £2,082. As most people would, she called the help desk to figure out the discrepancy, because surely her math couldn’t have been off by that much. It turns out it wasn’t. The help desk assisted her in figuring out the discrepancy was more than £4,000! Over the next nearly three years, this back-and-forth would continue, and despite her protests and her pleas for help from the support systems designed around the software, no help would come.  

American readers are probably less familiar with Jo and the Post Office Horizon scandal in the United Kingdom, but she became well known there as an advocate for other victims of the software’s accounting malfunctions. Looking back at her story nearly 20 years later, it feels like a bright flashing warning sign about what happens when organizations stop seeing software as a tool to help people and start treating it as something to defer to. The Post Office never paused to ask whether the system might be wrong. It treated the software's output as fact, and treated the hundreds of people questioning it as the problem. 

In the age of AI, when it is easier than ever to scale the speed and complexity of a system, this is the failure mode to watch for. Confusing an output for a judgment.

I have been in the government contracting and technology development space for more than two decades helping connect people to opportunities, and what I am seeing leaves me with an unsettling feeling that we are headed for a future where AI manages so much of the contracting process that it loses the humanity that makes it valuable in the first place.  

It’s a little too simple to say: “Government needs a widget, company makes a widget.” End of transaction. 

If you’ve been in the market for more than five minutes, you know the actual system is way more complex and more nuanced than that. Yes, the government has requirements, but those requirements exist inside a moving context: real-time evolving needs of the customer, capabilities already in use, wants constrained by resources, internal politics nobody writes down. The company side is just as complicated. What looks on paper like a clean match is almost always a layered conversation about needs, understanding, limitations, and competing motivations. These are human factors, and I have seen them in action. 

There is a story from my career I think about a lot. I worked with a small manufacturer in the Midwest who had built something genuinely useful: a purpose-engineered component, made on their own floor, that solved a real problem for a specific defense program. They had been trying to get in front of the right person for eight months. They had sent the capability brief. They had filled out the form on the website. They had asked for introductions through every channel they could think of, and nothing was moving. 

I had met the founder months earlier, and I knew the program office people from a separate set of relationships. The thing I knew that nobody else in the equation did was that those two worlds had never actually been in the same room together. The need was there. The solution was there. The understanding was not. So at an industry day later that spring, I made sure to be there with the founder at the spillover happy hour afterward where a few of the program office people ended up. I made the introduction, grabbed my Coke Zero, and stepped out of the way. He didn’t pitch. He sat down and talked about what he was seeing on the manufacturing side — the lead-time problems, the specific bottleneck his component addressed. Twenty minutes in, one of the program office people said, "Wait, you make that?"

That conversation moved more in 20 minutes than eight months of formal outreach. 

Here’s my fear: A future world where contracting is fully an AI sport looks a lot more like that manufacturer sending capability briefs into the void than it does sitting down in the room and talking. I have looked at many of the platforms promising to fix that with a button. 

Click here, get a 90 percent solution for a proposal. 

The first problem with most of those tools is that they assume a quality result can come from just knowing the solicitation and the company's past award history. That is basically just falling into the “government needs a widget, company makes a widget” trap. Contracting is nuanced and personal. Solicitation sections are complex because they are designed to show the evaluator that the user truly understands the problem and is prepared (and equipped) to solve it. Even if simple automation can clear that hurdle, do we all benefit from a world when those building the solutions don’t understand what they even submitted for?

The bigger issue is what happens at the system level. As more people use AI tools for submissions, the pace goes up and the diversity of inputs goes down. My good friend and colleague Jerry Ramey, said in an article for Federal News Network last September, this landscape creates a “situation where the best ideas are not necessarily positioned to be chosen. Instead, evaluators are left to choose from nearly indistinguishable submissions, meaning luck is as much a factor as persuasion or merit.” 

So the question is how to use AI well at a system level. This is an architectural question, and the reason I opened this article with the story of Jo Hamilton is because we already have the answer. Any system that scales like this needs to center the person, because it is people who actually do the work.

AI should speed searching for opportunities; a person should decide which solicitations to pursue. AI should accelerate market research, but humans should define what a real capability gap looks like. AI should gather and arrange requirements; a person should ensure inputs reflect reality. AI should help draft proposal language, but humans should make the argument credible.

I am not the only consultant or advisor with a story like the manufacturer's. Most of us in this space have a version. We know the X factor of the human side, because we have watched it convert eight months of dead silence into a 20-minute conversation that actually moves something.

We also know that with hundreds of billions of dollars of non-dilutive funding moving through tens of thousands of opportunities, automation is not optional. The pace of the system needs to go up. The government wants it to go up. The contractors need it to go up.

The challenge will be navigating that in such a way that we maintain the reason we’re doing all of this in the first place.

AI belongs in this work. Truly, it does. The question is whether we build systems that help people or ones that treat people as an afterthought.

I know which one I want to build.