I built an AI Digital Assistant
Here's how I did for myself, and my team, in three simple steps
Every week there’s a new headline about AI replacing jobs, changing industries, transforming how we work. Most of it is noise. Some of it is real. After spending months deep in it, I can tell you that almost everything I’ve tested still needs a human in the loop somewhere. The autonomous future people are selling you isn’t here yet.
What is here, if you’re willing to do the setup work, is something genuinely useful. Not magic. Not a replacement for thinking. A force multiplier for the person in your business who already has the most context and the most to gain from moving faster.
For me, that’s me.
As owner/operator of a commercial cleaning company the job description is broad. Strategy, clients, team, culture, growth. We’ve spent years building a management layer to handle the volume. Good people. Solid operators. But somewhere along the way I became more of a people and process manager than an individual contributor. I was spending more time in meetings and busy work than actually doing contribution work.
AI gave me a reason to reorient.
I’m not trying to automate my team out of jobs. I’m trying to take back the functions I know better than almost anyone in the building, do them faster, and stay closer to the work that actually moves things. The President/CEO has the biggest knowledge advantage in any company. Most of us underuse it because we’re buried in coordination. Claude’s skills and projects changed that for me.
Here’s how I built it.
Part 1: building the brain
The first thing I did was get Claude Teams and set up a dedicated project. Not a chat window. A project, with files, context, and structure. I called my project George’s DA (Digital Assistant).
I went through every folder I had, personal and company, and pulled the most relevant material. Old RFPs, proposals, competitor analysis, SOWs, industry certifications, sales strategies, Substack articles. I had Claude convert all of it to markdown files so the content was clean and readable, then loaded everything into the project.
Pro tip: I first had Claude convert these verbatim into .md files. This way it didn’t lose context or rewrite anything. But a lot of these files were very long and dense. The models have a hard time reading through everything and maintaining context. It also burns a lot of credits. So once I had them written verbatim I then worked to condense each file to the most important parts without losing context. Based on my research you want to keep .md files under 300 lines to make them super effective.
Then I moved through each area of the business the same way. Sales playbooks. External client docs. Our internal SOPs covering everything from account onboarding to health and safety to people operations. I rebuilt most of them from scratch to reflect where we’re actually going, not where we were three years ago.
By the end I had 63 markdown files covering the full business. That’s the brain.
Tip for owners: Only you should have access to your first DA project. Tell it everything you can about your business, your goals, and you personally so it gets a deep understanding of how you lead and operate. Once you feel like you’ve created 90% of the .md files you need, you then create a DA for your company. Copy in all files, except the sensitive ones, and give your people access. Now your people can operate more autonomously without derailing your vision/mission.
Then I built skills. This is where it got interesting. Skills are basically SOPs for Claude. I built ones for copy editing, reviewing contracts, building proposals, competitor battle cards, coaching cold calls, analyzing RFPs, interview screening questions, brand guidelines, and many more.
When I say “review this contract,” it doesn’t just read the document. It reads it the way a lawyer would, flags what I care about, and surfaces the risks I’d ask about. What would previously have taken me many hours, put me to sleep, and cost me money is now done within a few minutes, is essentially free, and doesn’t overlook any risks. Game changer!
Once the apps were connected (HubSpot, Canva, ZoomInfo, Email, Calendar) I now have one interface where I can do anything I need for any function across my organization. Mind blowing!
Part 2: account management
I run a lot of accounts personally. Relationships I’ve held for years, clients who expect me on the call. The admin that comes with that is real. Meeting prep, follow-ups, gap analysis on who I haven’t touched in months.
I set up a recurring task that runs every Monday morning. Claude goes into HubSpot, pulls every open deal, and gives me a rundown on each one. Where things stand, what’s been said, what’s next. No digging. No prep time. It’s just there.
The second task: on Mondays, Claude finds every customer I haven’t spoken to in at least three months and create a follow-up task for me in HubSpot. The longer it’s been, the higher the priority. That list used to live in my head. Now it’s in my CRM with tasks attached before I’ve had my first coffee.
These aren’t complicated automations. They’re just consistent. The kind of thing that sounds obvious and never actually gets done without a system behind it.
Part 3: Sales Coach
This one is newer and probably the most promising.
Your sales team will only ever be as good as their coach. You can’t just hire sales people, throw them into the wild hunt, and expect them to kill at a high rate efficiently.
You have to find the best talent, onboard thoroughly, give them the playbooks, provide targets/data, teach them the industry, and once that’s all done you basically have to keep coaching and course correcting.
I haven’t yet automated all those steps, but I have automated a lot of the coaching and course correcting. I have Claude listening to our call recordings in HubSpot. It tells me where the rep talked too much, where they missed an opening, where they lost the prospect’s attention, what they did well, and what deviated from our talk tracks.
I’m not in every call anymore but I’m still in every call. The feedback is specific and it’s fast. I can see patterns across calls without sitting through hours of recordings.
For a business our size with a lean GTM team, that’s a real edge.
The honest part
This took time to set up properly. The markdown files need to be clean. The ground rules need to be tight. The skills need to be written well or they’ll produce garbage. If you cut corners on the context, the output reflects it. Even though I am happy with my outputs there are many skills and context files that I edit often when new data or info arises.
But once it’s set up, the unlock is real. I can rip off work in an hour that used to take a full day and an admin to coordinate. Proposals, contract reviews, competitive analysis, account prep. I do it myself now, faster and with more context than anyone I’d delegate it to.
The President/CEO has the biggest knowledge advantage in the business. Most of us give that advantage away by spending our time managing instead of doing. AI doesn’t replace that advantage. It lets you use it.
That’s what I’m currently after.