Where the blame lands
Lesson video in production
The full lesson text below is complete — the video version lands with launch.
Where the blame lands
Ask a different question than everyone else asks. Not "what could AI do here," but "when something goes wrong here, who does the customer call, and what does that call cost?" The answer maps the deflection in a business with surprising accuracy, because the places where a human must absorb consequences are the places a machine cannot be paid to stand.
Liability is a moat
Think about what actually happens when work goes wrong. The cleaner breaks a faucet. The notary botches a page count and the refinance stalls. The biller mishandles patient data and the practice faces a privacy audit. In each case the customer's remedy is a specific, insured, identifiable person. Businesses that operate inside that structure carry a built-in resistance to machine substitution, for a blunt commercial reason: someone must be sueable, bondable, or strippable, and a language model is none of those things.
This is why the drone services course sits at 3 rather than lower despite the aircraft practically flying themselves now. The Part 107 certificate is a license a human can lose. The shot list is judgment. But the liability when a drone clips a wedding canopy lands on the pilot, and clients know it. Certification-heavy trades cluster in the low half of the catalog for exactly this reason: notary loan signing at 2, mobile beauty care at 3, with its license boundaries per service. Appliance repair sits at 3 too, where diagnosing a failed board wrongly costs the customer real money and you the callback.
Presence is trust you cannot prompt for
Physical presence does two jobs at once. It produces the outcome, and it produces the relationship. The person the customer waves at across the lawn, the cleaner who notices the leaking angle stop under the sink and texts a photo, the detailer who remembers which regular hates tire shine: this accumulation is local trust, and it is the asset that no amount of generation can counterfeit. Our lawn care course closes on exactly this point. The work is physical, local, and judged at the curb, and the parts software touches are the parts that got easier.
There is a boundary to draw here, and I want to draw it sharply. Presence inside someone's home or on their property is deflection. Presence inside a platform's interface is not. An assistant who is "always there" in a chat window is one redesign away from being replaced by the platform's own model. A house cleaner standing in your kitchen is not. When you audit your idea, check where your presence physically happens, and who controls that room.
Licensing and regulation as slow armor
Regulation is unpopular right up until it is the reason your business survives. A state commission, a cosmetology license, a food handler certificate, a Medicare supplier number: each one is a gate the machine cannot walk through, and each one slows substitution not because the technology cannot do the task but because the law requires a person of record. Micro bakery, rated 2, sells cottage food in a frame of state cottage laws. Mobile beauty care, rated 3, is bounded by license rules per service. Even medicare billing, whose daily work is data, carries its 4 rather than a 6 because compliance responsibility has to sit with a breathing entity.
The honest flip side: regulation slows, it does not stop. Rules change. Remote online notarization already exists as a lane inside the notary course, and if states someday accept machine-assisted verification broadly, parts of that trade's protection thin out. Regulatory armor is real armor and it rusts. Score it as protection today, and calendar a re-check.
Local relationships compound
The last deflection source compounds like interest. First customers come from neighbors, referrals, the church bulletin, the mom-group chat, the next-door app, and every served customer becomes a node that vouches for you. Machines can cold-message at scale. They cannot be the neighbor two streets over whose hydrangeas you did not kill. The recurring-route trades in this catalog, cleaning, laundry pickup, pet waste removal, all rated 2, run on this loop: physical work plus local trust plus recurring schedule equals a customer who would have to actively choose to replace you with a cheaper thing that still needs a human anyway. Laundry pickup runs on the weekly bag. Pet waste removal runs on the standing Friday yard visit.
Assembling the deflection side
So the deflection checklist, the things that push a score down:
Deflection audit for your idea
None of these alone makes a business safe, and their absence alone makes it doomed. What they do is anchor the score. You now have the tools for four of the five scoring questions: the task split, the deliverable test, and this deflection audit. The one remaining question, and the one people forget until it is too late, is where the demand comes from and who owns the gate the customers walk through, and the answer in 2026 is stranger than anyone predicted.
Keep going — you're working through Will AI Eat This Business?.
All courses are free ↗