Don't Automate the Trainer. Automate Around Them.

The economist Alex Imas published an essay recently arguing that AI won't eliminate scarcity — it relocates it. As AI commoditizes more production, the things that remain scarce migrate toward what he calls the relational sector: teachers, nurses, therapists, hospitality workers, trainers, childcare, clergy, guides. The places where a specific human's presence, judgment, attention, warmth, or trust is itself the product. He grounds the argument in René Girard's mimetic desire (we want what others want and can't have) and in his own experimental work showing that when subjects learn a good is AI-made, the exclusivity premium collapses — because AI-made things are perceived as inherently reproducible.

If you take Imas seriously, the implication for service small businesses is sharp and uncomfortable: the conventional AI playbook for SMBs is destroying the most valuable thing they own.

Service SMBs are relational businesses

Almost every small business Runbook serves is fundamentally relational. The trainer at the gym. The inspector at the property. The realtor showing the home. The chef in the kitchen. The aide visiting the homebound parent. What the customer is paying for is a specific human's attention, judgment, or trust applied to their particular situation. The transaction is the relationship. The administrative scaffolding around it — the scheduling, the waivers, the follow-ups, the billing, the compliance — is overhead.

This is the part of the economy Imas says is sitting on the most defensible value as AI advances. Not the least.

That is not how most "AI for SMB" pitches treat it.

The two AI strategies, and which one is the trap

There are two distinct AI strategies a service SMB can take. Imas's framework makes it suddenly clear which one is a mistake.

Strategy A: Automate the relational interaction. AI receptionist. Chatbot intake. AI-generated newsletters with the agent's face on them. AI-narrated marketing videos. Voice bot triage on the phone. Scripted conversational flows for every customer touchpoint. This is what most enterprise vendors and most "AI consultants" sell to SMBs right now, because it's the obvious thing — the customer-facing surface is visible, the labor cost is high, and the ROI math is easy to write on a slide.

It is the wrong move under Imas's framework. You are taking the part of your business with the highest exclusivity premium — the human relationship — and signaling to the customer "this is reproducible." The customer can feel it. They feel it when the receptionist is a bot. They feel it when the email obviously came from a model. They feel it when the trainer is following a script that was clearly generated rather than developed. Once they feel it, the premium collapses. They start comparing you on price.

You burned the moat to save labor cost. The labor cost was a rounding error compared to the moat.

Strategy B: Automate around the relational interaction. Policies handle the waivers, the scheduling, the routing, the follow-ups, the deadlines, the compliance, the documentation. The trainer, inspector, realtor — the human at the customer-facing edge — is freed from administrative remembering. When they show up for the customer, they are more present, more attentive, more reliable. Not because they suddenly became better humans. Because all the cognitive load that used to compete with the relationship has been moved off them.

This is a different shape entirely. The AI is backstage. The customer never sees it. They see their human, being more human-shaped because the operational drag is gone.

The exclusivity premium math

Imas's experimental finding is the key. The willingness to pay for an identical good roughly doubled when subjects learned a random subset of people would be excluded from purchasing it. And the same finding inverted on the AI side: visible AI involvement crushed the premium, because reproducibility is the opposite of exclusivity.

Apply this to a service SMB:

| | Visible AI (Strategy A) | Invisible AI (Strategy B) | |---|---|---| | What the customer perceives | "Reproducible. Generic. Cheaper." | "My person. Reliable. Worth what I pay." | | Pricing power | Pressure to commoditize | Premium holds, may grow | | Defensibility | None — anyone can buy the same chatbot | The specific human + the trust + the reliability | | Cost structure | High variable (LLM per customer interaction) | Low variable (LLM at the edges, behind the scenes) | | Customer churn driver | "I found a cheaper one" | They don't churn — the relationship is sticky |

The vendor selling Strategy A will tell you it's the future. They are right that it's a future — for businesses where the customer relationship was already commoditized at scale (call centers, transactional support, mass retail). They are wrong that it's the future for a gym, a residential services company, a real estate practice, a clinic, a restaurant. In those businesses, the human is the product. Visible AI in the customer interaction is the same as a famous chef putting a sign in the window saying "Our food is now made by robots." It might be true. The premium will not survive the announcement.

The Heroism Tax, reframed

We've been calling the cost of operations living in a founder's head the Heroism Tax. Imas gives us a sharper way to see what that tax actually costs.

The Heroism Tax is paid in two currencies. The visible one is founder bandwidth — late nights, missed weekends, the inability to take a real day off. That's the part everyone talks about, including us, because it's what the burnout owner feels first. But there's a second currency that's invisible and probably more expensive: relational quality at the customer interface.

Every minute the trainer spends remembering whether the waiver was signed is a minute they're not present with the kid in front of them. Every minute the realtor spends tracking which inspection deadline is slipping is a minute they're not building rapport with the buyer. Every minute the inspector spends second-guessing whether they ran every step of the procedure is a minute their attention is split between the customer in front of them and the checklist in their head.

These minutes don't show up on a P&L. They show up in the customer's gut feeling about whether you really see them — which is the thing they're actually paying for. The Heroism Tax is collected from the relational quality of every customer interaction, every day, in tiny installments nobody itemizes.

Runbook isn't just buying back the founder's evenings. It's buying back the quality of every customer interaction by removing the cognitive drag from the people delivering them. The substrate handles the remembering. The human handles the relationship. That's what the architecture is for.

Why this is what Runbook is shaped for

The architectural decisions we've made over the last two years — substrate first, AI at the edges, kill-switch test, multi-actor model, ledger as memory — all add up to a single product property: the customer never sees the AI.

When a homeowner messages Sabir Homes, they don't get an AI auto-reply. The message fires a policy that routes the work to a real human, on time, with context. The homeowner sees responsiveness, not a chatbot.

When a parent checks in their kid at the gym, they don't talk to a bot. The check-in procedure runs in the background; if a waiver is missing, the human at the front desk gets a flag and handles it personally. The parent sees an organized front desk, not an AI receptionist.

When a buyer gets a follow-up from their realtor, the realtor types it. The system tells the realtor that it's time to follow up — but the words are the realtor's. The buyer sees their agent showing up at the right moments, not an automated CRM blast.

This is not an accidental design choice. It is the only design that survives Imas's argument. Visible AI in the customer-facing layer is fatal for relational businesses. Invisible AI in the operational layer is what makes the human side of the business more defensible, not less.

The real positioning

The pitch that follows from this is not "we'll save you labor costs by automating your customer service." That pitch is selling the customer's premium for spare change.

The pitch is: "Your business is your relationship with your customer. We make sure nothing administrative gets between you and them. The AI runs the operation. You run the relationship. Customers will pay more for what you do — because they can tell the difference."

In Imas's frame, this is the only AI strategy that doesn't burn the moat to chase a margin. Substrate-first AI is what protects the exclusivity premium of the human relationship while still capturing every operational efficiency a model can offer. Agent-first AI does the opposite — it puts the model where the human used to be, and the customer immediately senses the substitution.

Close

The AI conversation in SMB world has been dominated by the Strategy A pitch for two years. Replace your receptionist. Automate your sales follow-ups. Let the bot handle your customer support. Each one is a small step that, taken together, finishes the job of commoditizing your customer relationship.

The economists have started catching up to what good service operators have always known: the human in the room is the product, and the product is becoming scarcer, not more abundant. The right move is to protect that relationship at all costs and use AI for everything else.

The AI you can't see makes the human you do see more valuable.

That is the whole positioning of Runbook in one sentence.

---

This post draws on Alex Imas's essay What Will Be Scarce. It continues the positioning thread from Architect Mode Needs a Substrate, The Economics of Authored vs Inferred, and Why Runbook Policies Aren't Claude Skills — adding the relational dimension to the architectural and economic ones.