Gartner says AI agents will outnumber sellers 10 to 1 by 2028. Most sellers say it hasn't helped yet

Gartner put a number on something a lot of sales leaders have been feeling but not saying out loud. In a forecast published July 28, 2026, the firm predicts that by 2028, AI agents will outnumber human sellers 10 to 1 inside a typical sales organization — and, in the same forecast, that fewer than 40% of sellers will say those agents actually improved their productivity.

Read those two numbers side by side and they don’t fit together the way vendor decks usually promise. Ten agents deployed for every seller isn’t a rounding error or a pilot program — it’s the default operating environment Gartner expects most sales orgs to be running inside of within two years. And in that environment, well under half of the people actually doing the selling are expected to say it helped.

This isn’t a story about AI not working

It would be easy to read the 40% figure as evidence that sales AI is overhyped. That’s not what the surrounding research says. A separate Gartner survey published in May 2026 found that AI already saves the average seller nearly five hours a week — a real, measurable reclaim of time that used to go into manual research, note-taking, and CRM data entry.

The catch is what happens after the five hours get freed up. The same survey found that 72% of sales organizations fail to reinvest that reclaimed time into anything that actually moves a deal forward. The agents are doing their job. The organizations around them mostly aren’t doing theirs.

The number that actually predicts whether it works

A third Gartner finding from the same research window is the most useful one for anyone deciding what to actually do about this. Sales organizations that provide AI-enabled next-best-action guidance — meaning the AI doesn’t just draft an email or summarize a call, it tells a rep what to do next and why — are 2.6 times more likely to achieve commercial growth than organizations that don’t.

Put the three findings together and a pattern emerges that has nothing to do with how many agents you’ve deployed:

• Agent count alone doesn’t predict productivity. Ten agents per seller by 2028 is Gartner’s forecast either way — for the 40% who feel the benefit and the 60%+ who won’t.
• Time saved isn’t the same as value created. Five hours a week is meaningless if nobody redesigned the week around what to do with it.
• Guidance, not generation, is what correlates with growth. The 2.6x organizations aren’t the ones with the most AI tools. They’re the ones where AI tells a person what to prioritize next, inside a process built to act on that guidance.

That reframes the “10 to 1” headline from a warning about robot sellers replacing humans into something more specific: a warning about tool sprawl outrunning process redesign. An organization can genuinely deploy ten agents per seller and still land in the 60% who don’t feel it, if none of those agents are wired into an actual next-action, and if nobody rebuilt the seller’s day to use the hours those agents free up.

Why this should sound familiar if you run outbound

The same failure mode shows up constantly in cold email and outbound specifically, and it’s worth naming directly: adding an AI drafting tool to a sequence-building workflow doesn’t automatically produce better replies. It produces faster drafts. Whether those drafts turn into pipeline still depends on the same fundamentals that predated AI entirely — a real signal to write about, a clean deliverable sending domain, and a list that hasn’t decayed since it was imported.

This connects directly to two things worth reading if this pattern interests you: the market concentration behind who’s actually building these agents — because three vendors controlling the underlying models shapes what “an AI agent” can even do inside your sales stack — and why buyers increasingly distrust AI SDR tools by default, which is the demand-side mirror of Gartner’s supply-side finding: a recipient can tell the difference between an agent that drafted something generic and a rep who acted on a specific, timely signal.

What “redesigning around the AI” actually looks like

Concretely, the organizations landing in Gartner’s 2.6x cohort share a pattern that doesn’t require exotic tooling — it requires deciding, in advance, what a human does with what the AI hands them:

1. The agent’s output has a defined next step, not just a defined output. A drafted email that nobody is required to review, personalize, or send within a set window is time saved that evaporates into nothing.
2. Someone owns the reinvestment, explicitly. If five hours a week frees up, a manager decides in advance what fills it — more account research, more calls to warm replies, more time on the accounts already showing intent — rather than letting it default to idle capacity.
3. The tooling stays scoped to a narrow, checkable task. Broad, general-purpose “AI agents” are harder to trust and harder to audit than a tool built to do one specific thing well, which is also the throughline in why rogue or poorly-scoped agents are becoming their own cybersecurity category — trust erodes fast when nobody can explain exactly what the agent is authorized to do.

A familiar shape: tools arriving faster than the process around them

Sales organizations have lived through a version of this gap before, and it’s worth naming because the outcome the first time around is instructive. When CRM software went mainstream in the late 1990s and early 2000s, adoption numbers climbed fast and the promised productivity gains lagged badly behind — a well-documented pattern at the time was reps treating the CRM as a reporting chore imposed on them rather than a tool that changed how they worked, because most organizations bought the software without redesigning the actual selling motion around it. The tool went in. The workflow around the tool didn’t change. Years of underwhelming ROI followed, right up until organizations started building actual process — defined stages, defined next actions, manager coaching tied to what the CRM showed — around the software instead of just requiring people to type into it.

Marketing automation went through a milder version of the same arc a decade later: platforms capable of sophisticated lead scoring and nurture sequencing got bought by teams that then ran them as glorified email blasters, because nobody redesigned the handoff between marketing and sales around what the tool could actually do. The lesson both cycles taught, expensively, is that a tool’s ceiling and an organization’s actual results are two different numbers, and the gap between them is entirely a function of whether anyone rebuilt the surrounding process.

Gartner’s 10-to-1 forecast reads like the AI-agent version of the same story, just compressed into a much shorter runway. CRM adoption took the better part of a decade to mature from “software reps resent” into “software that actually changes selling.” Sales orgs deploying AI agents in 2026 don’t have a decade to figure out the process side — competitors who get the reinvestment question right first will simply be faster and cheaper to sell against while everyone else is still treating the agents as a line item on a tools budget rather than a redesign of who does what.

The compressed timeline is the actual risk

That compression is worth sitting with for a second, because it changes what “wait and see” costs. With CRM software, an organization that took five extra years to redesign its process around the tool was still competing against other organizations moving at roughly the same slow pace — the whole category matured gradually, together. AI agents in sales aren’t following that shape. Gartner’s own 2026 research already shows a split forming between the minority pairing agents with next-best-action guidance and the majority just adding tool count, and that split is happening inside the same one- or two-year window, not spread across a decade the way CRM’s maturity curve was. An organization that spends 2026 and 2027 accumulating agents without redesigning anyone’s day around them isn’t just moving slowly — it’s falling behind competitors who treated the same forecast as a deadline to fix the reinvestment question, not a curiosity to revisit next planning cycle.

Where Norbelys fits this exact gap

This is the specific failure mode Norbelys’s AI campaign builder is built to avoid. It doesn’t hand a seller a generic “AI agent” and leave the next-action question unanswered — it takes a written brief and produces a staged sequence with drafted variants ready for review, which is a defined next step by design, not an open-ended tool a rep has to figure out how to use. And because it’s included on every plan rather than gated behind an enterprise tier, the reinvestment question Gartner’s research keeps surfacing — what do you actually do with the time saved — has a specific, immediate answer: less time drafting from a blank page, more time on the accounts already replying.

If the Gartner numbers above sound like your own sales org — plenty of AI tools, no clear sense of whether they’re paying off — the fix Gartner’s own research points to isn’t fewer agents or more agents. It’s fewer, better-scoped tools wired to an actual next action. See how Norbelys prices that in or start sending with a sequence built from a brief instead of a blank draft.

AI agents in sales — quick answers

Does Gartner's forecast mean AI agents will replace human sellers?+

No — the forecast is about the ratio of deployed agents to human sellers inside an organization, not headcount replacement. Gartner's own research emphasizes that the organizations seeing real gains are the ones pairing agents with human judgment on next actions, not removing people from the process.

Why would adding more AI tools not improve productivity?+

Because time saved by an agent only becomes value if someone redesigns the day around using it. Gartner found 72% of sales organizations don't reinvest the roughly 5 hours a week AI frees up into higher-value selling activity, so the saved time doesn't show up as more pipeline.

What actually correlates with better outcomes, according to this research?+

AI-enabled next-best-action guidance — tools that tell a seller what to prioritize and why, not just tools that generate content. Organizations using that kind of guidance were 2.6 times more likely to achieve commercial growth in Gartner's survey.

How does this apply specifically to cold email and outbound?+

The same pattern holds: an AI draft tool speeds up writing, but reply rates still depend on whether the email references a real signal, lands in the inbox instead of spam, and reaches a list that hasn't gone stale. Faster drafting without those fundamentals just produces faster generic email.