What Months of Building an AI Brain Actually Looks Like (and Where This is Heading)
This is part seven of our Putting AI to Work in Your Tour Business series. Watch the full series here and read part one here.
Months ago our AI Brain was a small folder with maybe fifteen lines of brand voice notes and a short list of team members. Today it is the thing we work from every single day. It runs scheduled tasks, drafts personalized follow-up emails using context pulled from three different places at once, checks our YouTube channel every week and updates its own index, and lets us direct it from our phones while we are away from the desk.
In this piece, we are doing something different. Less framework, more behind the scenes. What our AI Brain actually looks like after months of real use, what is working, what is still hard, and where we think all of this is heading.
The Brain Grows by Catching Itself
If you’ve been following along, your AI Brain is now a folder of markdown files. An instructions file the AI reads before doing anything. A nervous system of connectors. Skills, plugins, and agents. The self-correction loop where you correct the AI once and the correction sticks for next time.
“Every correction added is one the AI never has to be told twice.”
Our brand voice document started small. A month later it had grown by an order of magnitude. Not because we sat down and wrote it. Every time we caught the AI doing something we didn’t want, we added a line. Skip these words. Drop the hyperbole. Sign off this way. Those corrections compound. You can see every one of them sitting in the file.
If you want the full setup walkthrough for building this from scratch, the free AI Brain Builder covers it step by step.
The team file went the same way. We started with names and roles for a couple of contractors. Now it has every coach, their LinkedIn profile, links to any YouTube interviews we have done with them, plus contact info we hadn’t thought to include at the beginning. When a project needed that context, we added it. The brain grew with the work.
Context Built from Real Conversations
Two pieces of context have been disproportionately useful. The first is a YouTube index. This is a markdown file with every training video, a short description, and the direct link. We didn’t even have a centralized database for that before. Now whenever we want to reference one of our free training videos in a piece of content, the AI just knows where to find it.
“Capture how your customers actually describe their problems. Put it in the brain. Watch the quality of everything else improve.”
The second is a client language file. We pulled the exact phrasing our audience uses straight out of member profile surveys. “I do everything. I am feeling burnt out. Tired of being a one-man band. Cannot get time off. No one to ask questions. Too busy to work on the business.” That is not us guessing at how operators talk about their problems. It is the actual language coming out of the actual people we serve. Loading that into the brain changes the quality of everything the AI produces for marketing.
This builds directly on the context engineering work we covered earlier in the series.
That kind of context doesn’t show up on day one. It shows up around month two or three, when you hit a task and think, “Wait, the AI doesn’t know this yet, and it should.”
The Follow-Up Workflow That Used to Take Hours
Here is a concrete example of what the brain unlocks:
We had a year of strategy call notes sitting in Apple Notes. We exported them all as markdown files and moved them into the brain. Then we set up a single task: Go through Calendly and find every strategy call from the last 12 months that has not turned into an enrolled client. Cross-reference my notes for each one. Draft a personal follow-up email that references the specific business and the specific challenges they shared.
That’s pulling information from three or four different places. We didn’t spell out where to find any of it because the AI Brain already knew. It assembled everything, drafted around 30 personalized emails, and dropped them into Gmail as drafts. I reviewed the first two.
Then I had to take the girls to swimming lessons, so I approved the rest and the AI sent them one by one while we were gone.
“Hours of manual follow-up work, done in the background.”
That is what context plus connectors plus a clear goal makes possible.
Scheduled Tasks and Agents on a Timer
Inside Claude Cowork there’s a scheduled tasks panel. You give it a name, a prompt, and a frequency. Manual, hourly, daily, weekdays, weekly. You pick which model to use. Sonnet handles most day-to-day work because it’s efficient with tokens. Opus is what you want when building out the brain or creating a new skill.
I have two scheduled tasks running now. A weekly YouTube index update that checks our channel for new videos and adds them to the brain. And a weekly wins cleanup that goes through our task list, celebrates what got done, and clears completed items. Small jobs. They run themselves.
One important note. Scheduled tasks only run while your computer is awake. If you want a task ready when you sit down in the morning, your laptop needs to stay on and Claude Cowork needs to stay open.
Dispatch and the Remote Control
Dispatch is newer, but it’s a meaningful shift. It lets you communicate with Claude Cowork on your desktop from the Claude app on your phone. As long as your computer is on and Cowork is running, you can hand it tasks, give approvals, or kick off complex projects from anywhere.
This solves one of the genuine limitations of the AI Brain approach. The brain lives on a local computer, which means it’s not in the synced chat on your phone. Dispatch closes that gap. The brain stays where it is. You get a remote control.
Exporting the Brain to a Teammate
Something else worth sharing. We exported a copy of our AI Brain to Abby on our marketing team. We talked to the AI Brain itself and said, “give Abby the context she needs for her work.” It pulled out the relevant memory, stripped the personal context that doesn’t apply to her role, packaged it into a zip file, wrote the setup instructions, and we sent it to her over Slack. She dropped it into her own Cowork folder and was up and running within minutes.
“Onboarding context that used to take weeks now moves with a zip file.”
The two instances will get out of sync over time. We are okay with that. She will build her own skills specific to marketing. We will build ours specific to other work. We can always export updates in either direction. And a bigger shift is coming anyway. Within a year there will be team plans that let multiple people share an AI Brain natively. We are just doing it the manual way until then.
Honest Talk About What is Still Hard
We don’t want to pretend this is all clean. A few things are still rough.
Token usage and cost are real. Running tasks with a lot of context isn’t free. We upgraded from the $20 monthly plan to the $100 Max plan because we kept hitting limits. We would pay that many times over for the efficiency we’re getting, but it’s worth being upfront about. Especially during the brain-building phase, you will use Opus heavily, and Opus uses tokens fast. We recommend a temporary upgrade while building out the brain, then dropping back down once the heavy lifting is done.
Connectors are not always consistent. Sometimes a Zapier connection works one day and need a refresh the next. Sometimes a native MCP integration gains new functionality that requires reconnecting. We think this is just the price of being early in a fast-moving space, but it does mean some maintenance.
And the mental shift of having your AI Brain live on a local computer takes some getting used to. The computer has to stay awake. It can’t do its work while you’re on a plane with the laptop closed. Dispatch helps. But the underlying constraint is real.
Other Tools That Plug In
We have focused a lot on Claude Cowork, but there are other tools we use alongside it.
NotebookLM from Google is excellent for research. It pulls YouTube transcripts directly, which is something Cowork cannot do natively. We used it to build an advisory board of entrepreneurial thinkers we respect. We had it research their most popular podcast episodes, fed 40 or 50 transcripts into NotebookLM, and got a briefing of key ideas and language patterns. That output then migrated back into our AI Brain.
ChatGPT voice mode is still the best fluid spoken conversation with AI we have used. We sync our AI Brain into a Google Drive folder, then connect that folder into a ChatGPT project. That lets us talk through ideas while driving or stepping away from the desk, with the full context of our business available in the conversation.
Google has spectacular image and video generation tools. Eleven Labs is excellent for voice generation. None of these are competitors to your AI Brain. They are specialized tools that plug into it. The brain stays central while other tools become connectors.
Where the Pace of Change is Heading
Let’s zoom out for a minute, because the pace of change in this space is significant.
Open Claw is an open-source AI agent that went from zero to the most popular project on GitHub in about four months. It can control your computer, run tasks, and manage files. Plus it’s free and highly capable. It had real security concerns early on, which Nvidia addressed by releasing Nemo Claw, a security layer that wraps around Open Claw with sandboxing, privacy controls, and policy enforcement. Manus, acquired by Meta, is building similar capabilities as a desktop app. Perplexity Computer has its own version. ChatGPT and Gemini are heading into agentic territory next.
“The tools will change. The underlying structure will not.”
Within a year, most of these capabilities will likely be standard across all the major platforms. Scheduled tasks. Computer control. Remote phone access. Multiple agents working in parallel. AI employees handling specialized roles.
This is why we’ve been building around principles instead of features. Features change every month. The underlying principles do not. If you’ve built out your AI Brain, your nervous system, your skills, and your connections, you are ready to plug into whatever platform leads the pack next year, or whichever one is most cost-effective. The operators who are still bouncing between disposable chat conversations, still copying and pasting context into every prompt, are the ones who will fall behind.
Where to Go from Here
We don’t pretend to have all the answers here. We are figuring out a lot of this as we go, the same way the members of our coaching programs are. What we can share is what is working now, what has us excited, and what we are watching closely.
If you want to tag along, you can join one of our coaching programs. You can book a free strategy call with someone on the team.
You can join the free Tour Business AI Lab community, where operators share what they are building and where we share what is working.
If you want to set up your own AI Brain from scratch, our free walkthrough has everything you need.
If you got value from this, share it with someone in the industry who would benefit. That is how the good ideas travel. Thanks for being here.



