What I Learned Building 30+ AI Agents for Marketing Teams in One Year

In the past year I built 30+ agents for personal use, freelance clients, and the team at Adwise, a 150-person agency.

Here are a few personal learnings. Focused more on mindset challenges than technology.

FOMO is your enemy

I fell into this trap way too often and I'm not proud of it.

Working at an agency with 150 people, you need to approach things differently. A bit slower. With more thought. Access to data is easier for 1 person than for 150.

But then you look around on social media and see all those cool things. You constantly feel like you're falling behind.

In a way you are — but you're building a hotel, not pitching a tent. That feeling is strong though. You don't want to fall behind. You want to test all the new things. There's client pressure. "Are you guys using X and Y?"

So I was weak and gave into this feeling.

Due to this I sometimes sped up and took shortcuts.

• Positive: We learned a lot.
• Negative: We had systems scattered everywhere that didn't work at scale.

I ignored something important: feedback isn't linear.

When you get bugs in 1 tool, it's easy to manage. But bugs and feedback from 150 people about 14 tools? That's not 14x150. It's chaos and a big bucket of stress for you :D

Try to ignore the noise. There will be a new tool every week, but you can't move at solo-builder speed. Accept that.

Separate innovation from the main thing

You need room for innovation while building your company-wide AI system.

Ideally you want one main AI tool where all your context, data, and tools live. But building that takes time. It's slow.

In the meantime, you still need to test tools, frameworks, and models.

So leave room for people to experiment and play. The learnings we had from silly side projects and one-time experiments had an enormous impact on how we now design our main system.

Give 1-2 innovative people room to play (FAFO). Let them experiment, learn, and feed those insights back into the main tool.

This also helps to show your clients that "Yes, look we already tested this in this way." So at least you lower the client pressure.

The hotel analogy

Don't underestimate the pressure of FOMO. It took a toll on me last year.

Always remember — you are building a hotel, not a tent. So security, entrance, and other infrastructure has to be on the same level.

A solo builder can pitch a tent in an afternoon. Spin up a quick agent, test it, throw it away if it doesn't work. No big deal.

But when you're building for a team of 150? Every decision compounds. Every shortcut creates technical debt that 150 people inherit. Every "quick test" becomes a system that someone depends on.

The hotel takes longer to build. But when it's done, it houses everyone properly.

What I'd do differently

If I started today with everything I know now:

1. One main system first. Get the foundation right before adding anything else.
2. A sandbox for experimentation. Completely separate from production. Let people play there.
3. Slower rollouts. Test with 5 people before rolling out to 150.
4. Document everything. Every decision, every failure, every lesson. Your future self will thank you.
5. Say no more often. Not every shiny tool deserves your team's attention.

The technology is the easy part. The hard part is managing the humans — including yourself.