Skip to content
The AI-First Web Talking point

Cisco's Vijoy Pandey says AI agents can talk without sharing meaning

Pandey says agents must align terms first. He also warns that green dashboards can hide wrong outcomes.

W
WebPulse Newsroom
AI-assisted · 2 min read
Share on X LinkedIn
Cisco's Vijoy Pandey says AI agents can talk without sharing meaning
In brief
  • Vijoy Pandey of Cisco said agents can exchange messages without sharing the meaning of terms, so terms and goals need agreeing before work starts.
  • Pandey separately warned that dashboards can look green while outcomes are wrong, and he called such failures silent, with loud consequences.

Vijoy Pandey, who leads Outshift by Cisco, argued that AI agents can exchange fluent messages and still mean different things. So, he said, terms and goals must be aligned before work starts. He spoke on The Cognitive Revolution, in an episode published on 11 November 2026.

What was said

Pandey used a hypothetical deal desk, not a documented incident. Imagine agents from sales, marketing, legal and engineering. Sales passes on a request: build a feature for 100,000 users. Engineering reads that as 100,000 users in total. The customer meant 100,000 users at the same time.

Pandey said nobody in that chain knows what the number means. In his telling, human teams hit this problem too, and agent teams meet it constantly. That is his assertion; the excerpts give no measured rate.

The mechanism, in plain words: agents can send each other messages, but they may not share a definition of users or delivery date. Each may then act on its own reading. Pandey put it this way: "we've realized that connectivity is not the same as cognition."

He described three layers. Accord aligns the goal and the terms before work. Concord handles negotiation during the work, because dates and expectations keep changing. A delivery date might mean an alpha, a first drop or contract signing. A third protocol, for checking what was delivered, is not released yet. Scope sits underneath, like lane-correction software in a car. It watches for drift and triggers Accord or Concord.

Separately, he warned: "your dashboards might be green, but your outcomes might be wrong, completely wrong."

Why it matters

Our reading: the work shifts from picking a smarter model to writing down what your words mean. A budget signer buying or building agent workflows could ask four things. Which terms, such as users or delivery date, are defined in writing? Who owns those definitions? What triggers a renegotiation when targets change? Who checks the delivered result against what was agreed?

If a gap is only found at delivery, the cost would fall on the customer and the people accountable for the deal. That is our inference, not something Pandey said about this example.

The other side

Pandey is describing problems that Cisco's own protocols aim to solve, so he has a stake in the diagnosis. His evidence here is one illustrative scenario, with no measured failure rate.

The fix is also unfinished, since the verification protocol is still to come. Scope is meant to catch drift mid-work, so a gap found only at delivery is the case the design aims to avoid. And the same confusion happens between human teams, so agents may mostly speed up an old problem.

Written by the WebPulse Newsroom with AI assistance, and checked by our editorial review: every quotation was verified against the recording's transcript. How we use AI.

The conversation this talking point comes from