What AI Won’t Replace in a Local Service Business

Revised August 22, 2026

Picture a Tuesday in February. You are on your back under a house in Fenton with about eight inches of clearance, phone light in your teeth, working out why a supply line that was fine in November is weeping at a joint somebody soldered before you were born. Your phone has four texts on it. Two are from a customer in Spanish Lake who wants to know, politely, whether she really needs the whole run replaced or whether you are padding it.

You get out from under the house. You sit in the truck with a gas station sandwich, and the phone hands you a video: a man in a rented office explaining that artificial intelligence is about to make your trade obsolete, and also that you can clear $1,000 a day with it, and also that his course is $997 until midnight. You scroll. The next one says the same thing in a different room.

This is for the person who has to give a real answer to that — a plumber in Fenton, a stylist in Spanish Lake, a veterinarian or a general contractor bidding a job over in Troy in Madison County. Not a comforting answer. An accurate one. There is a real displacement happening inside a local service business, and there is also an industry selling you a story about it, and those two things are not the same size.

Which business can AI not replace?

A business whose product requires a licensed human to be physically present and legally accountable. That is not a claim about how capable the software is. It is a claim about who signs the permit, who carries the liability policy, and who a state board can discipline afterward. Software holds none of those three, so the work stays with a person.

It is worth being precise, because “AI can’t replace the trades” gets repeated as a comfort blanket and it is only about two-thirds true. Run your own work through three questions.

• Does somebody have to be there? Not on a screen. In the crawlspace, on the roof, hands on the animal, scissors in someone’s hair.
• Does the state issue a credential to a human being for this? A license, a registration, a certificate with a person’s name on it.
• Is somebody answerable a year later? When the joint fails, when the color goes green, when the dog is worse instead of better.

Work that answers yes to all three is structurally protected, and the protection has very little to do with the technology being bad at it. Work that answers no to all three — writing the estimate, answering the same question for the fortieth time, chasing a schedule — is exposed right now, today, in your business. Almost every local service company is a mix of the two, and the mix is the whole story.

Why is licensed work a legal problem before it is a technology problem?

A license is issued to a human being, not to a process. That single fact does more work than any argument about what models can and cannot do. In Illinois you cannot legally perform plumbing work of any kind without a license from the Illinois Department of Public Health, and the route to it runs years of documented apprentice hours under a licensed plumber before you sit the state exam.

Cross the river and the rule changes shape rather than disappearing. Missouri has no statewide plumbing license at all — plumbing is licensed by cities and counties, so a card issued in one jurisdiction does not travel to the next one. That is its own moat, incidentally. A licensing regime stitched together out of municipalities is a local-knowledge business by construction.

Roofing splits along the same river, and it is worth knowing which side you are standing on. Illinois licenses roofing contractors statewide under the Illinois Roofing Industry Licensing Act; 225 ILCS 335/9 makes it unlawful to hold yourself out as a roofing contractor without a license from the Illinois Department of Financial and Professional Regulation, and the license requires an examination, liability insurance, workers’ compensation and a surety bond — $10,000 for a limited license, $25,000 for an unlimited one. Missouri does it by registration instead. Since January 1, 2023, a roofing contractor working in Missouri must hold a certificate of registration from the state, and to get one you file proof of liability insurance, a current workers’ compensation policy and motor vehicle insurance.

Personal-service trades are no softer. A cosmetology license requires 1,500 hours of approved training in Missouri and 1,500 hours in Illinois before you are permitted to sit for the exam. Fifteen hundred hours, attached to a named person, verified by a school.

Veterinary medicine puts the whole point into one sentence of statute. Missouri’s definition of a veterinarian-client-patient relationship, at RSMo 340.200, requires that the veterinarian has recently seen the animal or made medically appropriate and timely visits to the premises where it is kept, and has assumed responsibility for the medical judgments. Assumed responsibility. That is a person accepting consequences, written into law, as the precondition for the practice existing at all.

Read those four together and the pattern is not “AI is bad at pipes.” The pattern is that the statute names a person, requires that person to have been present, and holds that person answerable. Even in a version of events where a model diagnoses better than you do, a licensed human still has to own the output, because owning the output is what the license is. Changing that is not a software release. It is fifty state legislatures and the insurance industry behind them.

Somebody has to be accountable when it goes wrong

Your customer is buying two things and only one of them is the work. The other is recourse. If the water heater floods a finished basement in March, there is a name on the permit, a policy number, sometimes a bond, and a state board that will take a written complaint. That entire apparatus assumes a specific human being can be found and held to it.

Notice how much of licensing is really insurance wearing a different hat. Missouri’s roofing registration requires proof of liability coverage, a current workers’ compensation policy and motor vehicle insurance before the certificate issues. Illinois stacks a surety bond on top of the insurance. An underwriter prices a risk it can attach to a person with a loss history and an address. No carrier writes a general liability policy on a language model, and no licensing board has the authority to suspend one.

Which is also why “the software told me to” will not function as a defense in a complaint hearing or a claim. If you run an estimate, a diagnosis or a treatment plan through a tool and hand the result to a customer, you have adopted it. The liability did not move. It cannot move, because there is nowhere for it to go. That asymmetry is permanent in a way that no capability benchmark is: you can be sued and the tool cannot.

The broader question of where these tools fit in a local business at all is covered in AI for St. Louis: how local businesses can actually use it. This page is about the narrower thing sitting underneath it — who the law can actually hold responsible, and why that is the line that does not move.

Where is AI genuinely taking work in a service business?

The exposed work in a local service business is the paperwork around the work, and there is far more of it than most owners like to say out loud. An owner who insists nothing is exposed is making the same error as the one who panics, just in a more flattering direction.

• Drafting. Estimates and proposals built from scrappy notes, job descriptions, letters to an insurer, the follow-up email you have been avoiding since Thursday.
• Summarizing. A forty-minute call, a warranty document, six months of maintenance history before you walk into the building.
• First-pass customer support. Hours, service area, do you work on that brand, do you take that insurance, are you licensed in Illinois too.
• Scheduling and dispatch admin. Confirmations, reminders, the reshuffle when a morning job runs three hours long.
• Marketing copy. Service page text, review responses, the seasonal post you have not written since spring.

If a person on your payroll does mostly those five things, that role is changing already. Say so honestly to yourself, then decide what you want that person doing instead, because the good version of this is usually redeployment rather than a headcount cut — the constraint in most service shops is billable hours, not admin hours.

The most-cited academic look at task exposure is by Tyna Eloundou, Sam Manning, Pamela Mishkin and Daniel Rock, first posted in March 2023 and published in Science in 2024. Worth stating plainly: several of the authors were at OpenAI, so read it as interested research rather than neutral measurement. Its headline estimate is that around 80% of the US workforce could have at least 10% of their work tasks affected by large language models, and roughly 19% could see at least half their tasks affected. The more useful detail sits at the other end of the distribution — the occupations that scored at the absolute floor, with no measured exposure, were the physically embodied ones: roofers, construction laborers, grounds maintenance workers.

Adoption is a separate question from capability, and the adoption numbers run well below the noise. The Census Bureau’s Business Trends and Outlook Survey, published May 26, 2026 and covering December 14, 2025 through May 3, 2026, put AI use at under 20% among firms with four or fewer employees, 32% among firms with 100 to 249 employees, and 37% among firms with 250 or more. A Federal Reserve note dated April 3, 2026 put it at about 18% of firms as of the end of 2025.

One warning about any trend line you are shown on this. In November 2025 the Census changed its question from whether a firm uses AI to produce goods or services to whether it uses AI in any business function, and reported rates rose between 47% and 159% depending on the industry. Any chart crossing that date is measuring a rewritten question as much as a change in behavior. Under the older wording, firms under 250 employees came in at 8.8% in August 2025, up from 6.3% six months before. The full mess of those numbers is unpacked in what percent of small businesses actually use AI.

And there is a figure that appears in no survey: what your own staff are already doing with these tools on their own initiative, on their own phones, without mentioning it. That one is worth knowing about before a customer finds out for you, and it is the subject of your staff is already using AI.

Wildly off-topic — those storm limbs hanging over the driveway: tree services.

What about the “make $1,000 a day with AI” videos?

They are selling the course. That is the answer, and it is not cynicism — it is the business model. The revenue comes from people buying the method, not from people running it, which is why the pitch is always urgent and the price always expires at midnight.

The Federal Trade Commission has been working this in public. On September 25, 2024 it announced Operation AI Comply, five enforcement actions against companies making deceptive AI claims; three of the five were business-opportunity schemes promising AI-powered income. The most instructive case came after. On March 18, 2025 the FTC sued the operators of Click Profit, alleging they charged consumers a management fee of at least $45,000 — plus thousands more for inventory — to run “AI-powered” storefronts on Amazon, Walmart and TikTok, and claimed supplier partnerships with companies including Nike and Disney that they did not have. The FTC put consumer losses at at least $14 million. In August 2025 the operators settled and were permanently banned from the industry, with a $13.6 million judgment against one group of defendants.

These persist for three unglamorous reasons. The claim is unfalsifiable to somebody who has never tried it. The product is information, so the margin on the ten thousandth sale is identical to the first. And when it does not work, the buyer’s most natural explanation is that he failed to follow it properly — so he does not complain, and a meaningful share of buyers purchase the next one.

There is something concrete you can do with this. The FTC’s Business Opportunity Rule, at 16 CFR Part 437, requires a seller of a business opportunity to hand you a one-page disclosure document at least seven calendar days before you sign anything or pay anything. If the seller has stated or implied what you can earn, they must also give you a separate Earnings Claim Statement showing the start and end dates of those earnings and the number and percentage of buyers who actually earned at least that much. Ask for both, in writing. A legitimate seller has them ready. One caveat worth knowing: a plain online course often does not meet the Rule’s definition of a business opportunity, which is part of the reason so much of this is packaged as a course.

None of that makes learning the tools a scam. The distinction is whether you are being sold a skill or a rented outcome. A skill you keep, and you can test it on a Tuesday against work you already understand. That is the whole idea behind the AI workshops — sit down with the tools, on your own jobs, and find out where they help and where they quietly get things wrong.

The judgment nobody sells a course on

The hardest parts of a trade to automate are also the parts nobody talks about, because they are the parts that were never written down. There are four, and every experienced operator will recognize all of them.

Diagnosing an unfamiliar problem in an old house. Much of this metro is century-old brick with three additions and four generations of plumbing stacked in one wall. The skill is not knowing the code; the code is a lookup. The skill is holding six plausible explanations at once and picking the one that is true in this house, from a smell, a stain pattern, the age of the fittings, the direction the floor slopes, and one sentence the homeowner said about last winter. It is also knowing when the honest answer is that you have to open the wall to find out, and saying that out loud instead of guessing confidently. A model is built to produce a confident answer from whatever it was given. You are allowed to say you do not know yet.

Reading a person. A customer goes quiet, and the silence means they do not understand and are embarrassed to say so. Another says “do whatever you think” and means “I cannot afford this and I am hoping you will notice.” A pet owner asks a question about cost that is actually a question about whether the animal is going to die. You have the room, the face and the pause. A tool has the transcript.

Deciding what not to sell. This is the least discussed and the most valuable. Telling somebody the $400 repair will hold four more years, when you could have written $6,000 of replacement that same afternoon, is a deliberate reduction in this month’s revenue in exchange for a referral base you cannot see on a spreadsheet. Every recommendation system ever built optimizes the transaction in front of it, because the transaction in front of it is the only thing it can observe. An operator with twenty years in one set of zip codes is optimizing something else entirely.

Improvising when the plan meets reality. The joist is in the wrong place. The part is discontinued and the supply house is closed. The chair in front of you has a home color job on it that nothing in the training covered. Every trade is really a long sequence of small recoveries, and the recovery is where the expertise lives. None of it is documented anywhere for a model to have read.

What this looks like from your side of the counter

Most owners in the service trades know the economics better than they say out loud. The trip or diagnostic fee is close to a loss leader — it exists to get the truck to the door, and the money is made on planned work, the labor rate and parts margin, not on the 11pm emergency. Two weeks of weather can make a quarter: the first hard freeze, the first run of 95-degree days. February and October are the months you stare at a thin schedule. One callback eats a job’s profit and the second one eats the next job too. And the complaint owners actually make to each other is never about the work — it is the six-quote price shopper, the no-show, and the ninety-minute estimate that went to a guy working out of a trunk with no insurance.

The one thing AI genuinely changes in that picture is how people find you. They ask an assistant, in plain sentences, for somebody nearby who does this — a shift explained in Google AI Mode, explained for a local business owner. Being listed, complete and consistent is how you turn up in that answer instead of the guy with the trunk. If you are not on St Louis Near Me Directory yet, you can add your business.

Frequently asked questions

What jobs will be gone by 2030?

Nobody can tell you honestly, and a source that names a year and a percentage is usually selling something. What is measurable now is task exposure, not job elimination. In the 2024 Science paper by Eloundou and colleagues, the occupations scoring zero exposure to large language models were the physically embodied ones — roofers, construction laborers, grounds maintenance workers. That is a description of today’s task mix, not a forecast.

What 5 jobs will survive AI?

The framing is off, though the instinct behind it is sound. Instead of five job titles, use three tests: does somebody have to be physically present, does a state issue a license to a named human for the work, and is that human answerable a year later. Licensed trades, veterinary medicine, cosmetology, nursing and skilled construction pass all three. Office work built on drafting and summarizing passes none of them.

What are 10 jobs that AI can’t replace?

Lists like this circulate constantly and most are guesswork. A defensible version is any occupation where the law requires a licensed person on site: plumbers, electricians, HVAC technicians, roofers, veterinarians, dentists, nurses, cosmetologists and barbers, home inspectors, and heavy equipment operators. The common thread is not difficulty. It is that a statute names a person who must be present, carry insurance, and can be disciplined.

What jobs will still be around in 10 years?

That is a prediction, and this page will not make one. The structural answer is more useful anyway. Work stays with humans where the law requires a licensed individual, where the task is physical, and where somebody has to carry insurance and answer for the result afterward. Those requirements sit in state statute and insurance underwriting, and they move at the speed of legislatures, not software.

What will be AI after 10 years?

Unknowable, and worth being suspicious of anyone who claims otherwise, particularly if a course is attached to the claim. What is checkable is current adoption. The Census Bureau’s Business Trends and Outlook Survey, published May 26, 2026, put AI use under 20% among firms with four or fewer employees and at 37% among firms with 250 or more. Plan against measured numbers.

What job is most safe from AI?

By the task-exposure research, the most insulated work is the most physically embodied. In the 2024 Science study by Eloundou and colleagues, the occupations that scored at the absolute floor for exposure to large language models were manual ones such as roofers, construction laborers and grounds maintenance workers. Add a state license and personal legal liability on top of physical presence and you have the most protected combination available.

Which jobs will AI replace in 5 years?

No credible source can name them with a date attached, and the ones that try are usually vendor marketing rather than research. The honest version of the question is which tasks are exposed today: drafting, summarizing, first-pass customer support, scheduling admin and marketing copy. If a role in your business is mostly those five, it is changing already. If it requires a person in a crawlspace, it is not.

Which jobs will not survive AI?

Nobody can list them responsibly, and the dated versions people quote are modeling exercises reported as findings. What can be said is narrower. Roles built almost entirely on producing routine text, summarizing documents and answering repetitive first-line questions are exposed to tools that exist right now. Roles requiring physical presence, a state license and personal liability are not exposed in the same way.

What job is most at risk from AI?

Judged by measured task overlap rather than headlines, the most exposed roles are text-heavy office ones. The 2024 Science paper by Eloundou and colleagues scored medical transcriptionists, customer service representatives, general office clerks and accountants among the highest-exposure occupations. Higher-paid, cognitive, screen-based work came out more exposed than manual work, which reverses the assumption most people start with.

About the Author: The St Louis Near Me Directory Team
Written by a dedicated team of St. Louis locals who live, work, and play right here in the St. Louis metro. Founder Lane Forman and team are committed to building the region’s most trusted directory by verifying listings and connecting local businesses with loyal customers across Missouri and Illinois.

Follow us:
Facebook LinkedIn X Pinterest YouTube

Does AI cite your expertise?

Run Your Free AI Audit

No credit card. No obligation.