Enterprise AI is easy to demo and hard to ship. I've done the second part.

At Kore.ai, the AI for Work portfolio put agents for HR, IT and recruiting in Slack, Teams and the inbox, where employees already work. At ADP, I led the first version of Workforce Now's HR and payroll integration, now a $1B+ revenue product, and I worked every segment from small business to enterprise. The constraint that shaped all of it: when it breaks, real people don't get paid.

Every product I've worked on has been about the people who actually do the work. The new hire lost in onboarding on day one. The recruiter with more roles than hours. The HR person fielding the same benefits question for the tenth time.

The hardest part of enterprise software isn't building it. It's everything that has to change around it.

My work lives in the messy middle, where the product has to serve both the employee using it and the leader buying it. Those two audiences want different things, and the interesting work is in refusing to choose between them.

To stay hands-on, I built a working employee assistant to test the product decisions enterprise teams face when moving from AI demo to production: orchestration, permissions, grounded answers, human approval, system actions and measurable evaluations. Policy search, parental leave and time off are live, and it's built in the open. Claude Code accelerated the implementation; the product decisions are the point. See the portfolio.