Cisco wished to make generative AI obtainable to each single worker in a safe, cost-optimized method. To do that, Cisco IT developed an inner AI-powered assistant — purpose-built with safety — designed to raise how workers worok. Extra than simply one other device, Cisco’s inner AI assistant drives vital effectivity and frictionless experiences for workers – with 73% of customers reporting elevated productiveness and a median time financial savings of 5 hours per week.
As AI instruments like ChatGPT emerged within the market, we noticed each alternative and threat. Whereas Cisco acknowledged the transformative energy of AI and the outcomes these instruments may allow for our group, the rising utilization of public AI options additionally raised crucial security and safety considerations for enterprises.
Seven in ten executives imagine generative AI within the enterprise introduces new safety dangers for firm information and are involved about workers inputting personally identifiable data (PII), shopper information, or propriety data into engines like ChatGPT and Google Gemini (Salesforce, 2023). Counting on exterior AI providers may expose delicate Cisco information to public coaching units — a threat we weren’t keen to take.
We would have liked an answer that made generative AI obtainable to each single worker, regardless of the place they work from, whereas holding Cisco information safe and inner.
First steps to a safe resolution
From day one, our precedence was holding Cisco’s information safe and in-house, not in another person’s coaching set. The unique structure Cisco’s Growth Crew launched in April 2023 was designed to supply a safe surroundings for workers to securely experiment with AI instruments. It was an enterprise-wide resolution powered by Azure OpenAI fashions that maintained a dialog immediate retailer, audited prompts in actual time, and routinely redacted personally identifiable data (PII) previous to processing.
Eight months after launch, the Cisco IT design workforce stepped in to reinforce consumer expertise and drive broader adoption. Initially, solely 30% of the workforce — primarily technical workers — had been utilizing the device. The objective was to make sure all workers, no matter their technical background, may gain advantage from it.
The workforce took a user-centric and data-driven strategy, conducting foundational and generative analysis to completely perceive what workers wanted from this device. By leveraging all of the obtainable information about consumer ache factors, figuring out what workers had been trying to find, and evaluating it with the highest areas of friction, the design workforce constructed out key use instances for the assistant. These included options to assist workers brainstorm, code, translate, generate content material, find out about Cisco merchandise, and extra — in the end lifting the load of on a regular basis work.
This analysis revealed that the primary factor workers had been on the lookout for was IT help — which means the true potential of Cisco’s inner AI assistant might be unlocked by means of ingesting Cisco information. This turned the best precedence.
Shifting ahead to help workers
We began by ingesting Cisco Folks and Communities and Assist Zone information (articles, web sites, and stories) into the device. We had Retrieval Augmented Era (RAG) options constructed for Cisco.com and SalesConnect (inner content material portal for all gross sales sources) purposes earlier. The identical capabilities had been added to the interior AI assistant as effectively to gauge the wants of gross sales customers — a audience our workforce wished to deal with.
With this high-quality Cisco information ingested within the device, workers can use AI for help case troubleshooting, inventive writing, information evaluation, questions, and even technical duties like community monitoring and configuration. It might additionally streamline the onboarding course of for brand spanking new hires by supporting laptop computer or cell phone setup and answering questions on expense stories.
Establishing an ongoing working mannequin
Cisco’s inner AI assistant was constructed for speedy integration, sturdy safety, and actionable insights, all at a worldwide scale. That is supported by an ongoing working mannequin the place our workforce constantly displays consumer suggestions and delivers options with agility, creating an iterative, full-circle course of.
To design an inner assistant greatest fitted to Cisco’s international workforce, we developed a layered structure round these elements:
- AI agent, MCP, and enterprise search platform: The AI Agent and Mannequin Context Protocol (MCP) platform hosts all of the AI brokers and MCP servers developed utilizing open-source AI frameworks. Its modular, plug-and-play structure offers the Enterprise AI/ML engineering workforce the power to onboard new AI brokers and search purposes with out operational overhead. The platform additionally gives LLM-as-a-service and RAG-as-a-service providing to over 4300+ groups throughout Cisco.
- Enterprise AI agent registry and MCP registry: This registry permits workers to register and join all distant AI brokers and distant MCP servers developed by groups throughout the enterprise with standardized agent and MCP safety, agentic protocols and entitlements in order that workers can securely entry this linked agentic mesh from the interior assistant.
- Hybrid multi-cloud mannequin and AI agent orchestration: Utilizing a federated strategy, our inner AI assistant connects to a number of Cisco AI brokers and main AI fashions comparable to Azure OpenAI, Claude, Google Gemini and Cisco’s inner giant language mannequin — Deep Community Mannequin. The platform routinely routes every request to the most effective obtainable AI mannequin for deployments primarily based on workload, latency, reliability, and use-case. This ensures customers at all times get quick, correct outcomes.
- Constructed-in safety, information governance, and observability: Safety is woven into each layer of the device, making certain our office stays resilient in opposition to evolving threats whereas enabling innovation. Cisco’s AI Protection guardrails are used to scan consumer prompts and MCP server instruments for malicious code and hidden or missed threats.
- Superior information operations and AI-driven enterprise impression: Area particular AI brokers have been developed to allow clever automation of enterprise processes by leveraging AI prepared enterprise information.
This platform, alongside our ongoing working mannequin, permits us to deploy new options and effectively and constantly refine our roadmap — making certain we keep on the forefront of AI innovation.
“Designed with a security-first mindset, our inner AI assistant has highly effective guardrails that routinely detect and redact delicate information earlier than it’s despatched to a public AI mannequin — giving our workers an clever associate they’ll belief.” – Sujith Joseph, Principal Engineer, Cisco
Driving higher effectivity throughout our workforce
Since its launch, Cisco’s inner AI assistant has dealt with greater than 21 million interactions, serving 100,000+ customers throughout our international workforce.
Cisco’s inner AI assistant is greater than only a chatbot. It has rapidly grow to be essentially the most extensively used AI device at Cisco, facilitating a median of 156,000 each day interactions and driving higher effectivity as 73% of customers reported elevated productiveness and a median time saved of 5 hours per week.
That is made attainable by options and developments at the moment obtainable within the device, together with:
- Entry to the newest AI fashions
- AI brokers and LLM orchestration
- Information and picture evaluation
- PowerPoint era
- Private RAG (RAG on consumer’s private information in OneDrive, and so on.)
- Limitless Deep Analysis (Analysis stories utilizing Cisco inner information and internet information)
- Information evaluation AI agent
- Enterprise AI Agent and Connector Registry (Integration of distant AI brokers and MCP servers)
Classes discovered alongside the way in which
AI instruments are solely ever as dependable as the information they’re skilled on. One problem we confronted early on was making certain that we had been solely ingesting clear, high quality, and structured information into the device. Something much less may actually lower the standard of the responses and the usefulness of the answer. The Dev Crew performed continuous and rigorous testing to assist be sure that the assistant doesn’t hallucinate and constantly gives updated and correct responses.
Moreover, driving adoption throughout our international workforce introduced challenges, particularly amongst AI-cautious or non-technical workers. Regardless of the answer’s approval for extremely confidential Cisco information, there was a stage of apprehension surrounding its use and potential misuse. This spurred an effort to upskill our workers by creating efficient demos, enablement sources, and rising normal AI consciousness.
Our user-driven, design-led strategy was essential for navigating these challenges and allowed us to construct a device that caters to the wants of our numerous workforce. We proceed to finish generative analysis to know how our workers use the AI assistant and the way it will resonate with them of their day-to-day work.
From product to platform
AI is remodeling industries at an unprecedented charge. Whereas most organizations acknowledge the pressing must undertake AI applied sciences and instruments, many aren’t absolutely ready to successfully leverage AI. At Cisco, we all know that AI is not a futuristic idea. We’re constantly innovating internally — creating AI-powered instruments, securing them, and in the end empowering our workers to do their greatest work on daily basis.
Constructing on its sturdy basis, Cisco is concentrated on evolving the AI assistant from a product right into a dynamic platform, empowering workers to have a hand in shaping their AI expertise. The Agent Registry and upcoming Agent Builder options will allow customers to tailor the device to greatest match their wants at work. Not solely will this enhance the general usefulness and worth of the AI assistant, however it’ll additionally assist upskill our workforce, constructing a extra technically proficient and AI-fluent worker inhabitants.
As a dedication to Cisco’s funding in AI, the Cisco Automation and AI Middle was not too long ago created to deal with increasing the adoption of AI. Trying forward, the imaginative and prescient for the interior AI assistant is to grow to be a dependable, always-on associate that may seamlessly combine into each workflow, anticipate wants, and allow groups to deal with what issues most.
The inner AI assistant isn’t simply altering how we work at Cisco; it’s a cornerstone of constructing a future-proofed office that’s safe, adaptable, and prepared for what’s subsequent.
Extra sources:
