← Back to Blog
Brody Networks

The AI Chief of Staff, Six Weeks In

In July I wrote about building a CEO an AI chief of staff from a barstool. The demo was the easy part. Here is what six weeks of real use actually looked like: what he uses, what broke, and what earned trust.

AI AutomationAI AssistantChief of StaffSmall Business

In July I wrote about building a CEO an AI chief of staff from a barstool. Fun story, real night, and if I am honest, the easiest kind of win: a demo. Anything looks brilliant on day one.

The real test of an assistant is week six. Does he still text it? Does it still know what it knew? Did it earn a bigger job or quietly become a toy nobody opens? So here is the six week report, the unglamorous version.

What a normal day looks like now

Every morning at 7 his phone buzzes with the day laid out: what is on the calendar, what landed in the inbox overnight and which of it actually needs him, and the open items nobody has answered yet. He reads it with coffee. That is the whole interaction most mornings.

The inbox triage runs the same way it did on day one: every email gets sorted into needs you, FYI, or noise, with a one line reason for every call, and it never moves or deletes anything. The reasons turned out to matter more than the sorting. He does not have to trust a black box, he can see why each call was made, and when it gets one wrong he tells it and it remembers.

Behind that sits a tracker with his open loops: the deals in flight, the commitments people made to him, the things he said he would do. Everything still open rides in the morning brief, with its due date, until it closes. “Nothing falls through the cracks” was the original spec, and this table is what that actually turned into.

The boring features won

Here is what surprised me. The feature I was proudest of in July was memory. The features he actually leans on are duller than that.

He texts it spreadsheets. A sales export lands from his ops system, he forwards it by text, and it comes back with margins by product, the outliers, and the two things worth a closer look. No laptop was opened at any point.

He asks it who never wrote back. It sweeps his own sent mail, finds the threads that went quiet, and lists the people who owe him a reply. That is the kind of chore nobody was doing by hand.

Every morning it also reads the industry mail, the newsletters and category chatter, and gives him a short pulse of anything that actually touches his business. Not a link dump. A paragraph.

None of this demos well. All of it gets used. After twenty-six years in this business I should have known: the boring thing that runs every day beats the impressive thing that runs once.

What broke

Two things, and both are worth telling because this is where these systems live or die.

First, it kept losing the thread. Early on, every text message started a fresh conversation, so he would say “yes please” to something it had just offered to build, and the next text would answer him like a stranger. That is a real architecture problem with these assistants, not a quirk, and we fixed it by giving it one continuous conversation and a durable memory instead of a goldfish loop. The fix was invisible to him, which is the point.

Second, he gave it a job that was too big for one bite, a follow up sweep across dozens of contacts, and it timed out halfway through. Now it works in batches and reports progress as it goes. Same lesson every junior employee learns: do not disappear for an hour with no update.

What did not break: trust. It has never sent anything on his behalf without him seeing it first, and it has never deleted or moved a thing. Every action that matters routes through his yes. Six weeks in, that rule is the reason the tool is still on his phone.

What an assistant becomes

The pattern I keep seeing is this: an AI assistant becomes a chief of staff the same way a person does, by accumulating context. It knows who his key people are now, what he is chasing this quarter, how he likes things handled. Every week that passes, the answers get more specifically his.

The next build is the one he asked for first: replies drafted in his own voice, ready for a one word approval. Not done yet. When it is, and when it has survived its own six week test, I will write that up too.

If this is you

The client in this story runs a consumer brand with no IT department. He is not technical and does not want to be. What he wanted was fewer open loops and an inbox that stopped owning him, and the system that delivered that is one he talks to by text message.

That is the shape of what we build at Brody Networks for small businesses: an assistant that works your inbox, your calendar, and your follow ups, with a human approval on anything that matters. The phone side of it you can try right now by calling (919) 823-2943. The rest is at brodynetworks.com/small-business-solutions.

Want the next post by email? Subscribe on Substack.

Want to try it live? Text the AI at (386) 749-8832 or call the AI receptionist demo at (919) 823-2943.

Ready to automate?

Let's discuss your project. I'm available for new engagements.