Which solo businesses are most AI-proof now?

What does “AI-proof solo business” even mean now?

The most AI-proof solo businesses today are businesses where better AI cuts the founder’s workload faster than it cuts the customer’s need to pay them.

That definition matters because simply asking which jobs AI can do gives us a moving target. The International Labour Organization examined almost 30,000 work activities and found that roughly one-quarter of workers worldwide already sit in occupations with some exposure to generative AI. Clerical jobs remain the most exposed, while improvements in voice, image and video models have pushed exposure higher in media, web and other professional work.

Yet only 3.3% of global employment landed in the ILO’s highest exposure category. Most jobs still mix tasks that software can handle with tasks that require judgment, physical action, context or responsibility.

AI agents make the picture even less comfortable for digital workers. METR’s latest evaluations show frontier agents handling increasingly long software, machine-learning and cybersecurity tasks. Those benchmarks still work best when the task is clearly specified and success can be checked objectively. Performance deteriorates once the work becomes messy, ambiguous or dependent on a broader real-world context.

Producing a tax draft and being responsible for a client's tax position are different businesses. Generating a plumbing quote and repairing a leaking pipe are different businesses. Building an appointment bot and taking responsibility for a clinic's entire intake process are different businesses.

When we call a solo business AI-proof, we therefore mean relatively resistant to economic replacement as AI improves.

Is AI already making generic freelance work less valuable?

Yes. Generic digital execution is already showing the clearest signs of price pressure from AI.

Upwork's Future Workforce Index, released in July, gives us one of the cleanest comparisons because freelance marketplaces reprice skills quickly. Contracts involving generative AI and creative production grew 90% year over year, yet earnings per contract fell 13%. Across lower-complexity AI execution work, earnings fell 28%.

More sophisticated work moved the other way. AI-augmented professional services grew 72%, with earnings up 22%. Freelancers doing more complex work with AI saw earnings rise 45%.

That is a much sharper split than the usual “AI is creating jobs and destroying jobs” discussion. Demand for AI-related work can explode while the value of individual jobs falls. More people need the output, more people can produce it, and prices get squeezed.

The vulnerable layer is easy to recognize: isolated pieces of text, simple designs, generic research, basic coding, presentations, transcription, summaries and other outputs where the client can judge the deliverable without caring much who produced it.

The latest Upwork data makes the point even clearer. In an analysis published recently, jobs asking humans to repair, check or improve AI-generated work were up 70% year over year. In software development, that type of work has grown more than eightfold since 2023; in design and creative work, it has grown almost eightfold.

The money is moving toward people who can catch mistakes, understand context and take an AI-generated first draft through the last difficult part of the job.

Type of solo work	What is happening now	AI resilience
Generic text, images or research	Supply is expanding while pricing gets squeezed	Low
AI output review and remediation	Demand is growing quickly	Medium
Complex AI-assisted professional work	Volume and earnings are both rising	High
End-to-end ownership of a business outcome	AI lowers delivery costs while responsibility remains	Very high

Are local hands-on businesses the safest solo businesses from AI?

Yes. If we care mainly about direct AI replacement, specialized local services currently sit near the top, and AI is making it easier for one person to run them efficiently.

The strongest recent evidence comes from Gusto's analysis of more than 500,000 small businesses. Gen Z founders, despite being the generation most comfortable with AI, are at least twice as likely as older owners to build in-person businesses. Their companies are 17% less exposed to AI overall.

The strictest measure is even more interesting. Only 18% of payroll at Gen Z-owned businesses goes toward roles that AI could mostly automate, compared with about 24% at companies owned by older generations.

Young founders are clearly using AI. More than 70% used it while starting their companies, and almost two-thirds said AI made the process faster and cheaper. They simply tend to build businesses where the actual customer work still happens in the physical world.

A model can answer an HVAC company's phone, qualify leads, write estimates, plan routes, chase unpaid invoices and ask customers for reviews. The air conditioner still needs someone standing in front of it.

Demand for that work also looks durable. Angi's latest survey of homeowners found that 63% of recently active homeowners had completed maintenance and 58% had completed repairs, compared with only 35% who had done renovations. Rising costs are pushing discretionary projects back, while essential work keeps moving. Sixty-six percent still expect to make a major investment in their home within five years, the highest figure Angi has recorded in this research series.

The competitive moat goes beyond physical labor. Scorpion surveyed 2,000 homeowners and found that 87% would reject a home-service provider rated below four stars. That figure had been only 64% roughly two years earlier in its comparable pest-control research. Reputation is getting more important even as AI changes how customers search.

Gusto also followed 1,593 small businesses that had adopted AI and compared them with 669 businesses that knew about AI but had chosen not to implement it. One year later, adopters had roughly 7% more employees than comparable non-adopters.

The effect was strongest among businesses with fewer than ten employees. Their headcount ended up around 10% higher.

What those firms hired matters more than the headline number. Health and social-assistance businesses added therapists and other care workers. Other businesses added teachers, technicians and cooks. The extra payroll went largely toward people producing the service customers were actually buying.

Community-service businesses with fewer than ten employees showed the strongest effect of all, with headcount around 19% higher among AI adopters.

Local customer expectations make the opportunity bigger. Scorpion found that 56% of homeowners want round-the-clock scheduling or some way to communicate after hours, while 66% of home-service businesses say after-hours service is a major challenge. AI can close much of that gap cheaply.

A solo electrician, inspector, pool specialist, appliance technician, mobile detailer or specialist cleaner can therefore run with a level of responsiveness that once required office staff.

Are tax, accounting and other regulated solo businesses safe from AI?

Specialized regulated services remain highly defensible, although the routine parts of those professions are being automated fast.

Thomson Reuters surveyed more than 1,500 professionals across legal, tax, accounting, risk, fraud and government and found organization-wide generative-AI adoption had jumped from 22% to 40% in a year. Its newer Future of Professionals research puts individual usage even higher: 74% of professionals use AI several times per week, and 44% use it several times a day.

Tax and audit are further ahead. Thomson Reuters found that 81% of professionals in those fields now use AI regularly.

The protection clearly doesn't come from accountants or lawyers refusing to automate. Their advantage comes from what remains around the automated work.

Tax filings still need correct facts. A disputed position still needs somebody who understands the client's situation. A business owner still needs someone to notice that a transaction has been categorized incorrectly, that the entity structure makes no sense or that an unusual cross-border rule applies.

Clients increasingly expect the professional to use AI as well. Thomson Reuters found that 89% of corporate tax clients consider AI-enabled quality improvements very important or essential. Nearly one-third were already reconsidering, or planning to reconsider, a professional relationship over the gap between what firms could deliver with AI and what they actually delivered.

That creates a fairly brutal outlook for commodity bookkeeping. Manual categorization, reconciliation, invoice extraction and routine reporting have weak long-term pricing power.

A specialist who owns the messy end of the problem has a stronger position: cross-border tax for online sellers, cleanup accounting after acquisitions, tax representation, healthcare compliance, regulated filings for a particular industry or financial operations for a narrow type of business.

The professional is still there, while a large chunk of yesterday's junior work happens automatically.

Solo professional model	Current AI pressure	Outlook
Manual bookkeeping	Very high	Weak
Generic tax preparation	High	Mixed
Specialist tax and compliance	Medium	Strong
Representation and dispute work	Low	Very strong
Ongoing niche financial advisory	Medium	Strong

Is AI automation consulting still a good solo business?

Yes, although generic AI setup work has a shorter shelf life than the current demand boom makes it seem.

The implementation gap is huge today. Goldman Sachs surveyed small businesses and found 76% already using AI, while only 14% had fully integrated it into their core operations. Seventy-three percent said more training or implementation support would help them.

That describes an enormous amount of unfinished work.

Upwork sees the same pattern from the hiring side. Its latest data shows rising value for people who can connect AI to actual business workflows, while basic execution work gets cheaper.

The safest version of this business is extremely specific. “AI consultant for small businesses” will become harder to sell as software gets easier. “We automate intake, estimates and follow-up for HVAC contractors” is much stronger.

The second founder learns where HVAC leads come from, which jobs are profitable, which information technicians need, how dispatch works, why customers fail to book, which integrations break and where human approval is still required.

The opportunity becomes more recurring when the founder stays responsible for performance. Building one chatbot creates a project. Running a contractor's lead-to-booking system creates an operating relationship.

Won't AI agents eventually replace the AI automation consultant too?

They will wipe out plenty of today's setup work, while specialists who own a real business process should have much more room.

Agent builders are already getting easier. Integrations are being packaged into products. Software companies are embedding agents directly into the tools businesses already use. Tasks that currently require somebody to connect APIs or configure workflows manually will become one-click features.

The valuable part is moving further into the client's business.

Someone still has to decide whether an appointment should be automated at all, which exceptions go to a human, what counts as a qualified lead, how customer data can be used, which mistakes are unacceptable and whether the system is actually improving revenue.

The recent surge in AI-remediation work on Upwork is useful here. Companies producing more work with AI are simultaneously hiring more humans to check and finish that work. Jobs focused on improving AI-generated output rose 70% year over year.

Software has one of the strongest versions of this pattern: remediation postings have increased more than eightfold since 2023.

Prompting skill will depreciate. Knowing one industry's workflows, economics, exceptions and customers has much more staying power.

Is vertical micro-SaaS still AI-proof when anyone can code now?

Vertical micro-SaaS can still be one of the best solo businesses, although code itself offers much less protection than it used to.

Gusto found that startups formed in AI-enabled industries are already operating more leanly. Twelve months after formation, one recent cohort had roughly 6% fewer employees than the previous cohort.

Stripe Atlas gives us the founder side of the story. Solo founders recently accounted for 63% of new C-corps formed through Atlas, an all-time high. AI-native solo startups also generated almost twice the revenue of other solo startups by their second year.

Building alone is becoming far more practical.

That same improvement makes mediocre software easier to attack. A founder who can build a scheduling app in a weekend has to assume that somebody else can too.

Stripe's revenue distribution makes this painfully clear. Four years ago, a top-decile Atlas solo founder generated about 34 times the revenue of the median solo founder during the first six months. That gap has widened to 61 times. Median early revenue for solo companies fell 23% year over year in the latest cohort measured, while top-decile revenue rose 19%.

Cheap building has created more founders without making average businesses better.

The vertical SaaS products we would trust most today live deep inside recurring work. Software for credential management at home-care companies, compliance records for small manufacturers, inspection workflows for a specific trade or claims documentation for one insurance niche can accumulate integrations, historical records and habits.

Stripe's solo-startup data backs that logic. B2B solo companies had more than four times the median revenue of B2C solo companies after two years, while the best B2B solo founders retained their earliest customers six times better than the median.

Retention is becoming a much stronger moat than code.

Are simple AI wrappers basically doomed?

Simple AI wrappers are one of the weakest places to build a durable solo software business today.

The test is easy: can the customer get roughly the same result by opening ChatGPT, Claude or another general-purpose AI product?

If the answer is yes, the founder sits dangerously close to the model provider.

An email rewriter, basic summarizer, generic image generator or simple research interface can still make money through convenience and distribution. The underlying feature remains easy to copy, and model companies keep absorbing popular use cases into their own products.

Workflow software has a different economic shape. Imagine a system that receives supplier invoices, extracts the details, matches them against purchase orders, flags discrepancies, emails vendors, pushes approved records into accounting software and preserves an audit trail.

AI may provide the intelligence inside that system. The customer is paying for the whole process to work.

The vulnerability comes from having almost no value beyond the underlying model call.

Can proprietary data create an AI-proof online business?

Yes. Proprietary niche data is one of the strongest fully digital assets a solo founder can own these days.

General AI systems are extremely good at explaining information they can access. That puts pressure on businesses built around repackaging public knowledge.

Unique datasets behave differently.

A founder might continuously collect industrial component prices, local planning approvals, commercial lease comparables, specialized procurement opportunities, insurance-policy changes, private-company metrics or regulatory updates from fragmented sources.

The newsletter or dashboard is only the presentation layer. The difficult asset is the collection history underneath it.

AI can make that operation dramatically cheaper. One founder can monitor far more documents, extract structured fields, flag changes, translate records and produce custom analysis for different customers.

Better models therefore increase the founder's ability to exploit the dataset while leaving competitors with the original problem of obtaining the information.

Historical depth can make the moat stronger over time. A competitor can start collecting today's prices tomorrow. It cannot instantly recreate five years of verified prices, corrections and relationships.

Among purely online solo businesses, we would rank this model unusually high.

Is e-commerce safe from AI?

E-commerce can be AI-resistant when the founder controls a real asset such as product, supply, distribution or brand; generic reselling has much weaker protection.

AI is already making storefront operations easier. Product copy, ad variations, translations, customer support, merchandising, image editing and basic analysis can all be produced cheaply.

That is useful for every merchant, including competitors.

The interesting question is what stays scarce after everyone receives those tools.

Exclusive manufacturing relationships stay scarce. Proprietary designs stay scarce. A trusted niche brand can stay scarce. An audience that repeatedly buys from the same merchant can stay scarce. A difficult sourcing network can stay scarce.

Generic product access does not.

This makes some small physical-product businesses more interesting today than they looked during the pure-software boom. A founder with an unusual supplier relationship can suddenly operate international marketing, customer service and merchandising with very little staff.

The technology compresses overhead while leaving the actual product advantage alone.

Dropshipping the same catalog thousands of other merchants can access has the opposite problem. AI lowers the barrier for every seller at once.

Can newsletters and content sites still survive AI search?

Generic information sites face severe pressure, while publications that own an audience or original information can still become excellent solo businesses.

AI assistants can already summarize public information, answer follow-up questions, compare sources and generate explanations in whatever format the reader wants. Publishers whose main advantage was organizing information more conveniently have lost a meaningful part of that advantage.

Traffic is especially vulnerable when readers arrive through one-off informational searches and have little reason to remember the publisher.

The stronger version owns something beyond the article.

A newsletter read by thousands of procurement directors can sell sponsorships, recruiting, research, events, introductions or software. An industry publication that runs original surveys can produce information unavailable elsewhere. A specialist with unusual access can publish reporting that models cannot retrieve from public pages. Personality also matters when readers actively want one person's judgment.

Content then becomes the distribution engine for a deeper asset.

Scorpion's local-business research already found 22% of homeowners using AI tools when researching providers. Search behavior is shifting outside traditional search pages even in categories where somebody ultimately hires a real-world contractor.

A solo publisher relying mainly on commodity search traffic is therefore carrying more risk today than a publisher with a direct email relationship, proprietary research or a paid community.

Is coaching or online education AI-proof?

Generic teaching is highly exposed, while coaching tied to accountability, observation or measurable performance remains much tougher to replace.

Explaining a concept used to be valuable because good explanations were scarce. AI can now generate unlimited explanations, examples, quizzes, practice exercises and customized study plans within seconds.

Charging primarily for access to information will get harder.

Customers often buy coaching for reasons that survive cheap information. A sales coach listens to actual calls and points out what the salesperson keeps doing wrong. An athletic coach watches technique. An executive coach knows the person's history and notices recurring behavior. An exam coach can impose deadlines and hold somebody to a study plan.

Those businesses revolve around observation and accountability.

The distinction changes how we would design a solo education company today. A generic recorded course has weak protection. A program that promises a measurable result, reviews the customer's real work and keeps the founder involved through completion has much more.

AI can handle explanations, exercises, summaries and preparation around that relationship, letting one coach serve more people without turning the service into a purely automated course.

Is specialist recruiting still safe from AI?

Niche relationship-driven recruiting still has a defensible core, while generic sourcing and outreach are being squeezed quickly.

Gusto's latest graduate-hiring research shows how fast recruitment itself is changing. Recruiter was one of the more common entry-level titles several years ago; it has now fallen outside the top 20 in its dataset. Software engineer, financial analyst and sales-development representative have also lost share, while hands-on roles such as field manager and service technician have gained ground.

Recruiting contains many tasks AI handles well: finding profiles, screening CVs, drafting outreach, scheduling interviews and comparing candidates against a job description.

A recruiter whose advantage mainly consists of searching LinkedIn and sending messages has less to sell today.

Executive search and deeply specialized recruiting have a stronger asset. A recruiter may know which semiconductor engineers are quietly willing to move, which CFOs have actually handled a turnaround, what compensation will attract a particular candidate and which companies have reputations that candidates privately avoid.

Those relationships take years to accumulate.

AI will make a good niche recruiter much more productive. It will also expose recruiters whose apparent network was mostly access to the same databases everyone else uses.

When will customers still insist on having a human involved?

Customers keep paying humans when an error has consequences and they want somebody clearly responsible for getting the outcome right.

We see this most clearly in professional services. AI use is already mainstream among lawyers, accountants, tax specialists and other professionals, yet clients are pushing firms to use AI more effectively rather than walking away from professional help altogether.

Thomson Reuters found organization-wide AI adoption in professional services nearly doubled in one year. It also found only 18% of organizations tracking AI return on investment. Adoption has moved faster than confidence in how these systems should actually be managed.

Tax and audit show the same tension at a more advanced stage. Eighty-one percent of professionals use AI regularly, while 35% told Thomson Reuters they were using tools their firm had not authorized. That creates obvious problems around confidentiality, audit trails and liability.

The more consequential the decision, the more valuable a clear chain of responsibility becomes.

Local services show another version of this. Homeowners increasingly use AI to discover contractors, yet 87% still refuse to hire businesses with ratings below four stars. AI can influence which provider enters the shortlist. Reputation still decides who gets trusted with the house.

Responsibility is therefore becoming one of the better economic moats for solo operators. Customers will happily let software do more of the work while still wanting a person or business whose name sits on the result.

Which solo businesses actually get stronger when AI improves?

The best AI-proof businesses are those where each jump in AI capability removes cost from the founder while leaving the scarce part of the business intact.

The newest Gusto evidence gives us a real-world example. Businesses that adopted AI subsequently hired about 7% more than comparable businesses that knew about AI but did not use it. Among firms with fewer than ten employees, the difference reached roughly 10%.

The new workers were largely customer-facing doers.

That is exactly the pattern a solo founder should want. AI handles more scheduling, documentation, research, marketing and administration; the founder gets more capacity for the thing customers genuinely pay for.

The same logic applies online. A proprietary-data business can use better AI to process more information. A tax specialist can research and prepare cases faster. A vertical software founder can build features with less engineering labor. A niche recruiter can research a market faster.

Generic output businesses face the reverse effect. Better AI increases the amount of competing supply while customers gain cheaper substitutes.

Solo business	What improving AI changes	Overall effect
Essential local service	Automates the back office	Strongly positive
Specialized regulated service	Compresses research and preparation	Positive
Vertical AI implementation	Automates some setup while expanding capabilities	Positive, with pressure to specialize
Proprietary data business	Makes unique information cheaper to process	Strongly positive
Vertical micro-SaaS	Makes development easier for founder and competitors	Positive only with workflow depth
Specialist recruiting	Automates sourcing while increasing recruiter capacity	Mixed to positive
Generic digital agency	Makes competitors dramatically more productive	Negative
Commodity content business	Makes the core output nearly free	Strongly negative

So which solo businesses are most AI-proof now?

The strongest solo businesses today are specialized services and products where customers still need physical execution, accountability, proprietary access, embedded workflows or trusted relationships after AI has done everything it can.

We would put essential local services near the top: repair, maintenance, inspection, specialized installation and other jobs where somebody eventually has to show up. Their resilience becomes even more interesting when AI can run much of the office around the technician.

Specialized regulated businesses belong in the same top tier. Tax, compliance, accounting and similar services are automating aggressively, but specialists who own unusual cases and accept responsibility for the result should be able to run increasingly lean practices.

Proprietary niche-data businesses are probably the strongest fully online model. Better AI helps the founder collect and exploit the information without giving competitors the underlying dataset.

Vertical AI implementation also looks excellent currently, particularly when the founder owns an outcome for one industry instead of selling generic AI expertise. Some of today's implementation work will disappear into software products, so specialization matters a lot here.

Vertical micro-SaaS remains attractive. The bar has risen sharply because building software is getting cheaper for everyone. We would want recurring workflow, integrations, historical data and genuine switching costs before calling a product defensible.

Relationship-heavy businesses such as specialist recruiting and brokerage sit below those groups but can still be strong when the network is real. High-accountability coaching can work for similar reasons.

Generic content production, routine digital agencies, basic bookkeeping, simple AI wrappers, undifferentiated online courses, dropshipping and informational websites sit much closer to the danger zone.

Stripe's recent solo-founder numbers capture the wider environment. Solo founders now represent 63% of new C-corps formed through Atlas, yet the gap between the top decile and median solo founder has widened from 34 times to 61 times early revenue in four years. Starting alone has become easier while building something customers genuinely care about has become more competitive.

The most durable solo company does not need to hide from AI. It needs to own a scarce part of the transaction that survives after AI has made everything around it radically cheaper.