First in Line for AI

For most of the past three years, the story about AI and work has been written from the top of the economy. A large company announces a cut, a headline assigns the cause to automation, and the owner of a ten-person firm files it away as a problem for someone else. The more useful reading is almost the reverse. Smaller firms may feel the change before the giants do, because a small operation can restructure around a new tool in the time a large one needs to schedule the meeting about it.

That is close to the case TIME laid out in May 2026, drawing on the work of Harvard economist David Deming. Some economists and entrepreneurs now expect the most significant AI-driven workplace changes to appear first in small firms, which can reorganize around new technology faster than larger competitors can. Deming made the underlying point in September 2025: "AI adoption is faster in smaller firms, including startups." About 46 percent of American workers are employed by small companies. If the pattern holds, the standard layoff coverage is looking in the wrong place.

If you run a CPA practice, a small law firm, or a boutique consultancy, the reasonable response is to start preparing now. The evidence is still contested, and economists disagree about how AI will affect firms of different sizes. What follows is the most defensible reading of imperfect data.

Adoption figures vary widely, and most of the gap comes down to how each survey defines "using AI."

The most conservative measure comes from the U.S. Census Bureau's Business Trends and Outlook Survey, which collects responses from hundreds of thousands of firms. Under its original wording, which asked about AI used in the production of goods or services, reported adoption rose from about 4.6 percent of firms in early 2024 to roughly 10 percent by September 2025. In November 2025 the Bureau broadened the question to cover AI used in any business function, and the national rate moved to about 17.3 percent. Later readings have held in the high teens, reaching close to 20 percent in early 2026 by the Federal Reserve's accounting. Small firms tend to sit below that average; under the production-based measure, fewer than 9 percent of businesses with under 250 employees reported using AI as of August 2025.

Industry and vendor surveys land much higher, and they should be read with care. The U.S. Chamber of Commerce reported that 58 percent of small businesses said they use generative AI in 2025, up from 40 percent a year earlier, though that figure is self-identified and self-selected, which tends to lift the count. Salesforce, surveying 3,350 business leaders worldwide, reported that 91 percent of AI-using small and mid-sized businesses credited the technology with higher revenue. The same study found that 83 percent of growing firms used AI against 55 percent of shrinking ones, which suggests the arrow may point partly the other way: a healthy, expanding firm has the slack to adopt, so some of the apparent revenue effect is really the kind of company that adopts in the first place.

A fair summary sits between the two poles. A clear majority of small firms are experimenting with AI in some form, while the share that has built it into a standing workflow is far smaller, nearer the Census Bureau's high teens than the trade groups' near-universal numbers. For planning, the distinction that matters is between trying AI and operationalizing it, because only the second changes a firm's cost structure.

Larger firms lead in raw adoption for understandable reasons: they hold the capital, the data, and the specialist staff to deploy new systems. The advantage available to a small firm is a different one. It lies in the speed of change once the decision is made. A sole practitioner or a small partnership can move from choosing a tool to using it within days, with no procurement queue and no legacy software contract standing in the way. That compression is what lets a small office get ahead of a larger competitor on a specific task.

There is supporting evidence for the mechanism. The Stanford study Deming cites, "Canaries in the Coal Mine," found early employment softening for young workers in AI-exposed roles at large, established firms. Its sample, though, was limited to companies that employed staff continuously from 2021 to 2025. Deming's argument is that this design misses the workers who move into startups and newer ventures, where adoption runs fastest. The shift may therefore show up first as a reallocation of work toward smaller, faster-moving firms, before it shows up as headcount changes at any single company.

A concrete case helps. TIME profiled Sonora, an online guitar school whose founder, Spencer Handley, rebuilt the business around AI agents after the release of Claude Opus 4.5 in late 2025. He replaced several paid platforms, among them HubSpot, Calendly, and DocuSign, with tools built for his company, saving an estimated $250,000 a year. Staff fell from 48 to 30 while revenue held steady. The employees who remained moved into supervising AI agents that handled outreach, customer follow-up, and onboarding.

Treat that as a direction of travel for a digital-first business. A professional practice is a different animal: Sonora automated scheduling, outreach, and onboarding, which sit a long way from trust accounting or a legal opinion. What transfers is the mechanism. A small team without legacy systems can compress months of reorganization into weeks once it commits. The work that goes first is the repeatable kind: first drafts, intake triage, document review, routine correspondence. The judgment a client pays for is the last thing to be touched, if it is touched at all. Firms that handle this well use AI to take routine work off people's desks under supervision, and they keep a named person accountable for everything that leaves the building.

One figure cuts against the replacement narrative, and it is worth stating even though it comes from the same self-reported survey that warrants caution elsewhere. In the Chamber's data, 82 percent of small businesses using AI said their headcount grew over the prior year. Self-selected as the sample is, it complicates any assumption that adoption and layoffs move together.

Start with one repeatable, low-risk task you already understand in detail, and keep a person signing off on every output. The right first task depends on the practice.

In an accounting firm, that might be drafting plain-language explanations of IRS or state notices for clients, condensing a client's prior-year return into a working memo for the engagement file, or producing first drafts of standard engagement letters. In a law firm, it might be summarizing discovery documents or deposition transcripts for an attorney's review, preparing first drafts of routine correspondence, or generating initial versions of standard-form agreements that a lawyer then checks line by line. In a consultancy, it might be turning raw interview notes into structured findings, drafting recurring status updates, or assembling first-pass competitive research for a partner to verify.

Run the pilot for a few weeks and measure one thing: hours saved with quality held constant. If the output is something you would be embarrassed to send a client, that is a useful result too. Retire the tool and try a narrower task. Only after one workflow has clearly saved time should you add a second. Working in this order keeps a firm clear of the common trap of paying for a stack of AI subscriptions no one uses, and it builds the internal evidence you will need to bring partners and staff along.

Expect to encounter dramatic claims: a precise figure for hours saved per firm each year, or the 91 percent revenue number above. Most come from vendors and trade associations with an interest in the result and a self-selected sample behind it. They are useful as signs of direction and unreliable as promises. When you put a number in front of a client or a partner, name where it came from and what it leaves out. Doing that consistently is what marks a firm building on evidence, and it is, quietly, the same discipline that makes AI safe to rely on in professional work: a human who can account for the result.