AI in practice · Dawson Russell

The work between “can you?” and “done.”

The most useful AI moment I’ve had lately wasn’t a better answer. It was getting the work done.

A client needed help with a Klaviyo issue and an update to their website. Normally, that means opening the message, working out what needs to change, moving between tools, doing the work, and writing back.

This time, I asked my AI assistant to read the request, handle the work, and respond. The exchange ended with a heart reaction and a thank-you.

That’s the shift I’m interested in. We’ve spent a lot of time using AI to produce another document for us to read. Browser control brings it into the places where the work actually happens.

This is one experience, not proof that every client request should run unattended. The details matter: what the assistant can access, what it’s allowed to change, and how you know the result is right.

For client systems, my takeaway is simple: make the task specific, preserve the things that must not change, and check the outcome. In email marketing, that includes respecting legitimate unsubscribes. “Fix the list” should never mean undoing someone’s choice.

I want to keep testing the space between receiving a request and delivering the result. That’s where a lot of an agency’s day goes. And that’s where these tools are starting to get interesting.