Hey everyone!

Here’s what I’m writing about today:

AI keeps showing up in this newsletter because I can't ignore what I'm seeing across our groups. I have a window into 100+ business owners and their day-to-day operations every month, and the gap between what's actually happening inside these businesses and what the general public knows about AI is widening fast.

This newsletter is my attempt to close that gap for you, 6-12 months before it becomes common knowledge.

If you want ongoing access to that kind of inside view, book a call with me and let's talk about a ScalePath membership.

Okay, let’s get into it:

This AI use case comes from our member Charles Miller, owner of Cooper Demolition in Colorado Springs. He wrote about this publicly on his twitter but gave all of our Scalepath members a live demonstration earlier this week of how his tool works.

_{Side note: every 2 weeks we bring in a Scalepath member to show off what they’ve built and are actually using in their businesses. What’s saved them time, money, given them better data to make decisions, armed their teams with better tools, etc. If you like reading this, you’ll see a whole lot more inside Scalepath. Charles talked about this in Scalepath months before you’re reading this.}

So what is this tool Charles cooked up?

An Artificial Intelligence Virtual Assistant (AIVA).

He calls his AIVA Peggy Olson… named after the Mad Men character. If you've seen the show, the name makes sense. Capable, always on, handles things so you don't have to think about them.

Peggy is essentially a Virtual Assistant that deploys task-specific subagents to complete tasks on a recurring and scheduled basis. She has her own email inbox, engages with clients, and when prompted either manually or automatically will create financial reports, estimate jobs, conduct sales activities, and book flights, manage Charles’s calendar, etc. Most everything a VA does, Peggy does and much more.

Peggy runs 24/7 on a Mac Mini in Charles's closet. The machine does nothing except run Peggy. It restarts itself every night to prevent entropy from building up in the system. It executes tasks on a schedule with no screen and no clicks required. Charles can access it remotely from anywhere via a free tool called Tailscale, which makes his laptop and the Mac Mini behave as if they're on the same desk. All of Peggy's memory lives in Airtable, so every part of the system reads from the same shared brain.

Total cost: $200 a month for Claude Max. $800 for a Mac Mini. He has not yet hit a point where he needed to pay more in token usage. (his company already pays for Airtable and Outlook, so no new costs besides Claude and a Mac Mini.)

Here's how Peggy is structured:

There's a Persona Layer at the top that acts as a filter on every output that leaves the system, keeping tone, format, and communication style consistent no matter which agent produced it. Below that is the memory context layer, a set of Airtable tables that Peggy reads before acting and writes to after acting. This is what prevents the hallucination and context-loss problems most people hit with AI assistants. Then underneath that are the agents themselves, each sandboxed to their own area of the business with access only to their own data sources.

The agents:

Finance Agent pulls live job data, generates the weekly Work in Progress report for his accountant, maintains a 13-week cash flow model, and runs AR aging and reconciliation on a fixed schedule. If his cash runway starts shrinking, say from week 12 down to week 10, he gets a high-alert email automatically. It all lands in the right inboxes before the workweek starts. He was actively searching for a controller when he first tested this. He told Peggy she was plugged into QuickBooks, gave her the vendor net terms, including handshake net-90 agreements that weren't in the system, and asked her to build a cash flow model. She did it in 20 minutes. He said it was as good, if not better, than anything he could have built himself in four to six hours.

Operations Agent scans email every 30 minutes, triages what needs attention, maintains a to-do list, and sends Charles a morning briefing before he wakes up. He deliberately set it to 30-minute intervals rather than something faster for two reasons: token efficiency, and so Peggy's response cadence looks like a person, not a bot. If he's on a flight, she knows his schedule, and upon landing he gets an email: “here's everything you missed.”

Sales Agent keeps the CRM current, monitors deal status, sends out proposals and bids, and follows up two weeks later. Peggy is the first point of contact for most of Charles's clients at this point. 

Estimating Agent is where it gets interesting for his specific business. Charles uploaded Walker's Building Estimators Reference Book, the industry standard, into Peggy's knowledge layer. Without getting too specific, it’s doing an absurdly good job. 

Every agent operates under Peggy Olson's identity, her own email address, her own tone. His team emails her like any other teammate. He interacts with her through Claude Dispatch on his phone, which has full visibility across all agents. One conversation, one point of contact, from anywhere.

AND Charles built an automated feedback loop that makes Peggy smarter every week without him having to do anything intentional.

Every time any agent does something, it logs what it saw, the pattern it matched, and the choice it made into a Knowledge Log table in Airtable. Every Sunday at 8pm, a scheduled task reads the whole week of logs and synthesizes patterns. Every Sunday at 9pm, a second task, called Skill Review, reads those patterns and proposes amendments to existing skills for Charles to approve. Because Peggy runs in Cowork, she can read, write, and delete her own skill files. She can update her own behavior dynamically. Charles said he can't think of a single skill file that's more than a week old in terms of its actual content, because they're all being continuously updated.

What that looks like in practice:

Charles sends one correction from his phone, Peggy writes a rule for herself, stores it in Airtable, and that situation never repeats. He corrected her once on a routine subscription invoice showing up on his to-do list. Never happened again. Stack that up over months and the shape of what Peggy flags today looks nothing like what she was flagging in January.

He also has a Decision Council tool he can invoke on demand. He asks Peggy to "counsel" a decision, and she spins up five separate AI personas, each coming from a different angle: devil's advocate, pragmatist, etc. They argue against each other, reach a consensus, and he gets back a summarized transcript of the debate in three to four minutes. That one runs on Opus, Claude's most heavyweight model. The inbox scanner, by contrast, runs on Haiku, the cheapest and fastest, because all it needs to do is read and flag. He was deliberate about matching each task to the right model, which is a significant part of why his token costs aren’t crazy.

He originally built Peggy because he was considering hiring a VA to get his time back, but then wanted to see if Claude could do it. His ultimate goal is to be able to lower his golf handicap with the time that’s been returned to him (it’s working).

2 months ago a member stepped back from Scalepath primarily because of bad timing. One month he missed a call because he was sick, another a family vacation, the next one a GM on maternity leave. He planned to be back in 2-3 months. I checked in with him recently.

AI has been really damaging his business in the last couple of months.

A large portion of his revenue came from financial templates, the kind people would find through Google, pay $20-200 for, and download. He got hit twice at once: the Google algorithm that was feeding him traffic turned on him, and separately, anyone who uses Claude regularly can now describe what they need, connect it to their actual numbers, and get something more customized than any template in about 20 minutes, for free.

He's probably the member with the most direct exposure, but he's not alone. Two categories this AI doom conversation regularly comes with is accounting/bookkeeping firms, and website agencies.

Over the last several months I've watched the owners in both categories oscillate between two positions, often in the same conversation:

Position 1:  "This is incredible, my team can move twice as fast, I can 4X revenue without adding headcount."

Position 2: "If I can automate 75% of what we do, my clients can figure that out too, eventually they'll to stop paying us."

There’s a large discussion to be had here. On one hand I just hired a website agency because I just wanted to hand off work to someone so they can drive the process and they’ve built thousands of websites, my team has not, and they’ve got a depth of experience here that we don’t have. Same goes for our bookkeeping/tax.

The expertise and the judgment are still worth paying for. The question is whether that stays true in 12 months, or 24, or 60?

I don't think anyone knows. But after talking through it in our professional services groups, here are a few responses that seem defensible:

Exit Horizon: Go one offense, build fast, reinvest in marketing, acquire new customers, double the company in 1-2 years and sell, get out while you can.

Technical defensibility: Go on defense, launch new product lines that have some defensibility against AI. A bookkeeping and tax firm launching fractional CFO work. A website Agency launching SEO, digital marketing, or just multiple services under one umbrella to become a one-stop shop for everything.

Niche dominance/Branding and Positioning: This is the one I find most interesting.

There's a book called 7 Powers by Hamilton Helmer about what actually makes businesses durable. The core idea is that lasting businesses tend to have some structural defensive position. One of the 7 powers is “branding” and while most of us won’t have the brand of Dyson, Tiffany, or Coke, we can still find it in niches.

One of our website agency members does 90% of his revenue locally, just from his city of around 250,000 people. Being the undisputed number one website agency in a mid-size city is a real position. It's built on relationships, reputation, and local trust that an AI tool or a remote competitor can't easily replicate.

One of our accounting firm owners has built a foothold in the franchise market, specifically major fast food franchisees. He knows exactly what those owners need, what they're worried about, how their books work, what the franchisor expects. People pay up for that. Not because accounting is scarce, but because that specific knowledge applied to that specific situation is.

I think we're all figuring this part out. If someone has a better idea or wants to jam on this, I'm all ears.

This one is a pretty simple observation.

Most of our group members acquired their businesses in the last 1-5 years. Over the last six months or so, I've noticed more conversations about selling, whether that's a full exit, rolling equity into a platform, or selling a majority stake while staying on. It's still only a couple handfuls of people, but it's come up more in the last six months than in my first two years of running peer groups.

These are good companies by the way. Profitable, growing, well-run. This isn't distress. So what's driving it?

It's never one thing. But after sitting in these conversations, a few themes come up:

Fatigue. Three years in, you just get tired of dealing with all the people problems. Tired of all the attrition, tired of feeling like you have a great stable team and having it crumble just to rebuild.

A great multiple. A few members are watching their industries go through consolidation cycles right now. When you’re looking at 6X - 10X valuation on $500K of ebitda, and you acquired at 3X, everyone goes through reconsidering their “buy-and-hold-forever” plan. In the last 3 months we’ve had members get 10X offer at $300k EBITDA, 6X on $1M of EBITDA, and 6X on $700k ebitda. PE is paying more and it’s real.

Taking chips off the table. Several members are sitting on businesses that are heavily leveraged from the original acquisition. Three years in, the equity is still thin. The appeal of rolling into a larger platform at a built-in multiple expansion, even at a modest equity stake, is that it converts paper value into something real. As one member put it: a small piece of a watermelon versus 100% of a grape.

Capital constraints limiting growth. This one comes up constantly. A member in commercial contracting described being capital limited on acquisitions he could otherwise execute. He's got two or three deals he could get done, doesn't have the money to do them. Selling equity to a family office solves that problem. 

Do something else. A few members have started to feel the pull of a next chapter, or simply a better industry they’d be happier working in or believe they’d be more successful at.

New opportunities. After several years of operating, you know most of the big players in the industry and some kind of opportunity will present itself that wasn’t there previously. And you have the industry know-how and the positioning to evaluate it and know it’s good. Some of our owners see that opportunity as being more “worth it” than just growing their current operating company or even trying to buy more businesses themself.

I’ll end this newsletter with a highlight of a recent discussion on the business owner who got a 10X offer on $300K of EBITDA.

His words on our peer group call when we talked through this: "I have no intention to sell. But it's an insane offer I’d be crazy to turn down.”

The group was split pretty cleanly on whether to take it. Half said take the 10X, even if they negotiate you down to 8X it’s still a great price, terms were also great, walk away with the money. The other half said wait, grow the business 50-70% over two more years, there’s a chance that 10X offer will stand for a period of time, sell after you grow it a bit more.

Either way, that owner had no intention of selling, and now his main question is “do I sell it now or in 1-2 more years?”

I bet you this will keep happening. Not necessarily at 10X on $300k ebitda, but many “buy and hold forever” folks will sell when they’re burnt out and get an above average multiple.

Seriously, our peer groups are amazing, I promise. If you’re a business owner book some time to talk to me here

Rand Larsen