SAN JOSE, California – Cisco Chief Product Officer (CPO) and President Jeetu Patel expects approximately 70% of the networking giant’s product portfolio will be fully AI generated by the end of next year, though humans will remain a key cog in that push.

Patel explained to SDxCental in an exclusive interview at Cisco’s headquarters in San Jose, California, that this “AI-first standpoint” has already begun. That spark was Cisco’s AI Defense product, which the vendor unveiled in early 2025, and that Patel said “has been built 100% with AI. No human lines of code.”

Cisco’s next goal is “we will by the end of the year probably have a half -a-dozen products, and by the end of next year, my thinking is we'll have about 70% of our entire portfolio at Cisco by the end of the calendar year ’27, they'll be fully built with AI, no human lines of code,” Patel said.

This push is backed by 85% of Cisco engineers already using AI “on a regular” basis, highlighting what Patel said was a “massive proliferation of AI across the entire stack.”

One source of this proliferation is through OpenAI’s Codex AI coding agent platform. Patel had previously noted Cisco was “exploring” Codex, testing the tool’s potential as it sought a 10-times increase in productivity from “the human workforce,” productivity aspirations that also include Anthropic’s Claude Code.

“When Claude Code writes Claude Code and Codex writes Codex, there is a snowball effect that happens. They go really fast,” Patel said. “And so all of our products … Codex is writing Cisco's products right now … and Claude Code is writing Cisco's products.”

And what about that 15% of engineers that are not using AI? “They might be deep down in the kernel, and … we're being careful about how we go there,” Patel said

AI code still needs humans to blame

But, back to that majority.

Cisco’s AI-fueled product push will require a complete re-think of how those products are currently assembled and how code is managed. Patel did provide a bit of light for humans, noting that “the human having ownership of an AI app, that’s an app that’s built with AI coding, is super important,” but later explained how that new model will be constructed.

“You have to prepare for a world where humans will not know the code that's being written because you will not be able to review all the code that's being written by AI,” Patel said. “So what you have to do is you have to actually make sure that you automate the test beds and harnesses for measuring the outputs of the code.”

This will be required as that code “within the next year becomes a black box,” Patel explained. “When I say black box, I mean if you want to change code, you’re just going to go to AI and say, ‘go ahead and change this piece of code to have this outcome.’ It will go refactor in the way that needs to be refactored.”

However, this model will require the requester to be “very good at saying ‘now, I’m going to measure the outcomes in a way that there’s no regression,’ and that I think is the way this will go,” Patel added. “Your automation will get to be much more effective on the outputs.”

Patel noted that this does lead back to having the right model structure in place as “you have to create a synthetic environment where the outputs can be managed, can be measured for the sanctity of the output much more so than actually evaluating every line of code. That’s a big shift in a way of thinking.”

And also having a human minder nearby as “you also need to make sure that whoever is evaluating the outputs is the one that's responsible for the code, even though they don't know the code,” Patel said. “Every piece of code needs to have a provenance of a human, where you can track it back to a human.”