AI Operations · Small Teams & Agencies

AI systems your team can run without hiring engineers.

I build the automations that take repetitive work off your team's plate, then write it all down clearly enough that you do not need me in the room to run it.

This is not a slide deck on prompt engineering. At JumpStart, instead of four different admin tasks eating someone's week every month, four small programs now handle them on a schedule. At AncestorIQ, instead of trusting an AI tool to always get a fact right, I built a check that catches it when it guesses instead of knows, before that guess gets saved as real data. Every automation drafts or suggests. A person decides anything that cannot be undone.

Builds: Automations, shared knowledge bases, and checks that catch AI mistakes before they ship Built for: Small teams and agencies without an engineer on staff Also available: Fractional COO and Chief of Staff engagements

Email me Book 20 minutes Full profile →

What I build

Every example below is from an actual engagement. Nothing here is illustrative.

Recurring admin

Work that used to live in one person's memory

Instead of someone manually checking every lead to see who has gone quiet, a bot scans them each morning and drafts a follow-up message tagged to the right person. It never sends on its own.
Instead of someone building next month's project list by hand, it gets created automatically on the 1st.
Instead of a stale page sitting unnoticed for months, the system flags it and opens a suggested fix for a person to approve. About 30 fixes opened so far, every one decided by a human.
Instead of running the same three-step client setup checklist by memory, one skill runs all three steps at once.

Result

Recurring admin now runs on a schedule, with a paper trail and a person in the loop on anything that matters.

Knowledge everyone can trust

Company knowledge, kept in one place instead of everyone's head

Instead of company knowledge scattered across docs, chats, and people's memory, it lives in one place both people and AI tools can search and rely on.
Instead of someone manually checking competitor websites for updates, a daily check does it automatically for about a dozen sites.
When that check hit a snag (some sites started blocking it), I fixed the underlying problem instead of letting it fail quietly. A failure now shows up the same day instead of going unnoticed for weeks.

Result

Research that used to take someone an afternoon now happens automatically, and if it breaks, someone finds out immediately instead of three weeks later.

Catching AI mistakes

Stopping a tool from guessing when it should say "I don't know"

Instead of trusting an AI system to always get a fact right, I built a check that catches it when it guesses instead of knows, before that guess gets saved as real data.
Instead of letting automation move ahead when it might make things worse, I have turned it down in writing, twice, and explained why to the client.

Result

The system stops making things up when it is not sure. Everyone knows exactly when a person needs to step in.

How the handoff works

The point is not to make myself permanently necessary. It is to leave your team something they can run and change on their own, without calling me first.

Everything is written down, not just built

Instead of an automation that only I understand, every one gets a plain-language name and a note explaining what it does and why. Instead of a decision that lives only in my head, the workspace or repo keeps a running record of what changed and why. I have inherited enough of these mystery systems from other people to know an undocumented automation is a trap waiting to go off at the worst time, for reasons nobody can explain.

Instructions the next person can actually follow

Instead of starting from a blank page every time, I write down the rules, tone, and facts once, so the next person, or the next AI tool, does not start from nothing. The same thinking applies to what should not be automated: instead of trying to catch a bad result after the fact, I set the right permissions from the start.

A person stays in the loop on anything that can't be undone

Every automation I build drafts or suggests. A person still approves the send, the publish, or the change to real data. That does not change as the system gets more capable. It is the same rule whether the team has three people or thirty.

Worth saying out loud

I have turned AI down in writing where automating would have made a client's data worse, and told them why. That judgment call is the actual product. The automation is just where it shows up.

Who this is for

Small teams and agencies without an engineer on staff, where the same admin tasks eat someone's week every month and hiring is not the answer yet.

What I build it in

Whatever your team already uses: Monday.com, Notion, or plain files in a shared folder. I do not make you switch tools just so I have something to build in.

What this is not

Not a hands-off robot. Every automation drafts or suggests; a person approves anything that can't be undone. Not a software company either. I am an operator who builds the tool because it is faster than writing a request and waiting for someone else to build it.