Picture your sales floor in 2028: ten AI agents for every seller, researching accounts, drafting emails, updating the CRM and booking meetings.
That isn't science fiction. It's Gartner’s forecast. The same forecast says fewer than 40% of those sellers will report that agents made them more productive.
Ten times the agents. Not ten times the outcomes.
If that sounds familiar, it should. In Signal 01, we looked at how AI made go-to-market busier without making it better. Agents are the next chapter of the same story, only faster.
Here's what the research says.
Gartner predicts that over 40% of agentic AI projects will be cancelled by the end of 2027, because of escalating costs, unclear business value or inadequate risk controls. It also estimates that only about 130 of the thousands of "agentic" vendors offer real agents. The rest are rebranded chatbots and automation.
Salesforce surveyed 4,050 sales professionals: 54% of sellers have already used AI agents, and 89% plan to by 2027. Yet 51% of sales leaders using AI say disconnected systems are slowing them down.
A second Salesforce study found that 84% of data and analytics leaders believe their data strategy needs a complete overhaul before their AI ambitions can succeed. They estimate a quarter of their data is untrustworthy.
And McKinsey describes a gen AI paradox: roughly 8 in 10 companies use gen AI, and about as many report no material impact on earnings. Fewer than 10% of function-specific use cases ever make it past the pilot.
The agents are arriving faster than the foundations they need.
The four numbers
<40%
of sellers will say agents improved their productivity, even at ten agents per seller.
84%
of data leaders say their data needs a complete overhaul before AI can succeed.
3 in 4
AI high performers redesigned their workflows. Only 1 in 4 of everyone else did.
We've seen this pattern before.
An agent doesn't fix your workflow. It copies it.
Give an agent a CRM full of duplicates and stale job titles, and it will personalise outreach to the wrong people, at scale. Give it a process nobody has written down, and it will improvise. Give it no definition of success, and it will optimise for the only thing it can count: activity.
I call this the mirror effect. An agent inherits your data, your process and your metrics. It doesn't improve them. It scales them.
Agents don't create discipline. They multiply what's already there.
Three foundations, before any handover.
Before handing a GTM workflow to an agent, it's worth checking three things.
-
Data you'd trust
Would you let a new hire act on this data without double-checking it? If not, an agent shouldn't either. High-performing sales teams are far more likely to prioritise data hygiene: 79%, against 54% of underperformers. Start with the fields this one workflow actually uses.
-
A workflow you can write down
If the steps live in someone's head, the agent will guess. Map the trigger, the steps, the handoffs and the exceptions first. Often the redesign is the real win: McKinsey's AI high performers are about three times as likely to have fundamentally redesigned their workflows.
-
A scorecard and a human checkpoint
Pick the one outcome this agent should move, the same discipline as Signal 01, and decide where a person reviews its work before it reaches a buyer. Gartner's cancellation reasons start with unclear value and inadequate risk controls. This addresses both.
None of this is about the model. BCG found that 70% of the challenges in scaling AI come from people and process, 20% from technology and data, and only 10% from the algorithms themselves.
The teams getting value from agents spend their effort accordingly. They redesign the workflow first, measure from day one, and treat the agent as the last step, not the first.
The agent is the easy part. The foundation is the work.
Navneet Kumar
Building Neolism in the open.