Cisco has made an individual AI agent available to nearly all of its 90,000 employees from the start of its 2027 fiscal year. This is not a limited experiment rolled out to a few departments; it is a uniform deployment, presented internally as a durable infrastructure building block rather than a pilot project.
Dubbed MyAgent, the tool is accessible from a desktop or mobile app and integrates with Webex, Cisco’s own video-conferencing platform, as well as Outlook, Jira, and SharePoint, according to Cisco.
The tool is built on Circuit, Cisco’s internal AI platform described as secure, governed, and model-agnostic, designed to give employees controlled access to approved LLMs, agents, and company data.
Each instance is strictly personal and only receives instructions from the user to whom it is attached. Any action that has an effect outside the system requires explicit human validation.
MyAgent operates on an intent-driven model. The user defines an objective, a context, and the expected outcome, and MyAgent independently determines the sequence of steps needed to achieve it. It also possesses persistent memory, intended to preserve preferences and the history of interactions with an employee over time.
Cisco notes that agent interactions on its Circuit platform advanced by nearly 350% quarter-on-quarter before the broad rollout of MyAgent (at the end of July).
Multi-model routing designed to control costs
To avoid an explosion in costs associated with the most powerful AI models, Cisco built an internal orchestration layer that distributes each request to the model deemed most relevant based on the nature of the task, the tolerated latency, and the required level of reliability.
According to available information:
- between 50% and 60% of requests are handled by “open-weight” models hosted in-house on Cisco-owned GPU servers;
- between 20% and 30% fall under classic software automations, without using a generative model;
- only a residual share of calls rely on external foundation models, such as Azure OpenAI, Claude from Anthropic, or Google’s Gemini.
“This won’t burn a ton of tokens with frontier models,” asserts Mark Patterson, the company’s chief financial officer, in an interview with Fortune, touting this architecture as a lever for budget discipline as much as for technical sophistication.
The tracking of token consumption is, meanwhile, handled by a supervising tool from Splunk, which Cisco acquired in 2023.
Governance and Access Security
MyAgent is not intended to have unrestricted access to the IT system. Cisco places it within a governance architecture that determines which models, systems, and data the agent may use, and which actions it may perform. The goal is to separate the ability of the agent to decide the steps in a task from its authorization to actually execute those operations.
This control layer is complemented by Cisco’s security mechanisms, grouped notably under the AI Defense banner, aimed at monitoring interactions between users, models, and agents. The challenge is to detect attempts to manipulate instructions, malicious behavior, or data transfers to unauthorized resources.
Uses focused on finance and regulatory reporting
The use cases highlighted by Cisco’s leadership are primarily in high-value, document-heavy functions:
- Regulatory drafting. According to Mark Patterson, between 80% and 90% of the first version of the Management’s Discussion and Analysis (MD&A) section of filings submitted to the U.S. Securities and Exchange Commission is now AI-generated.
- Financial governance. The CFO uses a dashboard named “CFO cockpit,” powered by AI, which aggregates performance by product, geographic region, and customer segment, and suggests corrective actions.
- Competitive intelligence. Patterson’s personal agent automatically compares Cisco’s performance to that of its peers on indicators such as revenue growth, earnings per share (EPS), and R&D expenditure.
- Mailbox management. More broadly, employees rely on their agent to summarize emails, filter spam, and pre-draft replies.
- Market analysis. Some agents are also deployed to synthesize financial news and outline expectations for stock movements.
A ‘zero client’ for Cisco’s AI offerings
Beyond internal use, this deployment also serves as a commercial showcase. Cisco positions itself as its own large-scale reference client for its AI agent offerings (notably Webex AI Agents) as well as for its AI-optimized network and security infrastructure.
But several questions remain.
First, about the program’s cost. Unconfirmed estimates point to an annual token bill of around $900 million, or roughly $200 per employee per week. A figure to be taken with caution given the absence of official confirmation.
Next, the medium-term impact on employment. Management denies any direct link between the deployment of agents and the 4,000 job cuts announced in May, described by Cisco as a “resource reallocation” toward strategic segments.
But the sequence of announcements, combined with statements from other industry leaders on similar topics, leaves open the question of a deeper transformation of roles within the group.
