AI Can Do the Work. Don't You Want to Decide What Good Looks Like?

The first AI workflow I ever built could draft a press release in under a minute. I never let it write a single executive quote.

Not because the technology couldn't produce something convincing; it absolutely could. But after years of working alongside organizational leaders, I knew that sounding like someone and understanding what they would actually say are two very different things.

I could give AI every speech, interview, and published statement a leader had ever made. What I couldn't give it were the countless hours I'd spent observing how they speak, what they emphasize, how they respond to uncertainty, and where their tolerance for risk begins and ends.

So I automated the press releases, and I wrote the quotes myself.

Knowing what to automate (and what to totally own yourself)

I didn’t build this loop because I couldn't write a good press release—but because I'd written so many.

I recognized three things: it was a recurring task, the format was relatively standardized, and I had years of successful examples that established my threshold for distribution.

I never asked AI to decide what made something newsworthy or ready for the wire. I'd already made that judgment and given it the structure, examples, and criteria to produce work that met my standard.

Knowing what to chase (and what to ignore as background noise)

That distinction has shaped how I approach other AI-enabled systems, too.

In the media engagement lead-scoring tool I built, for example, I deliberately treated a reporter's rejection as a signal of engagement. A thoughtful "no," particularly one explaining why a story wasn't right, told us more about a reporter's interests—and gave us more to build on—than an opened email that received no response.

The system could organize and display those signals. But deciding which ones mattered and how much weight each deserved required more than counting responses.

In both cases, the technology could execute the logic. Figuring out what that logic should be was the more interesting part—and the one that required knowing the work deeply before trying to automate it.

Knowing what to make (and why you're making it)

Automating the basics of the press release process didn't make me want to publish more press releases.

I built that loop because it created more room to ask better questions. What story were we actually trying to tell? What evidence supported it? Where did it belong? How could one piece of content reinforce another—creating an ecosystem of relevance rather than a collection of disconnected outputs?

AI has made it easier than ever to produce something. Deciding what's worth producing, what good looks like, and what deserves our attention is the work I want more time to do.