Cisco’s 90,000-Employee AI-Agent Rollout Tests Governance Before ROI
Editorial Team
••📖5 min read
Cisco is giving its roughly 90,000 global employees access to MyAgent, a supervised assistant that can coordinate work across company systems. The deployment’s strategic test is whether model routing, on-premises infrastructure and permission controls can make agent use affordable and auditable; Cisco has not yet disclosed enterprise-wide costs or productivity results.
Cisco has begun giving MyAgent, a supervised AI assistant, to roughly 90,000 employees worldwide.
The system can plan multi-step work and call more than 800 back-end subagents, but external actions need the assigned employee’s approval.
Cisco routes requests among open-weight models, automation and foundation models; it has not published total cost, error rates or time saved.
Cisco, the networking-equipment maker, is extending its internal AI program to its entire workforce. MyAgent is not an unsupervised digital employee: it is an agent layer attached to each worker’s existing permissions, with Cisco controlling the models, data paths and actions available to it.
Thimaya Subaiya, Cisco’s executive vice president of operations, described the rollout. He joined Cisco in 2018 and previously ran customer-experience operations and served as chief transformation officer, according to the company’s 2025 annual report. That operations role is directly relevant to a deployment that changes internal systems. Cisco says in its announcement that the program is rolling out to about 90,000 employees; a report published Aug. 27 described access reaching the full global workforce this week.
Cisco’s official internal AI assistant page uses a workplace scene to illustrate employees working with the tool. Source: Cisco Blogs.
Scale is the claim; supervision is the mechanism
MyAgent is available through desktop and mobile applications and Cisco’s Webex collaboration software. Secure enterprise connectors retrieve employee context in real time under existing user permissions. More than 800 specialized subagents can perform prompted jobs such as checking sales-forecast variances and sending an anomaly notification.
The boundaries are unusually explicit. The agents respond only to their assigned humans; they do not communicate with other agents or employees. Any action outside the system requires that employee’s signoff. Cisco also says a policy server can block destructive operations, including deleting datasets, and prevent Cisco data from being used to train third-party models.
SoftBank shares fell 10.8% as default-insurance costs reached a four-year high. Delayed listings and unfinished data centers complicate its AI investment outlook, while a $40 billion bridge facility has a March 2027 maturity and a repayment plan based on existing assets and other financing.
Editorial Team
The breadth is notable, but not proof of a unique capability. Enterprise agents are already being used for coding, customer support and financial analysis. The differentiator Cisco is claiming is deployment scale—one assistant for every employee—combined with a common control plane. Cisco’s announcement calls the rollout a blueprint for customers and partners, making the company both the operator and a seller of the architecture.
Subaiya described the intended labor relationship this way:
“I truly believe that you cannot replace humans entirely with AI. What you can do is very successfully augment people with AI.”
That is a company executive’s view, not an independently measured result. Cisco has not reported how often workers accept an agent’s recommendation, how many tasks are completed without rework or whether approval requirements slow the workflows.
Routing is Cisco’s cost-and-control bet
Agent runs consume more tokens than ordinary chats because they repeatedly plan, call tools and process intermediate steps. Cisco built an “intelligent router,” described as a model harness, to direct each request to what it considers the most capable and cost-efficient option. Its own data center uses GPUs to run open-weight models, while Splunk’s token-monitoring feature tracks usage; Cisco acquired Splunk in 2024.
Subaiya said roughly 50% to 60% of requests use open-weight models and 20% to 30% rely on software automation. He characterized the remainder sent to foundation models as “a very small percentage.” Those ranges leave a potentially material 10% to 30% unaccounted for, depending on the endpoints and rounding, so Cisco has not supplied a precise request denominator or a fully reconcilable model mix.
Mark Patterson, Cisco’s executive vice president and chief financial officer, has spent 26 years at the company and became CFO in July 2025 after serving as chief strategy officer. In a Fortune interview, he said his own agent benchmarks Cisco against peers on revenue growth, earnings per share, research-and-development spending and capital allocation. The comparison shows how the tool is being used inside finance, but it is not evidence that MyAgent improves Cisco’s results relative to those peers.
Patterson said AI now creates 80% to 90% of the first draft of the management discussion and analysis section in public filings. His team is refining an investor-relations tool that compares Cisco’s financial history with competitors’ earnings calls and a “CFO cockpit” that would synthesize performance data and recommend actions. These are management’s reported uses and a planned system, not an independent productivity audit. Cisco told Fortune that it does not separately disclose the agents’ costs outside its earnings reporting.
Circuit turns an internal rollout into a product lesson
MyAgent runs on Circuit, Cisco’s secure, governed, multi-model-agnostic platform. Cisco describes Circuit as a single architecture for approved large language models, agents and enterprise data. MyAgent adds intent-based execution: an employee supplies an objective, context and desired outcome, and the system determines a sequence of steps across tools such as Outlook, Webex, Jira and SharePoint.
Persistent memory retains preferences, prior interactions and context, which Cisco says keeps the assistant continuous and relevant. Cisco also says agentic interactions on Circuit grew nearly 350% quarter over quarter. The company does not disclose the starting level, the period’s request count or an independent measure of completed work, so the growth rate cannot be translated into adoption, savings or quality.
The deployment arrives as Cisco emphasizes networking, custom silicon, optical networking and security businesses aimed at AI infrastructure. An internal reference design can help Cisco demonstrate governance to customers and partners, but it also gives the company an incentive to present internal usage as evidence that the stack scales. No source reports the rollout’s total implementation bill or financial return.
The next evidence Cisco needs to publish
Giving every employee access establishes reach, not value. Cisco has described safeguards, routing choices and a few finance workflows, but the central business question remains open: do the controls and infrastructure produce reliable work at a cost lower than simply buying frontier-model access or separate enterprise tools?
The next decision is whether Cisco will connect MyAgent’s control metrics to audited outcomes. A credible case would disclose request volumes by model class, spending, task-completion and rework rates, approval frequency, and time saved against a defined baseline. Until then, MyAgent is best read as a large-scale operating test—humans retain authority, while Cisco’s platform determines the models, data pathways and costs beneath their requests.
U.S. prosecutors allege a global fraud combined AI impersonation, sham confidential deals and cryptocurrency laundering. A blocked bank transfer shows why convincing an employee and moving the money are separate hurdles.
President Donald Trump pledged support for states embracing AI without announcing project commitments. Existing federal incentives and Virginia’s data center review show why the returns depend on tax concessions, construction and who pays for electricity infrastructure.
Tesla plans to wind down Cybertruck production at its Texas factory around mid-2027 to expand Cybercab output. Weak pickup sales frame the retreat, while battery constraints and rising investment complicate the robotaxi expansion. Final truck output and customer order handling remain unclear.
EU regulators preliminarily found TikTok failed to adequately assess and reduce risks from repetitive video recommendations. The case challenges engagement signals and ineffective user controls, including for legal content. TikTok disputes the findings.
Activists breached the Amsterdam construction site intended for Microsoft, joining opposition to data center expansion with allegations about surveillance. The developer plans more than €1 billion in investment, while government records acknowledge competition for scarce grid capacity.
Airbnb plans to start rolling out an AI travel agent in the first half of 2027. Its wider ambition to bring partner bookings into its app could bring more revenue and support obligations, while Expedia’s abandoned Romie assistant shows the execution risk.
South Korea halted KOSPI share trading after a plunge of more than 8%, while Japan’s Nikkei touched 20% below its June peak. Investors fear a slowdown in AI infrastructure spending, but the market losses do not yet establish cuts to construction or supplier orders.
Kalepa’s $75 million Series B will fund more insurance workflows, a planned London office and a doubling of staff in 2027. The company says it has been cash flow positive for over two years; its next test is extending automation while preserving insurer control over consequential decisions.
The Cambridge startup has secured backing described as eight figures in capital and compute, including UK government support. Its wager is that cheaper, accurate quantum calculations can supply better AI training data; full performance results remain forthcoming.
Japan’s Nikkei fell 7.09% in Friday morning trading, with AI, chip and data center-related shares hit hard after Wall Street losses. Reports linked the rout to fears of U.S. regulation following Democratic election victories, without identifying an enacted restriction or a cut in AI spending.
Potential investors are pressing Anthropic for a lower valuation as a technology selloff threatens its November IPO timetable. Rapid revenue growth and substantial cash complicate a financing debate driven by losses and long-term infrastructure commitments.
Visa's new benchmark finds that AI shopping models often miss customer requirements and struggle to repeat successful decisions. The tests cover product discovery, leaving checkout and payment reliability for future rounds.
Agility Robotics began trading after its Churchill Capital Corp XI merger, bringing a commercially deployed humanoid robot maker to U.S. public markets. Its next test is turning conditional Digit 5 orders into deliveries and sustainable economics.
Goldman Sachs sees temporary halts potentially affecting about 45% of planned US data center expansion after Democratic election victories. In Texas, where approvals were already paused, the next fight is over how long restrictions last and what developers must pay for.
Core Scientific’s third-quarter colocation revenue rose 57.6% to $215.4 million. Its expansion beyond CoreWeave could diversify sales from 2027, but the next projects require additional financing rather than the same customer funding that supported its existing conversions.
Meta is rolling out voice conversations and animated avatars for Muse in the US. The update adds personality to a cloud-based task agent whose costs, permissions and reliability matter more than its expressions.
Anthropic is rolling out Personal Finance to paid Claude users in the US, using Plaid to read bank data. Its pitch rests on ad-free answers and privacy controls, while its own documentation warns that balances and budgets can be incomplete.
OpenAI president Greg Brockman predicts significant medical advances within 12 to 18 months and raises the possibility of death becoming a choice. OpenAI’s cell research and a separate early drug trial show narrower progress, with financial ties, safety questions and clinical testing still central to judging the promise.
Citrini Research is recasting its AI investments around relative value as it warns that financing and construction constraints threaten returns. Its reported Oracle–Nebius trade faces a crucial counterargument: Nebius has substantial customer commitments, although delivery remains conditional.
