I wrote thefirst commitmyself

Built first, asked later.

Capital One needed a way to build with agents without every team starting from zero. Nobody asked me to fix that. I wrote the first commit anyway, took it to a few teams, and let the results make the argument.

The platform runs the product development lifecycle with agents: product, tech, and design artifacts, fast, to a quality bar, with governance that is not optional, shaped so the next agent can pick them up. One rule sits above it all. Humans in the lead, not in the loop. People decide. Agents produce.

It worked. The platform is now a fully funded product with about 2,000 users, more than 50 contributors, and a dedicated team of 8 engineers. I established the cross-org executive steering committee that governs it. Teams that build on it reach market about 60% faster than they did before.

Agents fail quietly.

An agent that is wrong looks a lot like an agent that is right. So I designed the evaluation frameworks that decide which model handles each task and prove the choice with numbers. They cut our model spend about 70% without giving up quality.

Agents also forget.

Most hallucinations are missing context wearing a confident face. I built a context engine on a knowledge graph with repo hooks, so agents carry the context they need instead of guessing at it. It reduced agent hallucinations 57% against baseline.

Not what agents can do. What they may do.

I invented an agentic AI experience layer, patents pending, that distinguishes human users from automated agents and governs what each can access. That includes conditional access to high-risk actions like wire transfers. An agent should earn the right to act, case by case.

I have filed 40+ patents in agentic AI and identity. The warrant in the corner of this site is the same idea, playing itself out on you.

It refused to die.

The product behind those patents kept ending up on the cut list. I kept it alive each time, without authority, until the numbers made the argument for me.

It now carries eight-figure projected savings on a single use case, and it helps deter social engineering scams. The people it protects will never know it exists. That is the point.

Scale means other people can do it too.

I serve on our CPO's panel for AI literacy. My first workshop reached 1,500 people and scored a 78.2 NPS. Those trainings are now deployed centrally across the entire product organization.

Eight years of product first.

Fintech. IoT at Whirlpool. Enterprise platforms at the University of Rochester. Different industries, same job. Find the real problem, ship the fix, measure whether it worked.

It was never just the tech.

The RAPID Framework came out of my Georgetown research: 44 real enterprise AI deployments, studied for what made them work. The answer was rarely the model. Deployments succeed when they meet the organization where it is and keep a human element exactly where humans are required.

That research shaped the platform. The tech will always change. A sticky product brings people where they need to go.