Cisco Gave 90,000 Employees an AI Agent. Which Skill Expires First?
Cisco's MyAgent now handles scheduling, ticket routing, and cross-tool follow-through for all 90,000 of its employees, which puts a date on coordination skills that people spent entire careers building.
Key Takeways
- On August 27, 2026, Cisco gave all 90,000 of its workers an AI agent called MyAgent, built on its Circuit platform.
- MyAgent works across Outlook, Webex, Jira, and SharePoint, and agent use on Circuit grew nearly 350% quarter over quarter before launch.
- Coordination work is how people learn a company, so routing it through an agent moves that learning into the system.
- Nobody has 20 years of experience supervising agents, which puts a 25-year veteran and a 2-year analyst on closer footing than either expects.
Cisco handed the scheduling, the ticket routing, the document chasing, and the cross-tool follow-through for 90,000 employees to an AI agent. Thimaya Subaiya, Executive Vice President of Operations, announced it on Cisco's own blog on August 27, 2026. The system is called MyAgent, it runs on Cisco's Circuit platform, and it works across Outlook, Webex, Jira, and SharePoint.
The deployment math is the part that reports easily: one workforce, 90,000 seats, four systems. The number worth sitting with is what those 90,000 people now have somewhere to put. Enterprise AI agents at this scale absorb a specific category of work, and that work was never only work. It was how people learned the company they work for.
What Cisco actually built
Subaiya described MyAgent as moving "beyond chat-based interaction toward supervised autonomous workflows." An employee defines an objective. The system coordinates the steps toward it, remembers preferences and context over time, and comes back with the result. He called it "an intelligence layer across systems, shifting AI from passive assistance to active support for execution." Circuit, the platform underneath, is described as a secure, governed, multi-model agnostic AI platform. Cisco also reported that agentic interactions on Circuit grew nearly 350% quarter over quarter before MyAgent launched.
Subaiya went further and called MyAgent "the blueprint that helps our customers and partners implement Enterprise AI at scale," describing the move as going "from AI experimentation to enterprise execution." Agents left the pilot stage across the enterprise this year. This is what the stage after that looks like at full headcount.
The skill that just started its clock
I spent nearly 2 decades at Cisco, as Managing Director of Innovation Strategy and Head of Global Innovation Centers, where I shaped a $1.1B innovation portfolio. That work ran on knowing who to ask. Which team owned the broken thing. Which of the four meetings on a Thursday was the one where a decision actually happened. That knowledge took years to build, and it was worth building, because inside a company of that size it decided whether a proposal moved or sat.
That is the exact category of skill an agent absorbs first.
Coordination work is how people learn an organization. The person routing tickets finds out which team is drowning. The person booking the meeting finds out who holds the decision. Route that through a system and the knowledge still accumulates, inside the system.
This is what I mean by the half-life of skills: the interval between learning something and that something going obsolete, and how fast the interval is shrinking. Coordination by hand had a long half-life. For most of the last 40 years, a person who learned how a large company moves could count on that knowledge paying rent for an entire career. A rollout at 90,000-person scale puts a date on it.
"MyAgent moves us beyond chat-based interaction toward supervised autonomous workflows." Thimaya Subaiya, Executive Vice President of Operations, Cisco
Switchboard operators knew this story
At their peak, telephone switchboard operators were among the largest groups of employed women in America. The work was routing: the lines, the people, the shortcuts, the sequence. Automation took it over across a generation, and a craft that had been a dependable living stopped being one. Most of the women kept working. The specific skill expired.
The instructive detail is the timeline. That expiration ran over decades, slow enough that a person could finish a career inside it. An agent rollout to an entire workforce compresses the same movement into something a working life will contain two or three times over. Coordination used to be a career. Now it is a feature.
Supervision becomes the job
Subaiya's own phrase names the replacement: supervised autonomous workflows. Supervision. Somebody defines the objective precisely enough that the agent's version of finished matches the company's. Somebody notices when the output is confidently wrong. Somebody owns the outcome when it is. Ownership of that outcome is where agent programs tend to be thinnest, which I have written about before.
That work is real, and it is a different skill from doing the task. It is also newer than anyone's resume. There is no such thing as 20 years of agent supervision experience yet, which puts a 25-year veteran and a 2-year analyst closer together than either of them expects. For the veteran, that lands as unsettling. It is also the opening. Judgment about what finished means is built on exactly the organizational knowledge that took years to acquire, and that part transfers.
The catch is that the half-life applies to the replacement too. Supervising agents in 2026 will be a different skill from supervising them in 2029. Whatever anyone learns about MyAgent-class systems this year has its own clock running.
Questions for your next leadership meeting
Two questions I am taking into my own advisory work. Borrow both.
When an agent takes over a category of work at your company, where does the learning that used to come attached to that work happen now? If the answer is nowhere, you have moved a training program off the books without replacing it, and the cost arrives three years later in people who cannot judge the systems they are supposed to supervise.
Who at your company can tell when the agent is wrong, and what is the plan when those people retire? Most organizations have never had to answer that, because the people who did the work were the same people who judged it. Agentic AI separates those two jobs for the first time, and the separation is permanent.
Cisco has done the visible part at scale and published the pattern. The harder part is the one nobody announces. Your company's skills are on a clock now, whether or not anyone has said so, and the only question that matters is which ones, and how fast your people can learn what comes next. What is the first skill at your company an agent will make optional this year, and who is going to name it? I am easy to find.
What does multi-model agnostic mean for an enterprise AI platform?
Multi-model agnostic describes a platform that can route work to more than one underlying AI model instead of committing to a single provider. Cisco applies the term to Circuit, the platform beneath MyAgent. The arrangement matters for procurement. Model quality and pricing shift from quarter to quarter, and a company locked to one provider inherits that provider's roadmap. Security and governance controls sit at the platform layer rather than the model layer, so those controls hold steady when the models change.
What is the half-life of skills?
The half-life of skills measures how long a learned capability holds its value before the work it applies to changes underneath it. The term comes from physics, where a half-life is the interval over which half a quantity decays. Applied to a career, it asks one practical question: how many years does training bought today keep paying? Process knowledge inside large companies used to hold its value for decades. Workforce-wide agent rollouts cut that interval to a few years.
Which skills do enterprise AI agents absorb first?
Enterprise AI agents absorb the work that crosses systems without calling for judgment about the outcome. Calendar negotiation, ticket routing, status chasing, and document retrieval all fit that description. Each one follows rules a system can hold, and each leaves a record the agent learns from. Work that sits inside one application, or that depends on reading a person's intent in the moment, moves later. One useful test: could a new hire do it correctly from written instructions alone?
Here is what makes Alex a credible voice on this topic: Alex spent nearly 20 years at Cisco as Managing Director of Innovation Strategy and Head of Global Innovation Centers, where he shaped a $1.1B innovation portfolio on exactly the coordination knowledge an agent at 90,000-person scale now absorbs.
If your workforce is next, the useful conversation starts with which skills go first. Book a conversation →
.svg)
