OpenAI's small-business report: one chat box standing in for a payroll of specialists

Four million founders, and almost none of them building software

OpenAI's report opens by rejecting the Silicon Valley cliche it expects readers to hold. The four million people it counts using ChatGPT for a business in March 2026 are not writing code. They run consulting practices, online stores, home-service outfits, beauty practices, and restaurants.

The category breakdown makes this concrete: professional and agency services at 22 percent, retail and e-commerce at 21 percent, home and trade services at 12 percent, health and beauty at 11 percent, and food and hospitality at 8 percent. These are firms where the owner is still doing the work themselves.

The report frames the appeal precisely: for these owners AI is 'not primarily a product to sell' but 'a flexible source of capabilities that might otherwise require outside consultants, additional staff, or specialized software.' That reframing is the whole point of the piece.

The interface is the product here, not the model

What makes this a frontend story is the range of unrelated jobs funneled through one text box. OpenAI gives the example of a contractor who revises an estimate, explains a permitting requirement, writes website copy, and drafts a customer follow-up — four different specialist functions, one input field.

A large company would route those tasks to marketing, finance, legal, and operations teams. A sole proprietor has no such teams, so the demand OpenAI describes is specifically for 'general purpose assistance.' The value isn't a smarter model; it's that a single, undifferentiated interface can absorb tasks that would otherwise need four separate tools or hires.

For anyone building software for this audience, that's the tension. Purpose-built SaaS wins on depth for any one task. The chat box wins because the owner doesn't want to learn, pay for, and switch between five products to get a 'first approximation' of each capability.

Usage shifts from validation to operations — and so does willingness to pay

The report splits users into prospective (29 percent) and active (71 percent) entrepreneurs, and the tasks differ by stage. Prospective founders lean on branding, product development, and business validation — asking whether an idea has a market and what it might cost to launch.

Once a business is operating, the work changes: marketing and copywriting make up 26 percent of active entrepreneurs' activity, customer communication 11 percent, and legal and compliance questions 10 percent. The interface follows the owner from thinking about a business to actually running one.

The revenue signal sits in the plan tiers. Active entrepreneurs are 67 percent of entrepreneurial users on Free, 78 percent on Plus, and 81 percent on Pro. OpenAI's reading is that willingness to pay rises once the tool is 'connected to real commercial activity' — someone brainstorming a name pays casually, while someone juggling customers and deadlines has reason to buy speed and capability.

What a lower fixed cost of capability changes for small firms

OpenAI is careful about its claim. It states that generative AI 'cannot eliminate the risks of entrepreneurship,' only reduce the cost of a first approximation of many useful skills. That is a deliberately modest framing, and a more credible one than most adoption reports offer.

The specific implication is about firm viability, not growth. The report expects 'not a wave of high-growth startups, but an uptick in modest businesses that launch more quickly or survive because their founders can manage work that would otherwise overwhelm them.'

It may also make very small firms more viable by allowing a sole proprietor or two-person company to handle a wider range of tasks without immediately adding employees or purchasing numerous specialized services.Montana Labs

If that holds, the competitive pressure lands on the low end of the tool market — the entry-level bookkeeping, copywriting, and customer-service products a two-person shop used to stitch together. The chat interface doesn't beat any of them on depth; it beats the whole stack on being one place the owner already knows how to use.