Back to all articles

AI Workflows

My AI maternity leave experiment worked. Then the tools changed.

An honest recap of handing Watch Me AI to agents during maternity leave: what worked, what changed, and the surprise of exploring and delegating substantial work from my phone.

By Mollie Amkraut Mueller · · 8 min read

My AI maternity leave experiment worked. Then the tools changed. A cream and coral cover with a pause motif.

At the end of May, nine months pregnant, I spent three weeks setting up AI agents to keep Watch Me AI running while I had baby Louis.

By July, parts of the system I’d built already felt outdated. That is a very 2026 problem to have.

I’m back now, so here’s my honest recap of how the AI Maternity Mode experiment went.

TL;DR

What I handed over

I’d set up a newsletter agent to research topics and draft the weekly issue, and a LinkedIn agent to turn those ideas into posts. AI Mollie was there to answer visitors’ questions and point them to resources. Another system was meant to bring people to my products, especially the AI Business Playbook, and help test new things to sell. Research and metrics agents watched what was happening and brought me updates.

The idea was that these systems would keep working while I mostly stayed away from my laptop, with me stepping in for important decisions.

And parts of it really did work. Newsletters went out while I was offline. Two people I’d never met bought a small guide through a newsletter and tool the system had put together. I didn’t even notice the sales for two days because the notifications went to my Promotions tab. I was delighted, then immediately wondered whether the buyers were getting everything they needed.

It wasn’t a lot of money. But “someone bought something from my business while I was on the couch with a newborn” was one of the things I’d set out to test.

Then the ground moved

The first change came from AI itself. Claude Fable 5 arrived in June. GPT-5.6 followed in July, along with a new version of ChatGPT Voice that was actually good to talk to. Then Grok Bot came along toward the end of my leave, with a different way of handing off whole tasks. Each change made me rethink what I could ask AI to do and how I should set it up.

This was wonderful in theory. In practice, I’d just spent weeks building a system around the capabilities available in May, and I did not have a spare week with a newborn to start over.

I adapted parts of it as I went. But the specifics of which model I used for what, how I directed it, and what I thought was worth automating changed several times over the summer. The setup I started with was no longer the setup I’d recommend.

There’s a strange paradox in that. AI is getting better so quickly that it should make it easier to step away from work and delegate more. It did make that possible for me. But the same pace also made me feel as though I’d missed an enormous amount in just a few months. Luckily, I wasn’t entirely out of the loop, which I’ll come back to.

The second change came from me.

Before leave, I’d given the agents a goal of selling more AI Business Playbooks. They made strong plans for the marketing, content, and offers, and got to the point where they needed a few approvals from me. Meanwhile, I was having the kind of business thoughts that seem to show up when you finally step away from your business. Did I actually want to keep pushing the Playbook? What did I want to focus on when I came back?

I couldn’t quite bring myself to approve a campaign for a direction I might leave behind.

That’s a management problem, and I was the manager. These agents don’t need me to spell out every step of a task. They’re remarkably good at working out how to do something. But they do need to know what I’m trying to accomplish, and they need decisions from me when that changes. The experiment didn’t do wonders for my bottom line, and I think my uncertainty about the direction of the business was a big reason why. Classic management problem.

I learned something similar from the newsletter. As the models improved, the writing got much better. Some drafts sounded like me and were genuinely good. But I kept feeling uninspired by the topics the agent picked, even though I’d given it detailed instructions for finding them.

Eventually I realized that the most fun part of this newsletter, for me, is sharing what I’m actually trying and learning about AI. I was spending less time doing that because I was caring for Louis. The agent could produce a good piece of writing; it couldn’t give me a fresh experience I was excited to tell you about. So I sent fewer newsletters.

The experiment I didn’t plan

Somewhere in the middle of all this, Maternity Mode became something else too.

I’d started by asking, “How much of my business can AI run without me?” But improvements in the models and in voice mode raised another question: “How much can I do with AI, without sitting down at a computer?”

The new voice experience was a big part of it. I could have an actual back-and-forth conversation, interrupt when I wanted to change direction, and move between talking and typing in the same thread. I could give an agent a fairly involved task from my phone instead of saving the idea for the next time I opened my laptop.

On a two-hour walk with Louis and Cali, I kept voice mode open and talked through a strategic idea that had been sitting in the back of my mind. I could explore possibilities, push back, change direction, and keep walking. While nursing in the rocking chair, I could capture an idea, ask AI to investigate it, and pick up the thread later by typing on my phone.

This went well beyond dictating an email. When I started wondering whether to focus my consulting work on coaching and leadership development firms, I used AI from my phone to turn that thought into a market hypothesis. It researched 100 potential firms and found possible warm paths to their leaders. I then had it draft personalized requests for introductions using context from my past emails and meeting notes. I reviewed every draft and asked for changes where needed, but rarely had to write one from scratch. I even had it turn the research into a CRM I could use when I returned to work.

That was a surprising amount of substantial business work to do without opening my laptop.

For me, this mattered as a parent. I love my work. I wanted time with my baby, and I also enjoyed having a way to think, create, and follow an idea when one came to me. It meant I could keep my brain engaged with AI as it changed, when I felt like it, even though I wasn’t sitting at my desk.

To be clear, the majority of my baby-couch time was spent watching Love Island. But when a business idea came to me, I could follow it without putting Louis down and opening a laptop.

I went into leave trying to make the business more autonomous. I came out with a clearer sense of the split I actually want: autonomous where I want to hand the work off, hands-free where I don’t.

Could you do your own version?

I think so. But I wouldn’t hand you my June system and tell you to copy it. If I were having a baby this month, I’d build it again with today’s tools.

What I would carry forward is the approach: get clear on what should keep happening while you’re away; give agents the context and goals to pursue it; decide what they can do alone and what needs your approval; and check the work that actually ships. Leave room for both the technology and your own plans to change.

That’s where I’ve landed so far. I’m back at work now, still using agents, and much clearer about the work I want them to take on and the work I’m excited to keep for myself.

Also: AI Arts & Crafts is back Friday, October 16 at 12:30 p.m. Pacific, online. Come make something with us.

Mollie

P.S. I’ve put together the setup I’d recommend now, including the Opportunity Scout brief and a guide to handing off a substantial task. See the practical Maternity Mode setup.