Every time I teach an AI workshop here in Austin, someone asks the same question within the first hour.
“So when do we get to the agent part?”
And I understand why. That’s what all the headlines are about. Autonomous agents. AI that thinks for itself. The future where you just hand something to a robot and it figures out the rest.
But here’s what I’ve noticed after a year of building these systems: the “agent” stuff is actually a small fraction of what’s saving people the most time. Most of the real hours-per-week wins come from something much less glamorous — and much easier to build.
The Three Levels of AI Automation
When I first sit down with someone to help them figure out where to start with AI, I always walk through the same framework. I call it the AI Fluency Levels, and it’s basically a maturity map for how people use AI in their work.
Level 1: AI Assisted
This is where most people are right now. You open ChatGPT, type a question, get an answer. You’re in the loop the whole time. The AI is helping you think or draft or organize, but you’re steering every step.
This is not a bad place to be. People who use AI-assisted tools consistently tend to save somewhere around 5-10 hours a week. That’s real. But the AI isn’t doing anything without you.
Level 2: AI Workflows
This is where it gets interesting. A workflow runs in the background. You set it up once, and then it just… goes. Without you.
Email comes in and gets triaged and categorized automatically. A meeting ends and notes get summarized and sent to your team. A new inquiry arrives and a first-draft response is ready in your drafts folder before you even open your inbox.
These workflows are predictable. They go from A to Z in the same way every time. And that predictability is actually what makes them so reliable and so useful.
This is where 10-30 hours a week back lives. Not the flashy autonomous agent stuff. Just: the thing that was eating your morning is now handled before you wake up.
Level 3: AI Agents
Actual agents are different. An agent doesn’t follow a fixed path — it makes its own decisions to reach a goal. You give it an objective, and it figures out the steps. It can branch. It can adapt. It’s like handing a task to a real person and saying “go handle this” instead of writing a step-by-step procedure.
This is powerful. This is also rare in practice.
What My Own Setup Actually Looks Like
I have about 45-50 different things running in Lindy right now. When people see the full list, they usually assume most of it must be agentic. Super sophisticated. AI making decisions everywhere.
But if I actually count it up: five to eight of them are real agents. Maybe.
The rest are workflows. Predictable, sequential, runs in the background, gets stuff done.
And those workflows are doing most of the actual work. They’re the ones clearing an hour out of my morning. Handling the follow-up I’d have forgotten. Prepping me for a meeting I didn’t have time to prep for.
The Mistake That Slows People Down
The most common thing I see when people start building AI automation is that they try to jump straight to Level 3. They want to build something impressive and autonomous before they’ve built anything useful and reliable.
The problem is that agents are harder to build well. They’re less predictable. They require more debugging. And if you’re still figuring out what problems you actually want to solve, agent complexity makes everything slower.
Starting with Level 2 workflows is counterintuitive but much more effective. You build something simple, it runs reliably, and you immediately get time back. That quick win makes the next build easier. You start noticing other friction. You build that one. Repeat.
By the time you’re actually ready to build an agent, you have 15 workflows that have taught you exactly where the automation-ready problems are in your work.
Where to Start
Pick the thing that’s happening in your work every single day and costing you 30-60 minutes. Not something quarterly. Not something complicated. The thing that’s eating your morning right now.
For most people, that’s email. Or meeting prep. Or the manual report that gets generated the same way every week.
That’s a Level 2 workflow. And it’s probably buildable in a couple of hours.
You don’t need agents to make AI useful. You need one reliable workflow that handles the friction you feel most.
Start there. The agent stuff can come later, once you know exactly what problem it needs to solve.
P.S. The Two Hour Workday program walks through this framework in detail — starting with the highest-leverage Level 2 workflows and building up from there.
Recommended for you
Want the full system? 25X is the flagship productivity system we teach.
Explore 25X →