How I think about AI

AI can take on a lot of repetitive work, as long as reaching a person stays easy. Beyond that, I use it to help people do their jobs better.

Here's where I've landed:

• Individual workflows and shared standards are different problems. AI can be valuable for individual workflows. It shouldn't stand in for places where we need consistency, because it can't easily be maintained centrally. If something has to be the same for everyone, it belongs where one owner can change it once.
• Context does most of the work. Good context can identify whether something is likely a bug or comes from a user's browser or configuration. Depending on what the company uses, it can also pull in details like health scores and session recordings. Our chatbot at Workable answered 60% of tier-1 questions because it drew on 700+ articles, videos, and guides and worked with Zendesk, Confluence, and Jira. The articles and integrations did most of that work.
• Judge it by the outcome. If customers still need help, or leave more frustrated, it failed, whatever the usage report says.
• Watch for the confident answer. Grading AI model outputs on contract taught me the usual failure isn't obvious nonsense. It's an answer that sounds right, or one that's right for one type of account and wrong for another. Building Hey Vitae, tracing every AI call showed me where answers hold up and where they quietly fail.
• Adoption is a people problem. When we rolled out an agent-facing tool, it took a peer champion, not just training.
• Someone decides where it doesn't belong. AI shouldn't be the one resolving cancellations, purchases, billing disputes, or a request for a person. It can still help by routing and capturing context, so the person who picks it up starts ahead.

I build with these tools too, but what I bring is the judgment about what's worth building and how it should feel to use.

Here's what that looks like in practice: The AI guardrails I use.