The road to autonomous network goes through humans in the loop, says Venkat Pullela, Keysight’s CTO of Networking
With AI being on every CTO’s vision board, a question they often get asked is: how close are we to autonomous networks?
As much of the industry waits breathlessly for self-driving networks to arrive, increasingly, it’s starting to feel like — despite the milestones and the breakthroughs — a fully autonomous network, one that can think, learn, adapt, and take decisions all on its own without human intervention, might not be the goal after all.
Is AI enough on its own?
The concept of an autonomous network is seductive for good reason. AI agents armed with self-x capabilities performing sequential tasks without human help sounds like a fantasy. They can deploy, configure, monitor, maintain, and troubleshoot the network just as humans do — work on repeat and retire when they need to.
The construct, if you it break down, comes to three core elements — AI agents, closed-loop automation, and intent-driven networking — with no human in sight.
But as tempting as it is to leave the worries of a nebulous, ever-scaling network to an army of bots with capability and agency to get the job done, it has its risks. The part that no one’s talking about is that human oversight is still vital. It’s like a gut check to tell whether the work is getting done right.
Venkat Pullela, CTO of Networking at Keysight, described it well. He likened AI to an intern — it is here to help and learn from its mistakes, and should be treated as such.
While the rest of the industry rallies behind AI without holding back, Pullela likes to keep it real. “We’re not there yet, but we are kind of in that journey,” he said, talking about AI-driven fully autonomous networks.
Pullela who is associated with industry bodies like the Ultra Ethernet Consortium (UEC), sees both sides of the technology. As a public-facing executive working closely with hyperscalers, hardware vendors, and software firms that are actively embracing AI in all corners of their business, he’s privy to the real-world adoption scenarios.
“Practically everyone’s future plans at this time are based on AI. It’s here to stay. People are convinced. You don’t need to convince them anymore,” he said.
That marks a radical shift in mindset. He remembered not too long ago AI was an outlier that was thrown at infinite scale problems that humans could not solve: finding anomalies through pattern recognition, fuzz testing for spotting bugs and vulnerabilities in software programs, and mitigating errors through automation of repetitive tasks.
“AI was always part of it. We were only using it slightly differently,” he said.
Human-assisted autonomy
Post-ChatGPT, industries started considering tapping AI in earnest. Organizations started integrating the technology into various parts of the business and little by little it became part of its DNA.
For it’s part, AI lived up to the hype. It’s ability to proactively right-size and repair the network unlocked new levels of network autonomy not seen before.
Pullela is both optimistic and cautious of the developments. He acknowledges and applauds the role of conversational chatbots in democratizing data and intelligent multi-agent systems in lowering the technical bar for business personas.
However, he argues that the impact could be varied depending on the level of supervision. Unchecked implementation can lead to chaos, as has been witnessed by many companies in the past that have relied overmuch on AI coding tools for example, leading to irreversible mistakes, with losses sometimes running in millions; whereas a balance of AI-powered automation and human engagement can yield surprisingly positive results.
“The engineers still have to own the code they generate,” he said. “A solution engineer is able to show possibilities, but ultimately a middleware or a backend engineer has to come and build it. There is a hand off.”
In it’s third wave, AI is more intelligent and aware, but it’s still prone to biases and hallucinations, he cautioned. Agents can overdeliver in one task, and underdeliver in another.
Pullela emphasized that having guardrails in place is the next step as the industry progresses from network automation to autonomy — a matter that is still work in progress.
“The guardrails part of it is the biggest challenge. We are getting to explainable AI, but on the guardrail side, there is still not the same level of maturity,” he reflected.
Roles are evolving as well. As AI introduces low-touch networking, turning hands-on tasks like coding and testing into auto-complete functions, those human skills are going out of demand. Instead, employers are putting more emphasis on AI curation, prompt engineering, and strategic decision-making skills that can help tame the technology.
Pullela predicts that in the future, hardware will be shipped with out-of-the-box AI agents to ensure a white-glove experience right from Day 0. But that future will not be without humans in the loop.
“Autonomous networks and systems are the holy grail…We are not fully there yet, but my gut feeling is it’ll evolve with humans in the loop,” he concluded.
