Agents scale whatever you already have
By Dino Nokic · AUG 2026 · 4 MIN READ
Gartner's forecast for 2028: AI agents will outnumber sellers ten to one. From the same forecast: fewer than 40% of those sellers will say the agents improved their productivity.
Ten times the digital workforce. Under four in ten saying it helped. The gap between those two numbers is the entire subject.
Gartner's analyst has a name for what produces it — agent sprawl: more digital activity, little improvement in actual impact. And he says one thing I'd put on a wall. If the systems are fragmented, the agents will scale the fragmentation. If the systems are redesigned around human judgment and customer value, agents can create real capacity.
That's the whole mechanism, stated plainly by a research house rather than by a vendor. An agent is a multiplier. It does not repair the process it's pointed at. It runs that process more times per hour, at lower cost per run, with more confidence in the output. If the process was sound, you get leverage. If it wasn't, you get the same defect at volume — and now it arrives faster than anyone can inspect it.
The paradox underneath the paradox
The same survey — 210 chief sales officers and senior sales executives, early 2026 — turned up a number that explains a lot about why the productivity gains keep not showing up. 60% of those officers say their revenue is largely driven by factors outside their control.
Sit with that. The majority of the people accountable for the number believe the number is mostly set somewhere else. Then the response to flat productivity is to buy tooling that acts on the part they do control — activity. More outreach, more follow-up, more touches, more logged fields.
Agents are extremely good at producing activity. That is precisely the thing that was never the constraint.
Why "more agents" stopped being a strategy
Deploying an agent is easy now. That's not a complaint, it's the point — and it's exactly why deployment can't be the differentiator any more. When everyone can stand something up in an afternoon, standing it up is table stakes and the advantage moves somewhere else.
Gartner puts a number on where it moves: by 2028, sales leaders who overhaul their data, their workflows and the actual experience of using the thing will be five times more likely to see ROI from AI than those reaching for quick fixes.
I've made this argument before about the two curves — agents arriving fast in one forecast, agent projects being canceled fast in another, both true at once, and the ground underneath deciding which curve you're on. This is that same argument arriving in a different function. The technology is identical for everyone. What differs is whether anything was built to receive it.
The question that's actually worth asking
Not "where can we deploy agents." That question has too many answers and all of them are yes, which is how you end up with sprawl.
The better version is three questions. Where does friction actually cost us something we can count? Where would a decision get better rather than merely faster? Where would capacity genuinely appear — not hours theoretically freed, but work that stops needing to be done at all?
Those questions have far fewer answers, and the answers tend to sit in the same places every time: the handoffs nobody owns, the exceptions that get chased by hand, the decisions that pile onto one desk because nobody below has a framework they trust.
Fix one of those, then multiply it. That order matters more than anything in the tool selection, and it's the order almost nobody follows, because mapping the process doesn't demo well and buying the agent does.
The sellers in that survey aren't wrong, by the way. Fewer than 40% will say it helped because for most of them it genuinely won't have. Not because the models are weak — they're extraordinary — but because a multiplier applied to a fragmented operation returns a fragmented operation, ten times over, with better formatting.
An agent inherits your operation exactly as it is. Ten agents inherit it ten times.