Cisco's Four Pillars for Agentic AI Transformation

Cisco runs a $15 billion services business with 20,000 employees, and Liz Centoni, EVP and chief customer experience officer, is rebuilding it around agentic AI — starting with workflows, not tools. "We didn't just start bolting on AI," Centoni says. "We looked at the underlying workflows to understand how we could redo work."

A 26-year Cisco veteran who led the company's $28 billion acquisition of Splunk in 2024, Centoni now guides Cisco's push to become an AI-native company, both internally and in what it sells. Her governing principles are blunt:

• Every workflow constrained by human availability is a workflow for redesign.
• The status quo is not an option — you can't rest on laurels while revenue grows and margins look good.
• Today's quality is the floor, not the ceiling.
• Leadership, not technology, is the limiting factor.

What went wrong with Cisco's first AI agent?

When Centoni took her current role two years ago, she read through customer feedback. The biggest pet peeve was the path to the right expert — repeated handoffs from engineer to engineer, explaining the problem from scratch each time.

Her team's first instinct was to build an agent that sped up the handoff chain. It backfired. "It just upset our customers more because it showed that we actually did five handoffs behind the scenes and it surfaced it to them faster," Centoni says. "We just bolted AI onto the wrong thing and said, 'automation.'"

The real question, she argues, was why handoffs existed at all. Some workflow steps were designed decades ago to solve problems an agent can simply cut through. "Users are an essential part of the technology stack," she says. "You have to bring in the subject matter experts and the users to outline the workflow, and you have to get close to the work. If you can't, you have no right designing."

That requires drawing careful boundaries: where can an agent act autonomously, where must a human take over, and where can automation and AI hit 90%-plus accuracy?

How do you change the cost of expertise?

Historically, Cisco threw experts at customer problems — the more complex the engagement, the more experts. "When I came into this business, we treated every customer we had like a bespoke model," Centoni says. "But over the last 40-plus years, we've solved these problems tens of thousands of times."

Cisco has distilled those decades of engagements into a knowledge base and playbooks that both AI agents and human experts use to plan and execute deployments. Feed in a customer's data and project KPIs, and the system customizes value realization without stacking on human experts.

"You go into a new deployment with an expert agent that brings in all the prior experts and democratizes the knowledge base," Centoni says. "The outcome for the customer is consistency, security, and better time to project completion." Humans stay in the loop, freeing bandwidth for conversations on iteration.

Can AI prevent the outage entirely?

Cisco's focus used to be root-cause speed: if analysis took three hours, get it to 30 minutes. Now agents monitor customer environments continuously, watching case-history clusters that recur and recommending next best steps before issues turn critical. "In 12 months, I'd like to be able to sit down with a customer and say you would've had 10 outages last year and we prevented eight of them," Centoni says.

Which model, and why Mistral?

With boards now demanding ROI rather than adoption stats, Centoni insists on fit-for-purpose tooling. "Before we even go down the AI path, we ask, 'Can automation do this?'" she says. "If it's purely rules-based, I'm doing automation. If it requires bringing multiple signals together to drive a prediction, then AI." Often it's a mix of older predictive models, automation, and newer LLMs.

Cisco's AI assistant for renewals — combining a renewals agent, sentiment agent, risk agent, and install base agent — runs on Mistral. "They were one of the few who could give us an on-prem model that had the full capabilities," Centoni says. Cisco customized it on a private instance, with automation alongside.

How do you measure success without compliance theater?

Plenty of leaders track AI adoption. Centoni calls that a trap: users check the box, then finish the work with old tools — "compliance theater." Her metrics are whether workflows actually changed and whether the renewals team shows greater capacity since deployment.

The deeper point: agents never ship feature-complete. "With software, you build, test, deploy it, and it's 100% feature complete," she says. "With an agent, it's 60% of the way there in its first iteration in terms of usability, and the user is part of the tech stack that gives us feedback." Without a strong feedback channel, an agent can't become usable at scale.

Cisco's internal support agent shows what that looks like in production. The moment a customer opens a case, it pulls their full history and context — sparing the support worker the first 30 minutes of gathering details. "That's a lot of time and frustration saved up front," Centoni says. "It also provides the next best action."