How I Use AI

Treat AI like an intern.

I use AI in my design and project management work, and I review what it produces before it goes anywhere.

01 · In practice

How I work with AI.

The first 75% keeps getting easier.

Working with AI mostly feels solo, and calling it a "team" flattens something real. That framing is from a mentor of mine, who also named the 75% trap: the first 75% keeps getting easier, and the last 25% is where the work lives. My own projects bear it out. AI produces a rough, working draft embarrassingly fast, and getting from that draft to something I'd put my name on is still my job.

Reviewed twice before anything went in.

I had Claude Code rebuild this site from a design handoff and check its work twice before anything went live: once in a browser at desktop width and at 320 px, and once line by line against my writing rules. The browser check found tables that stretched the mobile layout. The writing check found copy that went beyond the facts I had given Claude, including a line that explained what a Jira board's side states were for. I never said that. Both were fixed before launch.

Prototypes fast, reviewed before they leave.

Agile Defense works the same way. We build concept prototypes fast, sometimes in days, because the fastest way to find out if an idea holds up is to put it in front of someone who knows the problem firsthand. AI gets the prototype into someone's hands quickly. My job (and my team's) is reviewing what comes out before it goes anywhere near a client, the same judgment, applied earlier.

Treat AI like an intern.

My rule is to treat AI like an intern: nobody at Agile Defense puts their name on what it gives on the first pass. I fact-check it, make it cite its sources when something doesn't sit right, and push back. (After pushback, AI sometimes swings to the opposite answer just to appease you, so the skill is knowing enough to recognize a good answer.)

Give the tool a skill that flags patterns.

One example from our program management course: the team noticed recurring patterns in how the scripts were written, and we gave our internal AI tool a skill that flags them for us or the subject matter experts to verify. It cut editing time by about half, because patterns that used to need their own review round get caught automatically.

Have a messy problem? Let's make it simple.

Get in touch