Most AI agents are brilliant at forgetting everything.
That is the inconvenient reality of deploying a general-purpose AI tool inside a business.
Each session begins blank. No memory of last month's customer complaints. No awareness of the sales calls that flagged the same objection six weeks in a row. No institutional context at all.
The capability is there. The knowledge is not.
Gartner predicts 40% of enterprise applications will embed AI agents by 2026, up from less than 5% the year before.
What most of those deployments share is a data problem, not a capability problem.
"Products are increasingly non-differentiated," says Benjamin Humphrey, CEO and co-founder of Dovetail Software. "Being able to compete is more and more about what you understand about your market and your customers to find an edge."
The Dumbest Smart Tool in the Room
Speed is the wrong thing to optimise for.
The prevailing assumption around agentic AI is that its value lies in how fast it can act.
Query something, get an answer, ship a decision.
The faster the cycle, the better the outcome. That logic holds up until you ask what the agent is actually reasoning over.
A foundation model processing a one-off prompt has no idea what your customers said last quarter. It does not know which product complaints are trending, which account tier is raising the loudest concerns, or which competitor name keeps surfacing in sales calls.
It can write a coherent paragraph about almost anything, but it cannot tell you what your customers are thinking right now, because it has never met them.
The agent's quality ceiling is set by the data feeding it. Raise the ceiling on the data, and the agent stops being a party trick.
The Feed That Never Sleeps
Dovetail Software was built around a specific insight: customer intelligence should be continuous, not periodic.
Traditional approaches to understanding customers run in cycles. A research sprint. A quarterly survey. An annual NPS pull. Findings get packaged into a document, shared in a meeting, and then gradually lose relevance as the business moves on. The intelligence decays almost immediately.
Humphrey describes what this looks like in practice inside Dovetail Software. An agent monitoring customer data can generate a weekly voice of customer report and land it directly in a CEO's inbox. It tracks week-on-week shifts. It holds memory across cycles.
"Another use case," he explains, "is our head of sales has a top deals report sent in Slack, which has transcripts from conversations with prospects, so we can touch on the top objections and which competitors are coming up."
The Foundation Decides Everything
Customer feedback does not arrive pre-organised. It comes in as call transcripts, support tickets, survey responses, app reviews, and interview recordings, distributed across Salesforce, Gong, Zendesk, Intercom, and a dozen other tools, each owned by a different team.
"Businesses are massively siloed between different departments," Humphrey says. "As a product team, it's very difficult to get your hands on verbatims and feedback from all these different departments because they're buried in tools."
The silo problem is not purely operational. When insights are fragmented, no agent, however capable, can synthesise a picture that does not exist. The intelligence has to be brought together before it can be acted on.
Dovetail Software addresses this through a data processing pipeline that takes unstructured, qualitative input and converts it into classified themes that can be tracked over time. Pull in 50,000 support tickets, and the platform does not simply summarise them. It samples the data, extracts key points from each entry, clusters those points into broad themes, and then quantifies how those themes shift week over week. The output is a chart. The story behind the chart is the synthesis of every piece of raw feedback that fed into it.
Humphrey calls it "quantitative meets qualitative." A product leader can click on a trending theme and immediately read the underlying customer verbatims that drove the movement. The agent has not flagged a number in isolation. It has explained what that number means, in the customers' own words.
The Compound Advantage of Knowing More
Generic AI tools will keep improving. Models will get faster, smarter, and cheaper. The gap in raw capability between a purpose-built platform and a horizontal AI assistant will narrow over time on almost every technical dimension.
The gap in institutional knowledge will not.
And that gap comes down to one thing: what the model actually knows when it's asked a question.
A platform continuously ingesting feedback from experience research, voice of customer programmes, revenue intelligence, and customer insights teams builds a body of structured knowledge that compounds with every new data point.
An agent running on that foundation does not need to be told what the company's customers think. It already knows, and it knows how that thinking has shifted since last week.
"The speed of iteration," Humphrey says, "is how you stay ahead. If we can heighten that for product managers and product teams, then our customers will have that advantage."
ⓒ 2026 TECHTIMES.com All rights reserved. Do not reproduce without permission.
