Agents That Think Together: Cisco's Vijoy Pandey on Semantic Alignment, Scope, Accord & Concord
Vijoy Pandey argues that connected agents are not yet thinking together, and that open tools for cognitive observability and intent alignment (Scope, Accord, Concord) are needed to make multi-agent systems reliable.
Watch on YouTube ↗Charlie's summary
- ▶ 6:42
Agents went from isolated to islands to connected, but connectivity is not cognition
Vijoy says agents moved from singular agents to islands inside one harness to connected agents from different vendors, clouds and organizations. Protocols like MCP, A2A and agency solve discovery, identity, communication and observability. They do not make agents reason together. The missing step is shared cognition, which Cisco calls the internet of cognition.
- ▶ 16:58
In multi-agent work, the value sits in the handoffs, not in any single agent
Vijoy uses a multimillion-dollar deal that touches sales, legal, finance, engineering and marketing agents from different vendors. No one agent owns the problem. “The value is actually in the handoff,” because an estimate becomes a commitment, then contract language, then a marketing claim. Optimize for the global outcome, not each agent's local result.
- ▶ 27:16
Language gave humans shared intent, shared knowledge and shared reasoning; agents need the same stack
Vijoy draws a parallel to the invention of language about 70,000 years ago. It let humans share intent and coordinate, share knowledge across generations, and then reason collectively. Agents are individually smart but isolated. They need alignment before the work (intent), during it (coordination and negotiation), and after it (verification against the global goal).
- ▶ 36:27
Dashboards can be green while outcomes are wrong: silent failures with loud consequences
Looking at public traces, Vijoy found that 100% of ChatDev sessions had semantic and cognitive misalignment failures. In a five-agent system, accuracy fell from 99% to 98% after one failure, to 81% after two, and to 16% after three. A travel-booking system was 66% up with 0% task completion. Scope, open source and built on OpenTelemetry traces, makes these failures visible.
Give your email to unlock the rest of the key points — for every episode.
Do more with Charlie
Free — sign in with your email, no password.