Stop Theorizing, Start Building

I don't code. I also just deployed my own AI agent application.

These two statements shouldn't exist in the same sentence, alas here we are.

I constantly feel behind. It gives me and my therapist a lot to talk about and I use it for good like experimenting and being willing to fail in order to learn. I'm literally building AI products at work. Not just throwing AI on top of products, building from the ground up. I'm also taking an AI evaluation class, working alongside engineers and a cast of very smart people. It's easy to believe my brain that I'm behind. Then there are times I'm reminded that I'm not.

Like when an engineer recently asked me about Claude Projects because I posted about a small UX improvement recently made on Claude. Me. The person who still has to look up JSON syntax every time I need to format a request.

Most AI agent tutorials assume you're already technical and naturally think in code architecture and deployment pipelines. They skip the part where you're a PM who knows what these things do but has never actually built one from scratch.

Here's the thing. Product managers need to get hands-on with AI tools, not just theorize about them. We can't keep asking "Can we add AI to this?" without understanding what it actually means. I mean we could but eww, gross, let's not.

After months of building AI products at work, I kept hitting the same wall: evaluating whether our AI features were actually good. The gap between 'it works' and 'it works well' felt huge, and I knew I needed better frameworks for this. Taking an AI evaluation class seemed like the obvious way to upskill. What I didn't expect was that building my own agent from scratch help with the imposter syndrome and feeling that I'm falling behind.

The AI evaluation class provided structure and accountability to build something from start to finish. I already have some of the basic skills. I use Cursor and Claude Code, I know how to handle API keys, I've cloned repos before. But having a clear framework pushed me beyond experimenting into actually shipping something is what I needed to level up.

Starting with the course's template provided, I transformed it into an Accidental PM Story Generator that uses GenAI for the stories. The process taught me real AI development constraints. Not because I was learning to code, but because I was learning to build complete products so I could get better at evaluating them.

The "Accidental" Advantage

My improv and theater background prepared me for this in ways I didn't expect:

Comfort with improvisation and uncertainty: In improv, you start with nothing and build something. AI development works the same way. You begin with a vague idea and figure it out as you go.

"Yes, and..." approach to technical learning: When I hit a wall with code, I didn't give up. I said "yes" to the challenge and found another way. "Yes, this is confusing, and I'll ask for help." "Yes, this broke, and I'll figure out why."

Pattern recognition from directing different personalities: Managing AI models feels like directing performers. You give them direction, they interpret it, and sometimes they surprise you with something brilliant you never expected.

My product manager instincts kicked in too:

User research mindset applied to context engineering: I approached prompts like user interviews. What does the user really want? How can I ask better questions to get better answers?

Stakeholder management skills for technical learning: I learned to ask the right questions, find the right people, and build relationships with those who could help me understand what I was missing.

Systems thinking from library science background: I instinctively organized information, created taxonomies, and understood how different pieces connected. All those years of organizing card catalogs translated directly to organizing code and data.

The Bigger Picture

More PMs should do this. AI is becoming core product functionality, not just a feature. The gap between AI hype and AI reality is huge, and the only way to bridge it is to get your hands dirty.

The meta-lesson is that the best way to learn something technical isn't to read about it or take a class. It's to build something. Start small, break things, ask questions, and keep going.

I constantly feel behind and then I realized that I'm doing what I always do: building my own path. I'm finding ways to learn and problems I want to solve.

PMs just start with one AI tool. Build something simple. Don't worry about being perfect. Worry about being curious. The confidence that comes from doing something you thought you couldn't do is irreplaceable. The best way to product-manage AI isn't to stay in your lane. It's to build your own AI tools and understand them from the inside out. Because at the end of the day, we're all just making it up as we go.