Gartner: AI Agents Will Outnumber Human Sellers 10 to 1 by 2028

By 2028, Gartner predicts, AI agents deployed across the commercial function will outnumber human sellers 10 to 1 — yet fewer than 40% of sellers will say those agents actually improved their productivity. VP Analyst Dan Gottlieb’s explanation is a warning as much as a forecast: “AI agents should not be viewed as a shortcut to sales productivity. They are only as effective as the systems they operate within. If those systems are fragmented, the agents will scale the fragmentation.” Sales productivity doesn’t stall because reps forget how to sell, he adds — it stalls because the system quietly caps them, and an agent layered on top of that system inherits the cap.

The gap between deployment volume and productivity gain is sharper because building the agents themselves has never been cheaper. Apollo CEO Matt Curl described a custom GTM reporting tool that once required RevOps analysts and a data-warehousing budget now built in about 30 minutes for roughly $30 in token costs, via a new developer surface exposing Apollo’s API, CLI, and MCP integrations to agents directly. Curl’s framing — “AI rewards context over credentials” — is the optimistic mirror of Gottlieb’s warning: cheap, fast agent-building doesn’t fix a bad data foundation, it just lets a company build ten fragmented tools instead of one.

That fragmentation risk is already showing up at the smallest end of the market. Salesforce reports three out of four small businesses are already investing in AI agents, deployed across two distinct categories — internal employee agents that draft follow-ups and surface knowledge-base answers, and customer-facing agents that resolve support cases around the clock. Both draw on the same CRM data, which is exactly the kind of shared foundation Gottlieb says determines whether agent sprawl compounds or cancels out.

For a consulting engagement scoping an agent rollout, Gartner’s number is the useful part to bring into the room before the deployment plan: the constraint was never how many agents a company can stand up. It’s whether the underlying data and workflow foundation can support the tenth agent as well as it supported the first.