How to Start Building Your Own AI Agents (No Master Plan Needed)

When I share my screen during workshops and community sessions, people go quiet for a second.

It’s the Lindy account. 40+ agents. Folders inside folders. Workflows that run on schedules, on triggers, on calendar events, on email arrivals.

Someone always says: “How long did it take you to design all of this?”

And the honest answer is: I didn’t design it.

There Was No Master Plan

I don’t have a document somewhere that maps out my complete AI system. I never sat down and said, “okay, I’m going to build a digital executive assistant suite covering email, calendar, content, outreach, and follow-up.”

What I did have was an email inbox problem. In early 2025, I was spending way too much time each day just sorting through emails — reading, categorizing, deciding what needed a response and what didn’t. So I spent about an hour building an agent to handle the triage. That was the entire plan.

A few weeks later, I kept forgetting to follow up on emails I’d sent. The recipient would go quiet and I’d miss it. So I built a follow-up reminder agent.

Then I kept going into meetings without enough prep time to remember who I was talking to or what we’d discussed before. Meeting prep agent.

Then I was spending too long writing intro emails — researching the people, finding talking points, drafting something personalized. Email drafting agent.

Forty times. One specific frustration at a time. Across roughly a year.

Why This Approach Works

The “design a full system first” instinct is actually backwards for AI automation.

The problem is that you don’t know what you need until you’ve lived with a partial solution for a while. If I had tried to design a 40-agent suite on day one, half of what I built would be wrong. I’d have agents solving problems I don’t actually have, and missing the things that actually cost me time.

Building in response to friction is different. You experience something annoying this week. You build the solution this week. You know exactly what problem you’re solving because you’re living in it right now.

At Asian Efficiency, we call this the difference between real work and fake work. Designing an AI system you’ll never build is fake work — it creates motion without movement. Building one agent that solves the friction you’re feeling today is real work.

What This Looks Like in Practice

When I show people my setup, they often ask which agent to build first. My answer is always the same: which part of your day is most annoying right now?

Not “which sounds most impressive.” Not “which is most strategically important.” Just: what’s costing you time, and is making you feel like you’re falling behind?

For most people, it’s email. Or meeting prep. Or the small administrative back-and-forth that takes 20 minutes to do manually every time. Start there.

The first agent doesn’t have to be elegant. Mine certainly wasn’t. It just has to solve the thing that’s actually bothering you.

The System Emerges

The weird thing about building one at a time is that you end up with a system anyway.

At some point I looked at my Lindy account and realized that I had email handling, meeting intelligence, content production, scheduling, follow-ups, outreach — all covered. I’d built a real system. Just backwards, in the sense that I built the pieces first and the pattern emerged later.

This is how most useful systems get built. Not by architects designing everything upfront, but by people solving problems until they look back and realize they’ve covered a lot of ground.

If you’re waiting to start because you don’t have a complete plan yet: you’re never going to have a complete plan. Pick the one thing that’s annoying you most this week. Build the agent for that. See what you notice after living with it for two weeks.

That’s how you get to 40.

P.S. The Two Hour Workday course walks you through this process — starting with the highest-impact agents, building them step by step, and expanding your system over time.