What is AI model routing? AI model routing is the practice of automatically sending each task to the cheapest AI model capable of handling it, rather than defaulting every query to an expensive frontier model. Cisco just put the idea in the spotlight: its CFO Mark Patterson told Fortune the company will give all of its roughly 90,000 employees a personalized AI agent starting at the end of July 2026, with a system that routes each task to the most cost-efficient model instead of always reaching for the top-tier one. The economics behind that decision are exactly what commercial real estate firms should study before rolling AI out across their own teams. It builds on the theme in our coverage of why CRE wins with a company brain, not a bigger model. Updated July 2026.
Key Takeaways
- Cisco will give each of its roughly 90,000 employees a personalized AI agent, one of the largest enterprise agent deployments announced to date.
- The core cost lever is model routing: sending simple tasks to lighter, cheaper models and only escalating to frontier models when the task truly needs it.
- Agent workflows can consume hundreds of thousands of tokens per task, so defaulting every step to a frontier model is how AI bills spiral.
- Cisco built much of its stack in-house and on-premises for cost control and data security, a tradeoff CRE firms with sensitive deal data should weigh.
- The lesson for CRE is to deploy AI by task and cost, not by always buying the most powerful model, and to start where the return is clearest.
What Cisco Actually Announced
Cisco is rolling out a personalized AI agent to every one of its approximately 90,000 employees, and the design choice that matters is how it controls cost. Speaking to Fortune, CFO Mark Patterson said the agents will route each task to whichever model is most cost-efficient, using a lighter, faster model for simple requests and a more capable model only when complexity demands it. Cisco built most of the infrastructure itself, favoring an on-premises approach for greater control over both cost and data. Patterson pointed to finance as an early proof point, noting AI now produces the large majority of the first draft of the company's management discussion and analysis, the narrative section of its public filings, and that his team built a tool to analyze Cisco's financials against competitors' earnings calls. You can read the original report at Fortune. The scale is notable, but the reusable idea is the routing logic, not the headcount.
Why Model Routing Matters for the AI Bill
Model routing matters because AI agents are far more token-hungry than chatbots, and paying frontier prices for every step is how costs run away. A simple chatbot exchange may use only a few thousand tokens, where a token is the unit of text AI models process and bill by, while an agent that plans and executes a multi-step task can consume hundreds of thousands or even millions of tokens in a single run. Frontier models are priced at a large premium per token over lighter models, so routing a routine task, such as reformatting a rent roll, to a cheap model and reserving the expensive model for genuine reasoning can cut the bill dramatically without hurting quality. This is the same cost pressure we covered when AI labs moved to tiered and metered pricing in our look at tiered AI pricing and what it means for CRE. The strategic shift across the industry in 2026 is from best model wins to best fit wins, and routing is how you operationalize it.
The CRE Translation: Route Tasks, Do Not Default to Frontier
For a commercial real estate firm, the Cisco lesson is to match each AI task to the cheapest model that does it well, rather than paying frontier prices across the board. Most of the AI work in a CRE shop is routine, such as summarizing a lease, extracting numbers from a rent roll, drafting a standard email, or reformatting an offering memorandum, and a lighter model handles that at a fraction of the cost. Reserve the frontier model for the genuinely hard work: complex underwriting judgment, nuanced investment memos, or reconciling messy financials. Ask your team, or an AI assistant, to sort your recurring AI tasks into a simple tier of light versus heavy, then route accordingly, and you capture most of Cisco's savings without building custom infrastructure. Cisco's own proof point, using AI to draft filings and analyze competitors, maps directly onto CRE asset management and investor reporting, a workflow our guide to AI agents for the CRE back office explores. The AI Consulting Network specializes in helping firms build exactly this kind of cost-aware AI deployment.
Build vs Buy and the Data-Control Question
Cisco built its own on-premises stack for control, but that is the one part of its playbook most CRE firms should not copy wholesale. Cisco is a global technology company with the engineering depth to build and run its own AI infrastructure; a typical real estate firm is not, and for most, the right answer is buying well-governed tools rather than building. The genuine and transferable question is data control: CRE firms handle sensitive material such as rent rolls, tenant financials, and off-market deal terms, and they need to know where that data goes when an AI tool processes it. You do not need an on-premises data center to get this right; you need clear data-handling terms, appropriate model settings, and governance. For firms with strict requirements, our coverage of on-prem AI and CRE data residency lays out the options, and the discipline of AI agent governance for CRE firms becomes essential as agents act on real data. Match your data-control approach to your actual sensitivity, not to Cisco's scale.
What CRE Firms Should Do Now
The practical move is to treat AI deployment as a cost and governance decision, not a race to buy the most powerful model. Start by inventorying where AI already helps your team, then tier those tasks by difficulty and route each to an appropriately priced model, whether through a platform that routes automatically or by simply choosing the right tool for each job. Next, set clear rules for what data can go into which tool, so you protect sensitive deal information from day one. Then pick one high-return workflow, such as lease abstraction or investor reporting, prove the value and the cost, and expand from there rather than deploying everywhere at once. For CRE investors who want help designing a cost-aware, well-governed AI rollout, reach out to Avi Hacker, J.D. at The AI Consulting Network, and see our broader AI tools for real estate investors guide for the tool landscape. Cisco's headline is the scale, but the strategy worth copying is deploying AI by task, cost, and data sensitivity.
Frequently Asked Questions
Q: What is AI model routing and why does it save money?
A: Model routing automatically sends each task to the cheapest AI model capable of handling it, using a light model for simple work and a frontier model only when needed. Because frontier models cost far more per token and agent tasks use enormous numbers of tokens, routing routine work to cheaper models can cut AI costs sharply without reducing quality.
Q: Should a CRE firm build its own AI stack like Cisco?
A: Almost never. Cisco is a technology company with the engineering depth to run its own infrastructure; most CRE firms are better served buying well-governed tools. The transferable lesson is not building, it is cost-aware model routing and disciplined data control, both of which you can achieve with commercial tools.
Q: How does Cisco's rollout apply to commercial real estate?
A: Cisco showed that most enterprise AI work is routine and can run on cheaper models, with frontier models reserved for hard reasoning. CRE firms have the same mix: routine lease and rent-roll work versus complex underwriting. Routing tasks by difficulty, and protecting sensitive data, captures most of the savings without Cisco's scale.
Q: What data-control questions should CRE firms ask before deploying AI?
A: Ask where your data goes when a tool processes it, whether it is used to train models, what the retention terms are, and which model settings protect confidentiality. Given rent rolls, tenant financials, and off-market terms, CRE firms should set clear rules for what data enters which tool and govern agents that act on real data.