I run about five client engagements at once, and roughly eleven AI agents run alongside me. Most of them live in a system I built on OpenClaw, my own agent setup rather than a product I bought. The agents handle research, drafting and operational work across all five accounts, and I keep the judgment.
I have started at nine or ten companies, and for most of that career the constraint was hours, not ideas. A fractional practice makes that constraint sharper: five clients means five sets of context, goals and people, and one of me. Before AI, serving that many meant serving each one worse. Now the agents carry the work that used to eat my evenings, and I show up to every call prepared.
None of these tools think for me. They fetch, draft, summarize and remember, which turns out to be most of the work.
How many AI agents does it take to run a fractional practice?
For me, eleven: a few per engagement and a few shared across the practice. Each one has a narrow job and a written spec. One pulls market research before a strategy session. One drafts the first version of a client update from my notes. One watches a shared inbox and flags the threads I need to answer. A couple carry the context of a specific account, so I do not start cold every Monday. The full build is in my OpenClaw setup guide.
Eleven sounds like a lot until you count what a fractional executive holds in their head across five companies. It is barely enough.
An agent should do one thing I would otherwise do myself, and it should do it the same way every time.
What I use AI for every day
Drafting, research and operations. In all three the work is real and the judgment required is low, which is where an agent earns its keep.
Drafting is the biggest. I no longer start documents from a blank page. I give a drafting agent a rough outline or a set of voice notes and tell it the shape I want: a strategy doc, a client update, a growth plan. It builds the scaffold and I bring the substance. The draft that comes back is never the one I send, but arguing with a draft is ten times faster than writing one.
Research comes before every client session. An agent pulls the market context, the competitive picture and whatever public signal exists on the question. I read it skeptically, because half the value is knowing what to throw away, but I walk in with a point of view instead of building one live in the room.
Operations is the quiet category: logging what happened, tagging it, tracking which client I have not touched in two weeks, turning notes into next steps. None of it needs me, and all of it used to cost me a Sunday. On a short engagement, every hour spent on admin is an hour taken from the mandate.
How I turn meeting transcripts into action
Granola records client and coaching calls, and an action-extraction workflow runs over the transcript. I keep capture and extraction separate on purpose. Granola lets me stay present in the conversation instead of typing. The extraction step does the part that used to slip: turning forty minutes of talk into the three things that have to happen and who owns them. I edit that list, because the agent often gets the nuance of a commitment slightly wrong, and then it goes wherever that client's work lives.
The payoff is memory. What quietly damages a fractional relationship is forgetting what you agreed to three weeks ago. A searchable record and a clean action list for every conversation means I stop dropping threads, and that matters more to clients than any single insight I bring.
Where AI fails in an advisory practice
It fails when you let it sound confident. These models are agreeable: they hand you a clean, well-structured answer that is quietly wrong, and the polish is what makes it dangerous. I treat every output as a draft and verify anything with a number or a claim in it.
Before I trust an answer, I ask the agent where it is most likely to be wrong. It is good at naming its weak spots when asked directly, and that one question has kept a bad figure away from a client more than once. The last twenty percent of any piece of work, deciding what is true and what matters, is the part I cannot hand off.
The subtler failure is doing the wrong work well. Point an agent at the wrong question and it produces a polished answer nobody needed, and the polish makes the mistake harder to catch. Direction stays with me; the agent only adds speed, and speed in the wrong direction is a faster way to be wrong. I wrote more about that tension in what your AI-obsessed CPO is not telling you.
Here is how the work splits in practice:
| Work type | What the agents handle | What stays with me |
|---|---|---|
| Drafting | Scaffold a doc from my notes and voice memos | The argument, the point of view, what to cut |
| Research | Pull market and competitive context before a session | Which parts are real and which to throw out |
| Meeting notes | Capture the transcript and extract open items | Whether a commitment actually got made |
| Operations | Log interactions, tag threads, flag stale accounts | Which client actually needs me this week |
| Verification | Surface its own likely weak spots on request | The final call on what is true |
Where to start if you run a portfolio of clients
Start with one narrow agent that does a task you already dislike, and get it working before you build anything else. Designing the whole system on day one is how people quit. Mine was drafting client updates. Hand exactly that task to an agent with a clear spec and live with it for a week. Once the hour it saves is obvious, you will build the second one without being told. That is the path I followed on OpenClaw, and the same logic I use in a fractional CPO engagement: prove the smallest useful thing, then build from there.
AI has not made me smarter. It has made my habits visible and faster. Because I know what good product and growth work looks like, eleven agents clear the noise and leave my hours for the one thing they cannot do, which is decide what matters.
