On AI for Work.
AI didn't solve enterprise software's hardest problem. Employees still don't have one place to go.
Despite a decade of new technology and a new layer of AI on top, the friction is the same as it's always been: integration, ease of use, and giving employees one straight answer about a hire, an onboarding step, a benefits question, an IT ticket. The technology changed. The problem didn't. Here's what I've learned about closing the gap between a demo and a habit.
Live where the work happens.
A feature lives in your roadmap. A habit lives in Slack.
The agents I've shipped live in Slack and Teams, not in their own apps. Employees shouldn't have to context-switch to get help. The harder integration work, sitting inside the channels people are already in and surfacing at the right moment, is what makes the difference between a feature and a habit.
Trust beats capability.
You don't win enterprise AI by being the smartest. You win by being the safest to depend on.
An agent that confidently does the wrong thing is worse than one that knows when to ask. I obsess over the seams: explainable scoring, smart escalation, clear handoffs, observable behavior. Capability gets the deal signed. Trust is what gets it renewed.
Two customers, always.
Most products optimize for the buyer and lose the user. The good ones refuse to choose.
Enterprise AI has two customers, and they want different things. The IT or HR leader buying it wants control, integration, and predictable ROI. The employee using it wants speed and a system that doesn't get in the way. Most products optimize for the buyer and lose the user. The good ones serve both.
The plumbing is the product.
Models are commodity. Integration is the product.
The hardest part of shipping an enterprise AI agent isn't the model. It's the integration. Workday. SAP. ADP. Oracle. SSO with Okta, Azure AD, Google. Approval workflows. Audit logs. Permissions. The agent is the visible 10%. The plumbing is what makes it real.
Demo-to-habit is the only metric that matters.
Adoption tells you they tried it. Retention tells you they need it.
I track adoption obsessively, but I look harder at retention and depth. The interesting question isn't "did 80% try it once?" It's "what fraction came back next week, and what did they ask the second time?" That's the only honest measure of whether the product is becoming part of how work gets done.
If any of this rings true and AI in your workforce is yours to solve, a short call is the best way to know if there's a fit.
Let's talk →