How AI Changed the Way I Run My Business

I'm an instructional designer. I've been one for more than 20 years. I am not a developer, and I still don't write code.

A year ago, I used AI quite a bit, mostly to generate content. Emails, course content, first drafts of just about anything. I'd ask, get something back, copy it somewhere, and move on. It was helpful, but I was still the one doing all the running around.

Today I have a small crew of AI agents that work while I sleep. I wake up to an inbox that's already sorted, a list of follow-ups from yesterday's calls, and a short note about how I showed up as a leader the day before. That's a pretty big change for a woman who runs a small business and wears every hat in the building.

I want to tell you what that looks like, what went wrong along the way (plenty), and what made it possible.

What It Looks Like Now

My email and follow-ups. Every night, an agent goes through both of my inboxes. It sorts what needs me from what doesn't and writes reply drafts for me to review in the morning. It never sends anything. I read every draft, fix what needs fixing, and hit send myself. Another agent listens to my call recordings and pulls out the things I said I'd do, so they land on my to-do board and don't live in my head. If you have ADHD like I do, you know what a gift that is. It also scans my inbox for important attachments and saves them to the right folder, so I'm not digging through email for them later.

My time. I used to rebuild my week from memory every Friday. Now my hours get pulled together from my calendar, my recordings, and what I worked on, and I review them. My team gets a reminder when their hours are due.

My invoicing, start to finish. Each week, an agent cleans up everyone's time entries and gets them ready to bill. On Mondays, another one prepares my client invoices in QuickBooks. I review every invoice and send it myself. On the other side, it finds my contractors' invoices in my email, saves them to the right folder, and checks them against the hours they logged. If something doesn't match, it sends a short note I wrote ahead of time asking about it. It never pays anyone. I still do that part. And every morning I get one short report that tells me if anything didn't run or the numbers don't agree.

Finding new work. This is the one I care about most, because business development is the first thing to slide when client work gets busy. Every night an agent looks for new contracts and RFPs that fit Taproot. For the good ones, it puts together a tailored resume and cover letter for me to review. We built a proposal system too, so a proposal starts from our real experience and our real numbers.

Coaching. Every morning I get a short note based on my own emails and messages from the day before. It tells me what's working and gives me one thing to try. Some mornings it's a pat on the back. Some mornings it's a little uncomfortable. Both are useful.

Our brand and website. We just rebuilt the Taproot website and moved it off our old platform. I made the decisions about how it should look and what it should say. AI handled the technical side - the build, the redirects, moving all of my old blog posts over - so I didn't have to learn any of it. You're reading the result right now.

What It's Worth

Subscriptions I didn't need. Managing products in GoDaddy overwhelms me, and I never knew what I was paying for. I love setting AI loose on complicated sites I can't get to work, so I pointed it at my account. Between GoDaddy and one other service, it found more than $1,600 a year in renewals I didn't need.

Hours I used to miss. My AI time tracker catches the work I used to forget to log, like short calls and quick emails. One week I saw my hours go up by ten. I estimate it adds up to $5,000 to $10,000 a year in billable time.

What Went Wrong

I'd be lying if I told you this was easy.

My biggest problem was agents that failed without telling me. I'd set something up to run overnight, assume it was running, and find out days later that it wasn't. It just... didn't happen.

Here's my favorite example. I did a time audit a few weeks ago and found I'd spent more than seven hours in two weeks working job postings by hand. I already had agents built to do that exact work. I just didn't trust that they were running, so I did it myself.

I also lost a lot of hours setting things up that didn't hold. I'd patch one problem, then another, then another, and end up with something so fragile I was afraid to touch it.

So I made rules, and they've saved me:

• Every scheduled agent sends me a message when it runs. If I don't hear from it, something broke.
• Agents write drafts and notes. Anything that goes to a client waits for me.
• Nothing gets deleted without asking me first.
• If it fails twice, we stop and find a simpler way.

What Made It Possible

I didn't figure this out on my own. I learned much of this in Leland's AI Builder Program.

I finished Level 1 in July and Level 2 in September, and I start Level 3 next week.

There are a lot of AI courses out there that show you how to set up one thing. You follow the steps, you get the one thing, and the minute you want something different you're stuck.

Leland didn't teach me that way. It gave me a foundation I can take in any direction. I learned what good practice looks like and how to apply it, so when I wanted an agent for email, then one for billing, then one for finding work, I didn't need a new tutorial each time. I already knew how to think about it. As an instructional designer, I notice when something is taught well, and this is taught well.

The cohort helped too. Learning next to other people who are building real things for their real jobs kept me going on the weeks when nothing worked. And somewhere in there I stopped saying "I'm not technical enough for this." I'm still not a developer. I just don't need to be one.

At Taproot we say we believe in what's possible and we spark curiosity. This past year I got to be on the learner's side of that. When you're working with AI, you can do just about anything. You just have to be curious and willing to try. And then keep learning...

Things You Can Use This Week

You don't need a program to start. Here's what I'd tell a friend.

Give it one job. My worst agents were the ones I asked to do five things. My best ones do one thing and stop.

Build the smallest version first. Run it once, by hand, and look at what comes out. Don't put anything on a schedule until you've seen it work.

Keep it to drafts. Let AI get you most of the way there, and keep the send button for yourself. Your clients and your team can tell the difference.

Make it report back. If something runs when you're not watching, it should tell you it ran. Silence should mean something's wrong.

Know when to skip it. Some things are a five-minute checklist. Not everything needs an agent.

Stay curious. Most of what I've built started with "I wonder if it could..." Try it. The worst that happens is you learn what doesn't work, and that's how we discover better ways.

If you want a place to start, pick one task you redo every week and try something like this:

I do [task] every week and it takes me about [time]. Before you build anything, ask me questions until you understand the one job this needs to do, what starts it, what you need from me, where the result should go, and when you should stop. Then show me the smallest version we could try once, by hand, before we automate anything.

That one prompt would've saved me weeks.

A Big Thank You to Leland

If you've been wanting to go from generating with AI to building with it, I can't recommend the Leland AI Builder Program enough. Use the code megan-aibp-60off for 60% off.

I don't get anything in return for sharing that. It's just that good.

Want to Talk About It?

A lot of teams are where I was a year ago. They have the AI tools, and they're still using them like a chat window and generating some content. The tools usually aren't the problem. Microsoft's 2026 Work Trend Index found that only 19% of AI users are both ready to work with AI themselves and backed by an organization that's ready to support them.

That gap is a learning problem, and learning problems are what we do. If that sounds like your team, we'd love to talk about what it could look like to help them build. We create together, and this is a fun one to create.