Cisco's CFO just told Fortune that 80-90% of the company's mandatory financial report narrative is now written by AI. Not assisted by AI. Not AI-suggested-then-rewritten. The first draft — the one that goes to lawyers, then to the SEC — is predominantly machine-generated. Mark Patterson, who has spent 26 years at Cisco and became CFO in July 2025, called this the most significant technology transition of his lifetime. Starting this month, Cisco is rolling out AI agents to all 90,000 of its employees. Every single one.
That's not a pilot. That's an enterprise commitment.
Here's why this matters: Cisco isn't just deploying AI for productivity theater. They're rebuilding the operating model of a Fortune 83 company around agents that autonomously route, decide, and act — and their CFO is the one architecting what that looks like for the finance function.
The Scale That Makes This Remarkable
Cisco has approximately 90,000 employees across every function — engineering, sales, finance, legal, HR, operations. Starting in its new fiscal year at the end of July 2026, each employee gets a personalized AI assistant. Not a shared tool. Not a departmental chatbot. A personal agent capable of handling tasks, answering questions, and dynamically routing to the most efficient AI model for each request.
That last part is critical from a cost architecture standpoint. Patterson was explicit: "It's not going to burn a whole bunch of tokens with frontier models." Cisco's system dynamically selects the right model based on the task. Simple queries go to cheaper, faster models. Complex reasoning goes to frontier models. "It knows which tool is most effective and most efficient," he said.
This is how mature enterprise AI looks. Not "we use GPT-4 for everything." Not "we standardized on one vendor." A tiered, intelligent routing system that optimizes for cost and capability simultaneously.
On-Premises by Design
Much of Cisco's AI infrastructure is built on-premises. Patterson's framing: "We feel like that's the most efficient way is to build our own AI stacks, which will go out and query the different models based on the particular use case."
For enterprise leaders debating cloud-first versus hybrid AI deployment, this is a meaningful data point. Cisco — a company that literally builds the infrastructure the AI economy runs on — chose on-prem for its own AI stack. The reasons are straightforward: cost control, data sovereignty, and the ability to optimize without being locked into a single provider's pricing model.
This is especially relevant given that agentic AI tasks consume dramatically more tokens than standard chat interactions. A typical conversational exchange uses a few thousand tokens. Complex agent tasks — planning, tool calls, intermediate reasoning, error recovery — can consume hundreds of thousands to millions of tokens per run. At 90,000 employees running agents continuously, token costs aren't an abstract concern. They're a line item that needs architectural solutions.
The Finance Transformation — What 80% Really Means
Let me give you the specific ways Cisco's finance function has already changed under Patterson.
MD&A preparation. The Management Discussion and Analysis section of public company filings is one of the most scrutinized documents a public company produces. SEC lawyers review it. Analysts dissect it. Institutional investors parse every word for signals. At Cisco, 80-90% of the first draft is now AI-generated. Human reviewers then refine and approve the language. That's not a small shift — that's a fundamental reallocation of highly-compensated finance talent from writing to reviewing.
Investor relations. Cisco built an AI tool that analyzes its own financial history alongside competitor earnings calls and anticipates the specific questions that specific analysts are likely to ask. Patterson uses this to prepare for earnings calls. Think about what that means: instead of finance teams spending days manually pulling historical data and researching analyst focus areas, an AI model does it in minutes and surfaces the most likely lines of inquiry.
Personal benchmarking. Patterson has his own agent. He uses it primarily for competitive benchmarking — quickly comparing Cisco's performance against peers across revenue growth, EPS, R&D spend, and capital allocation. What used to require a team of analysts and days of data aggregation is now a dashboard-style interaction.
The CFO Cockpit. This is the one that caught my attention. Patterson and his team are building what they call a "CFO cockpit" — an AI-powered dashboard that synthesizes performance data across products, geographies, and customer segments. Not just reporting what happened. Predicting where the business is headed and recommending actions. If that gets built and works as described, the CFO role shifts from financial historian to forward-looking operator.
Why Only 17% of Finance Teams Are Here Yet
Here's the gap: Cisco is an outlier. Across the broader enterprise landscape, only 17% of finance teams are actively using AI in their core workflows. A staggering 68% of CFOs report they don't know where to start. Meanwhile, at the aggregate level, only 5% of enterprises achieve substantial AI ROI.
The Cisco case illuminates why the gap exists. Patterson didn't just deploy tools. He committed to company-wide upskilling, internal knowledge sharing, and deliberately fostering internal competition as teams discover new applications. The AI deployment is paired with an organizational change management effort. Most companies skip that second part, which is why their AI pilots stay pilots.
There's also the governance reality. The SEC's Division of Examinations has made AI-related disclosures a focus area for fiscal year 2026. Any CFO using AI for financial reporting needs verifiable audit trails, data lineage, and security protocols. Cisco's on-prem infrastructure strategy provides those controls in ways that cloud-only deployments complicate.
The $234 Billion Wake-Up Call
On July 1, Gartner released research that every CFO and CIO needs to read. Their conclusion: up to $234 billion of enterprise application software spending is exposed to what they call "agentic arbitrage" between now and 2030.
The mechanism is straightforward: as businesses increasingly deploy autonomous AI agents to handle workflows, those agents bypass the point-and-click interfaces that traditional SaaS vendors built their revenue models around. Why pay for 200 Salesforce seats when an agent can query the Salesforce API directly? Why pay for enterprise software licenses tied to user-based pricing when agents don't have seats?
This is a strategic risk for buyers who are overpaying for user-count-based software licensing. It's an existential risk for vendors who haven't adapted their business models. For CFOs evaluating enterprise software contracts right now, this is a negotiating lever: the agent future is coming, and seat-based pricing is on borrowed time.
Gartner also noted that over 40% of agentic AI projects will be canceled by the end of 2027 due to escalating costs, unclear business value, or inadequate risk controls. That's the failure mode Cisco is explicitly trying to avoid with its on-prem architecture and dynamic model routing. They're engineering the cost side from the start, not discovering it after deployment.
The Playbook for Enterprise Finance Leaders
The Cisco deployment isn't magic. It's a series of deliberate architectural and organizational choices. Here's what they did that you can replicate:
Start with high-value, high-volume documents. MD&A wasn't chosen randomly. It's a document that takes disproportionate skilled time, has a predictable structure, and has a clear human review stage before it goes external. That's the profile of an ideal AI-first workflow: high cognitive load, structured output, human sign-off built in.
Don't default to frontier models. Dynamic model routing that matches task complexity to model capability is the difference between AI that generates ROI and AI that generates AWS bills. For enterprise-scale deployment, this isn't optional.
Build on-prem or hybrid where data sensitivity demands it. Financial data, customer data, M&A information — these belong in environments you control. Patterson's on-prem preference reflects the reality that CFOs carry legal fiduciary responsibility for how that data is used.
Make it personal, not just departmental. Every Cisco employee gets their own agent. Patterson has his own benchmarking agent. The power of AI at scale isn't in one team's shared tool — it's in 90,000 people developing institutional knowledge about what works and competing internally to find better applications.
Pair deployment with upskilling. Patterson explicitly cited company-wide training and internal knowledge sharing. The tools alone don't produce outcomes. The organizational change management does.
What This Means for Technical Leaders
For CIOs and CTOs reading this: the Cisco architecture is a case study in enterprise-grade agent infrastructure. The key decisions are model routing logic (not all tasks need frontier models), infrastructure ownership (on-prem for control, cloud for elasticity), and token cost management at scale.
Agentic tasks are token-intensive in ways that simple LLM queries are not. An agent that plans, calls tools, retries on errors, and produces structured outputs can consume 50-100x the tokens of a conversational interaction. At enterprise scale, that math determines whether your AI program is profitable or a budget drain. Cisco's architecture explicitly addresses this.
The dynamic routing layer — the system that decides which model handles which task — is becoming a core infrastructure component. It's not just about cost. It's about reliability, latency, and capability alignment. Some tasks need GPT-4-level reasoning. Most don't. The enterprises that build intelligent routing now will have a significant cost advantage over those running everything through frontier models.
The Bottom Line
Cisco's CFO is running the largest real-world test of enterprise AI agent deployment happening right now. Ninety thousand employees. Every function. Personalized agents for each person. And on the finance side specifically, AI that has already transformed financial reporting preparation, investor relations prep, and competitive benchmarking.
The question isn't whether your finance function will look like this eventually. It will. The question is whether you're in the 17% building toward it now, or the 83% that will be catching up.
Gartner's $234 billion figure tells you the software contracts you're signing today will look very different in three years. Patterson's on-prem architecture tells you the cost controls matter as much as the capabilities. And Cisco's rollout tells you that at this scale, you need organizational change management as much as you need the technology.
The CFO cockpit is being built right now. The only question is whether your company is building it or waiting to license it from someone who did.
Sources: Fortune interview with Cisco CFO Mark Patterson (July 1, 2026); Gartner press release on agentic AI enterprise software risk (July 1, 2026); Gartner prediction on agentic AI project cancellation rates (June 25, 2025); unframe.ai Enterprise AI ROI benchmark of 255 enterprise leaders.
