Evidence at a glance
Usage Growth Cannot Choose the Starting Point
The first barrier for a small business is usually not access to an AI tool. It is the time and capability required to place that tool inside a specific piece of work. A team with only a few employees may be handling customer acquisition, finance, marketing, supply-chain tasks, and compliance at the same time. It may have no specialist who can translate model capabilities into operating procedures, and no spare capacity to compare possible uses against their risks and benefits. For such a company, “what can AI do?” is still too broad a question.
OpenAI’s report says that during the week of September 9 to 15, about four million employees at companies with fewer than 500 people used OpenAI tools worldwide. Nearly one in five of those users worked at businesses with fewer than ten employees. That demonstrates reach among very small teams, but not that those teams have developed reliable ways of working. User counts show who opened the tool. They do not show who embedded it into an operation, or whether the result justified continued investment.
SBDC Is Providing Adoption Design, Not Just Promotion
The central feature of the partnership is not the publication of more online tutorials. It is the decision to split adoption into advisor training, hands-on practice, and follow-up support. Advisors will be trained and credentialed to help businesses choose an appropriate starting point, use AI responsibly, and assess whether the tool is actually helping. Participating SBDC networks then plan to host at least one Academy Workshop each, where owners can test ChatGPT on real business tasks rather than watch a product demonstration.
After the workshops, SBDC advisors will continue answering questions and providing one-on-one support. OpenAI and America’s SBDC also plan to create practical guides and an industry-focused companion guide covering possible areas such as marketing, operations, manufacturing, exporting, and government contracting. This creates a feedback loop in which problems discovered by owners can shape the guidance itself. It is closer to an iterative deployment organized around tasks and industries than to distributing one prompt library to every company.
Rising Agentic Use Means Tasks Are Becoming Longer
The most useful signal for a technical leader is not only the four-million user figure. It is the change in how the tools are being used. OpenAI says that in August, agentic output tokens for products such as ChatGPT Work and Codex accounted for two-thirds of total small-business output tokens, compared with one-third in April. In the material, tokens are defined as units that models process and generate. At minimum, the change suggests that more small-business work is moving from single-turn questions toward multi-step execution.
Token share cannot be treated as a direct measure of automation or business value. A longer process may mean that AI is taking on more work. It may also mean that the system requires more rounds of correction, or that a company is assigning complex tasks to a system that has not been adequately validated. Technical leaders should therefore break “agentic” use back down into a concrete workflow. They need to know where inputs come from, who checks intermediate results, and whether the final action can trigger a payment, quote, contract, or external communication.
Small Businesses Are Filling Missing Roles Before Replacing Staff
The material says that small-business interactions involving finance and accounting, supply chain, legal and compliance, and marketing were at least three times as prevalent as those in larger companies. This looks more like a signal created by organizational structure than evidence that small businesses have achieved more aggressive automation. Larger companies may have dedicated staff for these functions. Smaller firms often need a general-purpose tool to temporarily fill gaps in analysis, research, and content work.
Ted and Tami Taylor, who run three restaurants in New Mexico, used ChatGPT Work to compare competitor pricing while reconsidering their menu prices. The analysis showed that raising prices on some items could make them less competitive, helping them avoid skipping the competitor-comparison step. The example shows how AI can cover an analytical gap in a small team. It does not show that the model replaced managerial judgment. The owners still had to decide whether to change prices in the context of their own business.
The Last Mile Also Creates a Governance Bill
The strength and cost of the SBDC model come from the same source: its reliance on trusted advisors. An advisor can lower the owner’s cognitive burden, help select a starting point that can be evaluated, and stop a team from applying the tool directly to unsuitable situations. But training about 150 advisors, running in-person workshops, and offering continued support means that the model will not have the rapidly declining marginal cost of pure software distribution. Its ability to expand will depend on the number of advisors, the quality of training, and the reach of local networks.
The material does not specify the credential standard, the states or networks that will participate, or adoption rates, error rates, and cost savings after the workshops. The safer implementation choice is therefore not to pursue broad automation first. Assign an accountable owner to each workflow, begin with tasks where inputs and outcomes can be compared, keep human review and records, and expand only after the process proves reliable. For OpenAI, the partnership is about more than usage volume. It is an attempt to make a local advisor network infrastructure for reaching long-tail businesses. For those businesses, an advisor can lower the cost of getting started, but cannot assume final responsibility.