Small businesses are moving from prompts to delegated work

From assistance to completion

OpenAI reports that agentic output grew from about one-third of output tokens used by small and medium-sized businesses in April 2026 to two-thirds in August. In this context, agentic work means systems carrying out multi-step activity, rather than answering a single prompt and stopping.

The company’s analysis covered about four million employees at firms with fewer than 500 people during the week of 9–15 September 2026. Nearly one in five worked at a firm with fewer than ten employees. The data describe activity on OpenAI products, so they should not be read as a census of every small business or every AI tool.

Where small teams are leaning in

Compared with larger businesses, smaller firms devoted at least three times the share of their usage to each of four areas: finance and accounting, supply chain, legal and compliance, and marketing. These are functions where a small company may have recurring work without enough volume to employ a specialist for every need.

That does not mean the agent replaces the missing specialist. It can prepare reconciliations, organise supplier information, draft a compliance checklist or assemble campaign variations. A person still needs to define the boundary, supply the context and review consequences that the system cannot carry.

The small-company advantage

A small team can change a workflow quickly because there are fewer handoffs and legacy systems. It can also expose itself quickly if one person delegates too much without controls. The best first job is frequent enough to matter, narrow enough to evaluate and reversible when the result is poor.

The emerging opportunity is practical: build agents around work that small firms already need but cannot staff consistently. Measure completed tasks, corrections and time returned to the team. Token volume is evidence of changing behaviour; business value still has to be demonstrated in the operation.