← All case studies

AI Strategy · From near zero to 75%+ of engineers, inside the guardrails

Making AI Safe and Practical Across Engineering

Drove AI strategy and governance for engineering: usage policy, model access and quotas, risk and PHI reviews, enterprise platforms, and vendor strategy. Approved AI use grew from near zero to 75%+ of engineers, and the work fed the company's HITRUST AI certification.

The Project

In 2024, approved AI use in our engineering organization was close to zero. People were curious, a few were experimenting on their own, and nobody had a safe, sanctioned way to use these tools on real work at a healthcare company. Today at least 75% of our engineers actively use approved AI tools. I treated getting there as one project: make the safe way to use AI the practical way to use it.

Making the governed path the easy path

I formed an AI-focused strike team to do the early hands-on work, then helped turn what we learned into best practices and governance. I own the usage policy and engineering standards, including what data can go into which tools and how PHI is handled. I run the approval path for model access and quotas, so teams get capacity through a governed path instead of around it, and I administer the enterprise AI platforms we rely on. I also helped form the AI Adoption Committee that now governs our AI adoption, and I work closely with our development teams so the infrastructure side of their AI needs is covered.

Safe for production, ready for audit

We build software for healthcare, so safe has a concrete meaning here. It means clear production standards for where AI can and can’t act, security and privacy reviews of the tools we bring in, and work that holds up under the same scrutiny as everything else we ship. That work fed into the company’s HITRUST AI certification.

Spending like it matters

Strategy includes the bill. When our enterprise AI renewal came up, I negotiated the vendor’s top pricing tier, with no seat fees and a lower usage rate, at half the commitment they normally required for it. When adding a second AI platform came up, I compared the cost of real usage on each and recommended against it: it would have added a large annual contract and bought far less usage.

Meeting engineers where they were

My first adoption timeline was unrealistic. I had been using AI for years and was far ahead of most of our engineers, and ideas I was excited about, like AI coworkers, were not near term for them. So I changed my approach: start from the work people already do and add to it slowly and strategically. I shared short posts about useful things I had done with AI and exactly how to do them, and I talked about AI in 1:1s. I started an AI community of practice, and the AI team I created grew into an AI chapter of champions who help their own teams. I mentor engineers through their first serious AI work and use surveys to see where adoption is working and where it is stuck. Some of our most hesitant engineers became some of our best adopters, because they learned it for themselves.

What changed my mind on risk

Early on I pushed hard. I wanted more tools and a third AI provider, and I was quick to wave off risk concerns about the tools we already had. I learned that the business did not need three providers, and that an agent running with broad permissions next to a browser with full network access is a real risk on an engineer’s machine. I come from security engineering, so I stepped back, pulled back the edgier work, and went deep on security practices and governance.

What is next: Zero Trust Agents

My current work is an idea I call Zero Trust Agents: give agents more freedom to do real work, but isolate them so they pose no risk to the engineer’s machine. Capabilities get added one at a time, like Lego, on a 12 to 18 month plan. Right now I am laying the groundwork and having the conversations that will make that plan succeed, slowly and steadily. It pairs with the platform I am architecting for AI-built internal tools.

Wrap Up

The measure I care about is that AI use across engineering looks like engineering. Judgment stays with the engineer, accountability stays with the team, and the policies, access paths, and platforms make the responsible path the default one. That’s how AI went from near zero to routine practice without anything having to be walked back.