Insights

The clones started managing me

Published Jul 8, 2026

About the author
Picture of Rachel Huber

Rachel Huber

Rachel is the founder of Bare Digital. Two decades across marketing, operations and technology, now spent building the assets and systems (and the occasional AI clone) that help clients deliver their best work and get recognised for it.
Connect on LinkedIn ›
Share
LinkedIn
Facebook
Email

In March I wrote about cloning myself. It’s still the most read thing on this site, and something I get follow-up questions about regularly.

So here’s the update.

The clones are fine. Dave still does data, Chad still runs the pipeline, and they have a new friend, Terry. The main difference is that somewhere around May, the system stopped being something I run and became something that runs me. I mean that in the best possible way. Mostly.

Four months of upgrades boil down to one move, repeated: catch a mistake, turn it into a process, let the system enforce it.

The briefing grew a bookend

When I wrote the first post, the crown jewel was the morning routine. Every day it scans the whole agency, checks what’s due, and briefs me before I’ve finished my coffee. A simple action, but such a timesaver.

It had a weakness: a morning briefing is only as good as yesterday’s records. And yesterday’s records were only as good as whether I remembered to update them at 5:47pm with one eye on dinner. (I did not.)

So the morning routine got a sibling. The daily wrap runs at the end of the day. It reviews every action taken on my computer that day, reads every client tracker, proposes what should be marked done, asks me what tomorrow’s priority is, and writes the answer to a handoff file. The next morning, the briefing opens with that file. Fridays get a bigger version that reads the whole week. There are monthly and quarterly variants too.

Which means this whole tech system rests on the least glamorous feature imaginable: a text file that one routine writes and another one reads.

Closing that loop killed most of the omissions and my “no, Brobot, we did that yesterday” conversations. Yes, the operating system has a new nickname. It’s Brobot. I change its nicknames regularly.

The optimisation process

Another patch landed on the weekend. I’m now that person who cancels plans because I have access to a new AI model and need to use all the tokens while I have them. “Sorry, I can’t do dinner, I only have extended Fable 5 access until Monday.” Sorry, not sorry.

I asked Brobot to perform a detailed system analysis to uncover performance improvements across both the system outputs and how the system itself operates. My objectives were two-fold: continuous improvement is good, and token use needs to be efficient. If I wanted to avoid hitting my usage caps (which I can do when developing complex tasks or software), I needed to trim the fat.

One of the most helpful findings was that the daily wrap had a blind spot. It could see everything I did on my computer. It couldn’t see my inbox. So a day spent mostly on email rather than file development, to the system, looked like a day where nothing shipped. Same move as always: catch the mistake, build the process. I plugged email in. Now the wrap reads all my sent items from the day, and the morning briefing opens by flagging the emails still sitting in my inbox waiting on me. On security and privacy, the assistant only reads my inbox to keep the task and action registers straight. It doesn’t write or send anything, I draft every email myself. And it runs on a top-tier business plan that isn’t used to train the models or shared beyond my own systems.

There were some other really helpful catches too. We cleaned up duplicate files, messy code and contradictory instructions, and archived old ways of working so they don’t confuse the system.

More AI meant more process, not less

You’d assume handing more work to AI means less structure. The opposite happened. As the assistants took on more, I formalised more.

Each AI-assisted deliverable now has a stringent QA process that goes beyond my manual checks. For example, every website we ship goes through a 55-item launch checklist. The build methodology maps to six quality gates.

One agent does the task. Another runs a critical review against a QA checklist relevant to that task, and they feed each other: issues get flagged, sent back to the first agent, and resolved before anything reaches me for final review. The daily wrap logs the whole exchange, and if the same issue shows up twice, we treat it as a process problem and fix the process.

Rework you don’t learn from is just rework you’ve scheduled for later.

What the process layer bought

Since the first post went live, nine new client workspaces have been onboarded into the system. Same structure every time, so a new client’s folder works identically to one that’s been there for a year.

A favourite from last week: Brobot had been nagging me to launch a digital consulting line since our new site launched earlier this year. Agent Dave had discovered in his scheduled SEO and performance dashboard reviews that our impressions for digital consulting-related searches were growing rapidly. But I had no service page for it, so as those searches climbed we had nowhere to send them. Every time I had a spare hour, Brobot would pipe up with “the digital consulting web page is still in the backlog, should I add it today?” The response was generally no, clear it from the list. Then Dave would see another big spike in related web traffic, and the nagging would recommence.

So when an enquiry came in for digital consulting work, I conceded that Brobot was onto something. We stood up the entire service line in three hours. I asked for an analysis of our existing services, went into a meeting, and came back to a positioning paper, a full draft services page, mockups in our template’s styles, and three sample documents for the types of work I could deliver based on my experience. I reviewed, redlined, refined, published the new web page and issued a proposal. If it converts, it’s my biggest value win in two years. I’ve issued two more digital consulting proposals this week, and one is already signed and scheduled for kickoff. That’s the boring layer, making serious bank.

In June, a client came to us with an urgent launch: new business, no web presence, brand and website needed fast. I still design every logo the old-fashioned way, because AI isn’t anywhere near good enough on logo development (yet). The website, though, I turned around in two days with AI support. Request to ready-to-launch in under a week, no late nights, no heroics.

Neither of those wins came from a clever prompt. They came from months of processes and system improvements. None of this came off a shelf. Brobot runs on my cadence, my quirks, my way of doing things and my nicknames. A system you haven’t shaped to your own workflow is just someone else’s process you’re renting.

The first post said the real capability isn’t in the prompts, it’s in the system you build around them. Four months on I’d go further: the capability is in how aggressively and how continuously you hunt your own mistakes. Every single day. The system is never finished. Every week it catches something new and gets a little tighter. The week it stops improving is the week it starts rotting.

The trackers, the wraps and the checklists are the machinery for doing that at scale. The model underneath matters least. I know this because I keep swapping the models out. Earlier this year, I moved the whole operation from ChatGPT to Claude. More recently, I shifted our image generation from Google’s Nano Banana to ChatGPT Images 2.0.

Things change. Models change. Build the system, shape it to how you work, tend to it constantly, and you can carry the whole thing anywhere.

Enjoy this post?

Subscribe for monthly insights on marketing, BD, and growth strategies.