Short answer: most small employers are no longer deciding whether to use AI agents, they are already using them, and the gap that matters now is oversight. The Small Business & Entrepreneurship Council’s October 2026 Check Up Survey found that 58% of small employers use AI agents regularly or occasionally. A Pax8 survey released four days later found that 35% of AI-engaged small and midsized businesses discovered AI tools, workflows or agents running without approval in the past six months.
If you run a roofing company, an HVAC shop, a dental practice, a lawn care crew, a restaurant, or a retail store in Northeast Indiana, that second number is the one to sit with. It does not describe a company that ignored AI. It describes a company where AI showed up in the office faster than anyone wrote down who owns it, what data it can see, and who checks what it says.
This post walks through what the two surveys and a St. Louis Fed tracker actually report, what the numbers suggest about cost and time, and a one-page setup a small team can finish in a week. We stick to what the sources say and mark where we are offering our own judgment. We also flag the limits of these surveys, because two of the three come from groups with a stake in how the story is told.
Key Takeaways
- SBE Council’s October 8, 2026 survey of 516 small business owners found 58% use AI agents regularly or occasionally, with another 14% considering or planning to.
- Pax8’s Q3 2026 SMB AI Pulse (October 12, 2026, 1,150 leaders in four countries) found 35% of AI-engaged businesses discovered shadow AI in the past six months.
- In the same Pax8 report, 67% of AI-engaged businesses have built at least one AI solution now in regular use, and 46% are building or testing agents that complete multi-step tasks on their own.
- Oversight is thin: in the United States, 36% of AI builders have formal, documented review, and 41% rely on consistent but undocumented review.
- Costs are hard to predict: 49% of businesses that paid for AI got at least one bill higher than expected (57% in the United States), and 55% of decision-makers spend three or more hours a week checking AI output.
- The St. Louis Fed’s tracker shows worker use of generative AI rising from 28.2% to 39.2% since Q3 2024, with self-reported time saved up from 1.6% to 2.2% of work hours.
- The fix is small and concrete: an inventory, an owner for every tool, a data rule, a review habit, and a spending cap.
How Many Small Businesses Are Really Running AI Agents?
The most direct number comes from the SBE Council’s Small Business Check Up Survey, released October 8, 2026. TechnoMetrica Marketplace Intelligence ran it online from September 11 to 19, 2026 among 516 owners of businesses with 2 to 99 employees, with a credibility interval of plus or minus 4.4 percentage points. Among its findings: 58% of small employers use AI agents regularly or occasionally, and another 14% are considering or planning their use.
The same survey asked owners whose financial performance improved or held steady what helped. Twenty-eight percent cited adopting technologies such as AI, behind higher or stable sales (38%) and cost cutting (31%). That is a self-described reason, not proof that AI caused the result, but it tells you how owners themselves are framing the choice. The survey also reports that only 16% of owners say AI is eliminating more jobs than it creates, which fits the framing SBE Council’s own release uses: AI is creating, transforming and redesigning jobs.
A broader view comes from the Federal Reserve Bank of St. Louis. Its FRED blog describes a Generative AI Adoption Tracker built by economists Alexander Bick, Adam Blandin and David Deming from the Real-Time Population Survey, a nationally representative online survey of adults aged 18 to 64. The share of employed adults who used generative AI for work in the past week grew from 28.2% in Q3 2024 to 39.2% by Q2 2026. The share of work hours assisted by generative AI grew from 4.1% to 6.3%, and the share of work hours saved grew from 1.6% to 2.2%. The St. Louis Fed notes the time-saved figure is self-reported, so it is “inherently approximate.”
Put those together and the picture is consistent without being precise. Use is widespread, individual use is deepening, and the measured time savings are real but modest. Two and a half hours saved per hundred worked is not nothing, but it is a long way from the “replace a department” claims that float around social media. If you want the sober version of that argument, our earlier post on what Gartner’s agentic AI failure forecast means for small businesses makes the same case from the project-failure side.
What Is Shadow AI, and Why Did 35% of Firms Find It?
Pax8 released the Q3 2026 Pax8 SMB AI Pulse Report on October 12, 2026. It surveyed 1,150 business owners, executives, functional leaders and managers involved in technology decisions at companies with 5 to 499 employees in the United States (400 respondents), the United Kingdom (350), Germany (200) and Sweden (200). Pax8 defines shadow AI as “tools, workflows or agents operating without official approval,” and reports that 35% of AI-engaged businesses discovered some in the past six months.
The word “discovered” is the important one. Nobody set out to run AI nobody approved. Someone found it. In a company of eight people, that usually looks like one of four things: an employee pasting customer emails into a free chatbot to draft replies, a booking or review tool whose vendor quietly added an AI assistant, a scheduling automation a helpful manager built over a weekend, or the owner’s own experiment that never made it onto any list. None of these are malicious. All of them can touch customer data.
The report also shows how loosely most firms check what they build. Formal, documented review of internally built AI ranges from 23% in the United Kingdom to 45% in Sweden. In the United States, where 81% of AI-engaged businesses have built solutions now in use, 36% of builders have formal, documented review and another 41% rely on consistent but undocumented review. That middle group is the one to watch. “Consistent but undocumented” means a particular person remembers to look, and that works right up until the day they are on vacation, quit, or get busy during your spring rush.
Pax8 also lists the top-cited risk of employee-built AI in every market it surveyed: data security or privacy exposure, named by 30% to 35% of respondents depending on the country. For a dental or medical practice, that risk is not abstract. Patient names, appointment details and insurance questions flow through the front desk every day, and where that information ends up when someone pastes it into a tool is a question you need to be able to answer. We are not offering legal advice here. If you handle patient or other regulated information, ask your compliance advisor before any AI tool touches it.
Why Are Small Businesses Building Their Own AI?
The headline claim of the Pax8 report is that small businesses “are no longer just using AI tools, they are building them.” Among AI-engaged businesses, 67% have built one or more AI solutions now in regular use and another 19% are prototyping. Only 13% rely exclusively on off-the-shelf tools. Among those who are building, 52% are developing automated workflows that connect AI to other business systems, 46% are building or testing AI agents that complete multi-step tasks on their own, and 39% have built custom integrations using an AI provider’s API.
Pressure explains some of it. In the same survey, 66% of small businesses say their business will not remain competitive without AI in the next three years, and 65% say they need to move immediately or within six months. Pax8’s own framing of what that means is worth repeating, because it is the part owners tend to skip: a business that deploys an agent or connects AI to its finance system takes on work that used to belong to software vendors, namely maintaining what it built, securing it and paying for the compute it consumes.
Our reading, and it is judgment rather than data: this is the right instinct pointed at the wrong place. Wanting AI to connect to your calendar, your invoices and your phone is sensible. The risk is that each connection becomes its own little project with its own login, its own permissions and no owner. The diagram above shows the shape we would rather see: every agent passes through one approval point before it touches the systems your business depends on. That single gate is nothing more than the habit of writing it down and naming a person.
The report adds one more finding that helps owners who feel behind. Businesses that use AI extensively are more likely to feel ahead of competitors than those still experimenting, including 63% versus 40% in the United States. That is a feeling, not a measured result, and it comes from a survey of people already invested in the answer. Still, it suggests the payoff comes from using AI deeply in a few places, not lightly in many. For most of our buyers, that argues for picking one workflow and doing it well. Our post on where AI agents get stuck on Fort Wayne business websites shows what that groundwork looks like on the customer-facing side.
What Does Unapproved or Unwatched AI Actually Cost?
The Pax8 numbers point to three costs, and none of them is the subscription price on the sign-up page.
The surprise bill. Among businesses that paid for AI in the past year, 49% received at least one bill or usage charge higher than they expected. The share rises to 57% in the United States and 65% in Sweden. Pax8 says usage-based charges are the hardest to forecast and that they grow as businesses build more agents and workflows. If an agent can run on its own, it can also spend on its own.
The checking tax. Sixty-nine percent of AI-engaged businesses say AI has freed up at least some capacity. But 55% of decision-makers personally spend three or more hours a week reviewing, editing or fact-checking AI-generated work, and 7% spend ten hours or more. For an owner-operator, those are the same hours you would otherwise spend quoting jobs or working the schedule. A tool that saves your office manager two hours and costs you three hours of review is not saving the business anything.
The control premium. Seventy percent of surveyed small businesses say they would pay at least 6% more for greater control over how their data is stored and used. That is a stated preference, not a purchase, but it shows where owners expect the trouble to come from. When your data lives inside a tool nobody approved, you cannot say where it is stored, who can read it, or how to get it out if you stop using the tool. A short list of approved tools and what each can see fixes most of that.
None of this argues against using AI. It argues for the same basic discipline you already apply to a new truck, a new software subscription, or a new hire: someone is responsible, there is a budget, and you check the work. If your marketing runs through paid ads, the same logic applies there, as we covered in why PPC AI agents fail without your business data.
What Does This Look Like in a Northeast Indiana Shop?
The surveys cover the whole country and, in Pax8’s case, several others, so none of them tells you what is happening in Auburn, Kendallville or Fort Wayne specifically. What follows are illustrations, not case studies and not client results. They show where the survey patterns tend to land in the kinds of businesses we work with.
- A lawn care or landscaping crew. A crew lead starts using a free chatbot to write estimate follow-ups and pastes in the customer’s name, address and what they asked for. Nothing is wrong with the follow-up. The question is whether the owner knows that customer addresses are now sitting in an account the owner has never seen.
- An HVAC or plumbing shop. The office manager turns on the AI answering feature that came with the phone system, and it begins booking service calls on its own. It may be working well. Nobody has written down what it is allowed to promise about pricing or arrival windows.
- A dental or medical practice. The front desk uses a writing assistant to reword insurance and billing letters. The letters read better. The practice cannot say whether any patient detail went into the prompt.
- A restaurant. A manager sets up an automation that drafts replies to reviews. Replies go out in the owner’s name, and the owner finds out when a customer mentions one.
- A retail store. A vendor’s inventory app adds AI-driven reorder suggestions that quietly start placing draft orders. Usage-based pricing means the first sign is the invoice.
Each of these is the Pax8 pattern in miniature: a reasonable tool, a helpful person, no record. And each is fixable in an afternoon with the setup below. If you want a primer on the customer-facing side of AI, our lightweight AI governance framework for SEO content shows the same approach applied to marketing copy.
What Is a One-Page AI Setup You Can Finish This Week?
This is our own checklist, built from the gaps the surveys describe. It is not a compliance framework and it does not replace advice from your attorney or insurer. It is meant to get a small team from “we think we use a few AI things” to “we know what we use and who owns it.”
The One-Page AI Setup
- Run an amnesty inventory. Ask everyone, including yourself, to list every AI tool, assistant, automation and agent they use, plus any vendor feature with “AI” in its name. Say plainly that nobody is in trouble for what shows up. The goal is a complete list, not a confession.
- Give every item one owner. One named person is responsible for each tool: who can change its settings, who pays for it, and who answers if it goes wrong.
- Write a data rule in three lines. What may never be entered (for example patient details, payment card numbers, passwords), what needs approval, and what is fine. Keep it short enough to remember.
- Mark what acts on its own. Separate tools that suggest from agents that send, book, buy or post. Anything that acts on its own gets a human check at the start and a spot check afterward.
- Set a review habit and write it down. Pick who checks the output, how often, and what a failure looks like. This turns “consistent but undocumented” into documented.
- Cap the spend. Use monthly limits wherever the tool allows them, and put the first two invoices on your calendar to read line by line.
- Pick one workflow to prove. Choose a single task, record the hours it takes today, and compare after 30 days. Include your review time in the result.
- Re-run the inventory quarterly. Vendors add AI features by default, so the list will drift.
If you handle regulated information, such as patient records, have your compliance advisor review steps 3 and 4 before you finalize them.
Step 7 is where we would put most of your effort. Button Block’s workflow optimization service starts from the same idea: map the current workflow, find the bottleneck, define success metrics up front, and automate one thing before you automate ten. Whether you do that with us or alone, measuring hours before and after is what separates a useful agent from an expensive habit.
How Much Should You Trust These Numbers?
Three cautions, so the takeaways stay proportionate.
The sources have different samples and different incentives. The SBE Council is an advocacy organization and its poll covers 516 owners of firms with 2 to 99 employees. Pax8 is a marketplace that sells AI and cloud products to small businesses through IT partners, and its release itself says the findings show why small businesses need trusted partners. Its survey covers 1,150 leaders at companies with 5 to 499 employees in four countries, and Pax8 says results for smaller subgroups should be read as directional. The two surveys also define their groups differently, so 58% and 67% are not measuring the same thing and should not be compared.
Self-reporting runs through all of it. The St. Louis Fed says plainly that time saved is self-reported and approximate. Owners who say they use AI agents may mean a chatbot with a few automations, or a fully autonomous workflow. Neither survey tells you how well any of it works.
Nothing here is local. We found no Northeast Indiana-specific figures for AI agent use among small businesses, and we are not going to invent any. What the surveys can do is tell you what to check at your own shop, and the inventory above does that for free.
Get Your AI Inventory and First Workflow Mapped
Button Block helps Northeast Indiana businesses sort out what AI they already use, choose one workflow worth automating, and set it up with an owner, a data rule and a review habit built in. Start with a conversation about what is running in your business today.
Frequently Asked Questions
- What is an AI agent, compared with a chatbot or a regular AI tool?
- An AI agent is software that completes a multi-step task on its own instead of answering one question at a time. Pax8's October 2026 report draws the same line: it separates people who use AI tools from businesses that are building agents that complete multi-step tasks, building automated workflows that connect AI to other business systems, and writing custom integrations through an AI provider's API. For a small business, the practical difference is that an agent can touch your calendar, your invoices, or your customer messages, which is why it needs an owner and a boundary.
- What is shadow AI?
- Pax8 defines shadow AI as tools, workflows or agents operating without official approval. In its Q3 2026 survey, 35% of AI-engaged small and midsized businesses said they discovered some in the past six months. In a small company this is rarely sabotage. It is usually a helpful employee, a vendor feature that switched on by default, or an owner's own weekend experiment that nobody wrote down.
- Do I need a formal AI policy if I only have a few employees?
- You need something written, but it can fit on one page. The survey data points to why: in the United States, Pax8 found that 36% of AI builders have formal, documented review and another 41% rely on consistent but undocumented review. Undocumented review works until the person who does it is out sick. A one-page list of what tools you use, who owns each one, what data each can see, and who checks the output gives you the core of a policy without a consultant.
- How much time should I expect to spend checking AI output?
- Plan for real time. Pax8 reports that 55% of decision-makers personally spend three or more hours per week reviewing, editing or fact-checking AI-generated work, and 7% spend ten hours or more. The same report says 69% of AI-engaged businesses have freed up at least some capacity, so the savings are real, but they are net of checking. If you are the one reviewing, count those hours when you decide whether a workflow is worth keeping.
- Why do some AI bills come in higher than expected?
- Usage-based pricing is hard to forecast. Pax8 found that 49% of businesses that paid for AI in the past year received at least one bill or usage charge higher than they expected, rising to 57% in the United States, and it notes the charges grow as businesses build more agents and workflows. Ask every vendor whether pricing is flat or per use, set a monthly spending cap where the tool allows one, and check the first two invoices line by line.
- Are these survey numbers reliable enough to act on?
- Treat them as directional. The SBE Council poll covers 516 owners of businesses with 2 to 99 employees and carries a credibility interval of plus or minus 4.4 points. Pax8's survey covers 1,150 leaders at companies with 5 to 499 employees across four countries, and Pax8 sells technology to small businesses through IT partners, so it has an interest in the story. The St. Louis Fed's time-saved figures are self-reported. None of this shows what is happening at your shop, which is why the inventory in this post matters more than any statistic.
- Can Button Block help us set this up?
- Yes. Button Block works with Northeast Indiana businesses on AI workflows and automation, and the first step is the same as the one in this post: list what you already use, pick one workflow worth keeping, and define how you will know it is working. You can start with a conversation on our contact page.
Sources & Further Reading
- Small Business & Entrepreneurship Council: sbecouncil.org/2026/10/08/new-small-business-check-up-survey: New Small Business Check Up Survey, October 8, 2026
- Pax8 via GlobeNewswire: Pax8 Research Finds Small Businesses Are Building AI Faster Than They Can Govern It: Q3 2026 Pax8 SMB AI Pulse Report, October 12, 2026
- Federal Reserve Bank of St. Louis, FRED Blog: fred.stlouisfed.org/blog/2026/08/does-generative-ai-save-time-at-work: Does generative AI save time at work?
- Under30CEO: under30ceo.com/small-business-outlook-sbe-council-survey-2026: Small Business Outlook Is Steady but Cautious, survey sample and fieldwork details