6 Min Read
"Self-driving network" is a term that executives at Juniper Networks were throwing around long before the dawn of agentic AI to describe a fully autonomous system, capable of making decisions and executing tasks without any human involvement. Now, about 16 months since the company was acquired by HPE in a $14 billion deal, and with agents mushrooming across the planet, it suddenly appears a lot closer.
Among other things, the Juniper acquisition in July 2025 came with an AI called Marvis, attached back then to Juniper's Mist portfolio of network technologies. The goal was to use Marvis across other products, as well, said Rami Rahim, the president of HPE Networking (and former Juniper CEO), during a press conference at the Winter Olympics in Milan, where HPE was flaunting some of its expertise earlier this year. "Marvis now is not exclusive to Mist," he told reporters. "Marvis is the AIOps engine that will operate across all of HPE networking technology."
To an outsider, Marvis is probably best imagined as the brain of the entire system. Experts at Mist, which Juniper itself bought in 2019, began training it about a decade ago on vast quantities of data so that it could figure out network problems and fix them without recourse to people.
"We have been putting support tickets into our AI model for ten years plus now," said Mittal Parekh, a senior director of product and solutions marketing for HPE. "Back in 2016, we started with only wireless – only Wi-Fi. We took the tickets, we put them in Marvis, Marvis was able to solve them, and we delivered self-driving networks on the heels of the trust that we built."
By 2019, when Mist was acquired by Juniper, Marvis was being fed with data relating to the portfolio of wired and switching technologies. In 2024, Marvis gained fresh expertise when Juniper pointed it at Apstra, an intent-based networking system designed for data centers that feature technologies from multiple vendors.
Reverse hydra
HPE evidently preferred Marvis to any brain it had tried nurturing before the Juniper takeover. But using Marvis to control elements of the Aruba Central portfolio – which offers the same kind of enterprise Wi-Fi and related products sold by Juniper Mist – would still have kept it within the perimeters of the campus edge. Networks outside it would not have looked very autonomous.
The day envisaged by Rahim in Milan now seems to have arrived following an important update this week. It brings EdgeConnect, HPE's lineup of software-defined wide area network (SD-WAN) and secure access service edge (SASE) products, into the same fold as Aruba and Mist. That creates a kind of reverse hydra – not one serpentine body sprouting multiple heads but a single brain controlling various entities and limbs.
It also gives HPE something unique, insisted Parekh. "As of today, none of my competitors, not Cisco, has a truly self-driving network," he said. "We are the only ones in the industry right now able to deliver this." ServiceNow, a US cloud-computing company, has already been able to eliminate about 90% of helpdesk tickets by adopting the HPE technology, he said.
Cisco might have something to say about such a bold claim, but Marvis does already seem to have a long history in the context of AI, and the use of it across the combined portfolio appears to have been an objective ever since the Juniper deal was completed. Marvis could even be viewed as the answer to the question of what stays and what goes in the Mist and Aruba portfolios following the HPE and Juniper integration.
"I want to dispel whatever you have heard in the market that only one platform will remain, and we are writing ten-year letters of commitment to our customers on the Aruba side as well as the Mist side," said Parekh. With the same brain for each product set, HPE can feasibly avoid having to make other adjustments and maintain both Aruba and Mist in their entirety.
The big attraction for customers, according to HPE, is the ability to control disparate technologies and environments from a single interface. "While all my competitors are forcing them to access multiple different consoles – multiple different systems and data stores in the back end – in order to understand why someone is having a bad experience, we are bringing it under one single roof," said Parekh. All that should result in savings for chief financial officers who are keen to trim IT budgets, he thinks.
Learner driver
Even so, in the wake of some increasingly negative publicity about agentic AI, taking people entirely out of the loop might seem like folly. Anthropic and OpenAI, the US developers behind the world's frontier AI models, continue to issue warnings about the danger posed by unfettered AI. In recent weeks, as if to illustrate the risks, agents built by those companies have reportedly hacked into other IT systems, despite the fact that human programmers had apparently not given explicit instructions to do so.
Critics have dismissed the doomsday scenarios as a scaremongering attempt to make AI seem more capable than it really is ahead of high-profile initial public offerings. But talk of ceding control to AI has clearly generated anxiety, not least because it has implications for the role of people.
Parekh says Marvis was trained as rigorously as a teenager being taught to drive by a watchful instructor or parent. "First, I will sit next to you in the daytime on small streets, then I will sit next to you in the daytime on a freeway, then I'll sit with you at nighttime on small streets," he elaborated. Wary organizations can also adopt versions of the HPE technology that keep people in the loop. Promotional slides describe "human-assisted" or "controlled self-driving" options for companies worried about letting go of the wheel.
As for the necessary guardrails to prevent agents from running amok, Parekh says HPE has been careful to withhold certain privileges from its agents. "The keys to the kingdom will only open one door," he said. "You don't have a master key. You can't run wild because you have a defined job based on your tools and skills."
Notwithstanding a drop in early October, HPE's share price has performed strongly throughout 2026, rising more than 140% since the start of the year amid the excitement about AI. Sales were up a third year-over-year for the July-ending quarter, to more than $12.2 billion, while net profit quintupled, to $1.54 billion. Networking was the fastest growing and most profitable part of the business, with margins of around 22%. In that networking contest against Cisco, Marvis could prove vital.
About the Author
International Editor, Light Reading
Iain Morris joined Light Reading as News Editor at the start of 2015 -- and we mean, right at the start. His friends and family were still singing Auld Lang Syne as Iain started sourcing New Year's Eve UK mobile network congestion statistics. Prior to boosting Light Reading's UK-based editorial team numbers (he is based in London, south of the river), Iain was a successful freelance writer and editor who had been covering the telecoms sector for the past 15 years. His work has appeared in publications including The Economist (classy!) and The Observer, besides a variety of trade and business journals. He was previously the lead telecoms analyst for the Economist Intelligence Unit, and before that worked as a features editor at Telecommunications magazine. Iain started out in telecoms as an editor at consulting and market-research company Analysys (now Analysys Mason).