I Built 54 AI Agents. Here's What I Learned About Humans.

Over the last year, I've built 54 AI agents. Content writers, market researchers, financial planners, legal reviewers, design generators, sales assistants, coding copilots, and everything in between.

The technical lessons were interesting. The human lessons were the ones that changed how I think about everything.

Lesson 1: People don't want more capability. They want less confusion.

When we first launched the agent army, we were proud of the number. 54 agents! They can do anything! We put them all in a grid and said "pick one."

Usage was terrible.

Not because the agents didn't work — they worked great. Because 54 choices is not a feature. It's a burden. People would open the grid, feel overwhelmed, and close it.

What worked? Removing the grid entirely and letting Cameron — our main AI — route people to the right agent automatically. "I need to write a blog post" → Cameron calls the content agent. "Help me analyze my competitors" → Cameron calls the market research agent. The user never picks from a list. They just describe what they need.

The lesson: The best technology is invisible. People want outcomes, not options.

Lesson 2: The agents people love most aren't the smartest ones. They're the most opinionated ones.

Our most-used agents aren't the ones with the most capabilities. They're the ones that push back, have a point of view, and tell you what to do instead of asking what you want.

The content agent that says "this headline is weak — here are three better options" gets used 10x more than the one that says "what kind of headline would you like?"

The strategy agent that says "based on your data, you should focus on X and stop doing Y" gets loved. The one that says "what would you like to focus on?" gets ignored.

The lesson: People are drowning in options. They don't want another tool that asks them what to do. They want a partner that tells them what to do — and explains why.

Lesson 3: Trust is built in the first 30 seconds and lost in one bad response.

Users give AI exactly one chance. If the first response is generic, irrelevant, or obviously wrong, they close the chat and never come back. There is no "give it another try."

This is why we obsess over the first message. Cameron's greeting includes your name, your business context, and a specific observation. Not "How can I help you?" but "I see you're working on your marketing strategy — want me to audit your current approach?"

The lesson: Personalization isn't a nice-to-have. It's the difference between a tool people use daily and one they try once.

Lesson 4: The agents that failed taught me more than the ones that succeeded.

Out of 54 agents, about 7 were basically useless. Not because the AI couldn't do the task — because the task itself wasn't something people wanted AI to do.

Our "AI Therapist" agent? Barely used. People don't want to process emotions with AI. Our "Meeting Notes" agent? Low engagement. People would rather just... pay attention in the meeting. Our "AI Receptionist" agent? Hated. People calling a business want a human, full stop.

The lesson: Just because AI can do something doesn't mean it should. The best AI handles the work people don't want to do. The worst AI replaces the work people actually enjoy.

Lesson 5: AI makes good teams great and bad teams worse.

This was the most uncomfortable discovery. Teams that already communicated well, had clear processes, and trusted each other became dramatically more productive with AI. Teams with broken communication, unclear roles, and internal politics? AI amplified the dysfunction.

AI doesn't fix organizational problems. It accelerates whatever already exists. Give a high-functioning team AI tools and they'll 10x their output. Give a dysfunctional team AI tools and they'll 10x their chaos.

The lesson: Before implementing AI, fix the humans. Clear roles. Clear processes. Clear communication. Then add AI as the accelerant.

The meta-lesson

After 54 agents and thousands of user interactions, I believe this: AI is not a technology product. It's a human product delivered through technology.

The companies that win with AI won't be the ones with the best models or the most features. They'll be the ones that most deeply understand what humans actually need — and build AI that meets them there.

That's what we're trying to do with Waymaker. Not build the most powerful AI. Build the most useful one.