Most organizations with network operations today believe they have a good handle on network observability—they’ve progressed past legacy tools and have excellent data collection within the four walls of their data center. They are masters of packets, latency, faults, and performance logs for their internal, managed infrastructure. Yet, when you ask them to assess their maturity, they rate themselves not as highly.
According to the State of Network Operations :2026 from the Broadcom NetOps group, the reason for this self-assessment is that two-thirds of respondents report persistent cloud and internet blind spots. Organizations are outsourcing the majority of their service delivery path to networks they don’t own, and 87% of them still cannot see beyond the user’s wireless router. This lack of visibility renders network operations teams virtually helpless when a user calls to complain about voice drops or application slowness. Without seeing the external network hops—the Verizon Fios, the Comcast, the path through various ISPs to Google Cloud or Microsoft Azure—triage is impossible.
Organizations haven’t taken that visibility outside of their data centers in part because there aren’t many service providers that can help them gain that visibility. According to Jeremy Rossbach, chief technical evangelist at Broadcom NetOps, there are two: Cisco, which provides outside-in visibility, and Broadcom, which offers inside-out visibility into blind spots. “There are multiple, multiple hops” along the network to connect someone at home to a meeting application, he said. “The first network hop would be your wireless router, and then the second would be to the residential ISP outside of your house, all the way to connecting to a cloud provider and the WebEx. So I have blind spots right now, because I’m not running the software that would monitor this. But my blind spots are everything after my wireless router, which means if I call and complain to Broadcom that I just got interviewed, and the guy said that my voice was dropping out every 30 seconds, the Broadcom guy would be like, I don’t know how to help you. You’re not on our network.”
Meanwhile, according to the survey, 92% of network leaders plan to deploy AI-enabled network observability solutions. But if that effort comes before you have a foundation for data collection, it’s “like putting the cart before the horse,” Rossbach said. AI is only as good as the information it consumes. If you are feeding your AI engine a network dataset with massive blind spots, the result will be an AI remediation assistant with some blindness.
The answer, Rossbach said, is to create an “AI-ready network.” This involves creating inside-out visibility, using automation to scale, and using AI for predictive analysis.
Inside-out visibility involves deploying a software agent on the client or in the cloud, capable of mapping the entire network path from the client to the application destination. This capability generates what we call “network innocence.” It’s the ability to pinpoint the source of an issue, whether it’s a router on a public IP address or a server in a cloud network having 60% packet loss. This real-time, undeniable evidence builds the trust required for the next stage.
As for automation, the survey found that only 27% of organizations have a mature automation practice. The primary hurdle is a lack of trust—no one wants to automate a remediation that could take down a branch office based on incomplete or suspect data. By resolving the visibility crisis, we create a data set that is trusted, allowing the AI to provide predictable, trustworthy recommendations. Once you establish that trust, you can automate the responses.
Finally, instead of acting blindly, the AI system can ingest a complete, trustworthy data set to find solutions in seconds that would take human engineers hours. This is how AI shifts from being a mere troubleshooting tool to a system that provides true foresight, suggesting necessary config rollbacks or network path changes before a failure impacts the customer.
Looking to the near future, Rossbach envisioned a time when “maybe one day we won’t have to collect data and store it. Maybe one day somebody calls and says, ‘Something isn’t working.” And I say, ‘Hey, my AI Assistant, can you figure out why? Someone is calling and saying she can’t get to Gmail,’ and it’s smart enough to go get real time performance data or utilization data and map out the network hop. But for now, we’ve got to give it a head start, which is data. It’s got to learn. We’ll get there one day.”
