Cisco equips all 90,000 staff
with a personal AI agent

On August 27, Cisco announced it is rolling out the personal AI agent MyAgent to all ~90,000 employees — not a pilot, a full company-wide launch. This is a structural re-organization of a Fortune 500 company around agents, and the cost and governance architecture it chose is becoming the template every peer is benchmarking against.

Not a chat box — an execution entry point

The notable thing about MyAgent is that Cisco is trying to turn internal AI from a chat box employees open voluntarily into an execution entry point that remembers continuously, calls tools across systems, and advances tasks in the background. It retains employee preferences, past interactions and work context; instead of typing specific instructions, employees describe a goal, context and expected outcome. MyAgent decides the steps, coordinates across Outlook, Webex, Jira and SharePoint, and keeps moving under appropriate human supervision.

It acts more like a chief dispatcher than a universal executor: behind it sit 800+ specialist agents covering inbox triage, spam filtering, reply drafting, market-intelligence aggregation and more. The underlying Circuit platform already hit 90% employee adoption by July, with 21,000 engineers using AI coding tools — over 80% of them weekly.

The economics: not every request needs the strongest model

The binding constraint of company-wide agents is the bill. Cisco's answer is layering: roughly 50-60% of AI requests are handled by open-weight models running on GPUs in Cisco's own data centers; 20-30% go to traditional software automation (rules, APIs, deterministic flows); only a small share calls external frontier models. Tool selection matches capability and cost rather than chasing model call volume and token consumption.

For observability, Cisco uses Splunk to track token usage and an intelligent router that assigns each task to a model with sufficient capability at lower cost; anything a deterministic flow can do is not forced onto an LLM. This is the most pragmatic part of the MyAgent architecture: the LLM is the planning and judgment layer, and high-determinism steps sink down to rules and APIs. Otherwise 90,000 always-on agents would turn the technical vision into an out-of-control compute bill on day one.

The financial signal runs on the same curve: Cisco's AI orders jumped from $2B in FY2025 to a $9B FY2026 guide, with the stock up about 52% year-to-date as of July. Management and Discussion sections in public filings are now 80-90% AI-drafted — even the CFO uses his own agent to benchmark against peers.

Governance: Agent Registry and MCP Registry

At this scale, governance has to be a mechanism, not a slogan. Circuit runs an enterprise Agent Registry and MCP Registry: agents, connectors and MCP servers built by any team are registered in a unified directory and must pass identity, permission and security review before other systems can call them. That is the infrastructure that keeps shadow agents out — no registration, no review, no call rights.

This design answers the hard governance question of company-wide agents: when every employee has an always-on agent, how are permissions granted, scoped, revoked, and how are an agent's actions explained and rolled back? Registry plus identity review plus permission boundaries turns governance into a workflow rather than after-the-fact auditing. Cisco is now selling the same governance model it validated on 90,000 employees — the move it describes as going from experimentation to execution, and from execution to durable advantage.

Read this separately from the layoff narrative

On May 14 Cisco cut 4,000 jobs, officially framed as "realigning resources" toward silicon, optics, security and AI rather than pure cost-cutting. Stanford SIEPR data adds a colder background — the "junior-gap paradox": AI is hollowing out entry-level knowledge work, and enterprises that automate foundational tasks are losing the pipeline that trains the next generation of experts.

Cisco's official position is that MyAgent does not replace the 90,000 employees but expands their capabilities, freeing people for issues that require judgment and impact. But when 80-90% of filing drafts are AI-generated and 4,000 people have been cut, the line between "expanding capability" and "replacing work" is one every enterprise will have to draw for itself.

Four questions for your enterprise

Once employees habitually describe goals to a personal agent that then calls Outlook, Jira and SharePoint, the center of gravity in enterprise software shifts: vendors once locked users in through interfaces, workflows and data formats; tomorrow the thing employees touch most may be the agent, with the underlying SaaS retreating into data, permissions and APIs. Whoever controls this new entry point between people and all enterprise systems may hold the most important position in the next generation of enterprise software.

References

• PYMNTS: Cisco Deploys Custom AI Agent to Entire 90,000-Person Workforce (2026-08-27) — https://www.pymnts.com/news/artificial-intelligence/2026/cisco-deploys-custom-ai-agent-to-entire-90000-person-workforce/
• 36Kr: Cisco Deploys Personal AI Agents for 90,000 Employees (2026-08-31) — https://eu.36kr.com/en/p/3960230958693507
• Forkast: Cisco Is Deploying AI Agents to All 90,000 Employees. Here's What the Economics Look Like. (2026-08-08) — https://forkast.news/cisco-is-deploying-ai-agents-to-all-90000-employees-heres-what-the-economics-look-like/