In an era where artificial intelligence is reshaping everything from healthcare to entertainment, networking is no exception. But beyond the buzzwords and marketing gloss lies a world of deep technical innovation, where real problems are being solved for real customers.

To explore this intersection of AI and enterprise networking, we sat down with Jose Tellado, VP Fellow of AI Ops at HPE Aruba Networking. Jose brings a unique perspective as both a founder of an AI start-up, and a veteran of wireless communications. In this wide-ranging conversation – with Carlos Gomez-Gallego, CTO APJ, HPE Aruba Networking, and Dion Ryff, Engineering Manager, Matrix CNI - he dives deep into use cases, development cycles, self-healing networks, and even how AI helps his kids with homework (and don’t be surprised; stats vary across research, but it’s anywhere from 75 – 86% of kids using AI to assist with schoolwork!).

Here’s what we learned.

Q: Jose, how did you end up at Aruba and what sparked your interest in AI for networking?

Jose: I founded an AI for networking company back in 2014, but as a small company, we struggled with access to the kind of data we needed to train our models. Aruba approached us in 2016, offering access to a vast dataset—something we couldn’t pass up. Since then, I’ve focused on how AI can address real-world networking problems, especially in wireless environments where configurations are complex and under-optimised.

Q: There’s a lot of hype around AI. How do you cut through the noise?

Jose: The key is use cases. We’re not building AI for the sake of AI. We start with a customer pain point, gather anonymised data, build models, deploy them, and - most importantly - track whether users find them helpful. AI should focus on solving problems, not chasing headlines.

Q: What were the first big AI-driven use cases at Aruba?

Jose: We started with wireless connectivity - issues like roaming failures and suboptimal RF profiles - because wireless networks tend to be more variable than wired ones. Over time, we expanded to wired networks, SD-WAN, and even into data centre insights. Today, AI also helps monitor the health of Aruba’s own cloud services, which are increasingly part of the customer experience.

Q: What does the AI model development lifecycle look like?

Jose: If telemetry is available, we can go from idea to production in about two months. For example, our DFS (Dynamic Frequency Selection) channel recommender was built in that timeframe. If new telemetry is needed - say, for application health metrics - it can take three to six months. However, we always follow a structured approach: define the use case, collect and clean the data, build the model, test it, and monitor feedback.

Q: How are customers involved in model refinement?

Jose: Increasingly, we work directly with our global field teams and SEs to validate and improve models. We also reach out to customers when we detect anomalies and ask them to confirm our findings. This feedback, especially in Education environments (as an example), where seasonal network patterns are common, helps us fine-tune the AI to avoid false positives.

Q: Beyond connectivity, what other AI use cases are you excited about?

Jose: Firmware recommendations are a big one. Most customers are running outdated or vulnerable software. Our AI models analyse upgrade and downgrade patterns, known bugs, and CVEs to recommend the safest firmware for each device type. We’ve also automated client device classification using distilled LLMs, replacing thousands of hand-written rules with a model that identifies device types with 99% accuracy.

Q: Are you seeing interest from customers in training their own models?

Jose: Some! We’ve had a couple of requests from universities with computer science departments. They requested access to Aruba’s telemetry data on their own environment for academic research or internal model development. It’s not common, but it’s growing. It is also worth adding that we won’t ever broadly share customer data with other customers, and in the case of their own data all end user information was stripped out.

Q: What are the essential skills network professionals should focus on as AI becomes more prevalent?

Jose: AI isn’t replacing network engineers - engineers who use AI are replacing those who don’t. Soft skills like critical thinking, curiosity, and willingness to engage with AI tools are key. Technically, familiarity with data and reasoning models will help. Internally, we’re also seeing a shift toward managing AI agents, not just hardware or VMs. The landscape is evolving fast.

Q: How close are we to fully self-healing networks?

Jose: We’re getting there. Today, we offer partial automation with human-in-the-loop systems. For instance, we can auto-reset APs in bad states, and recommend CLI commands to fix misconfigured gateways - but we still ask for user confirmation. As confidence in the models grows, full automation will follow. But we’re always cautious with changes that could impact thousands of users.

Q: How has 6 GHz adoption impacted your AI recommendations?

Jose: Funny story - about 75% of 6 GHz-capable APs initially had the band turned off because admins forgot to enable it. We built a model that detects capable clients and predicts the potential throughput gain. It now helps customers understand the benefits and adjust configurations. Still, many have 6 GHz set to 160 MHz bandwidth, which isn’t ideal in all environments, so we guide them to smarter defaults.

Q: Any fun or personal ways you’re using AI outside of work?

Jose: Honestly, I don’t have much free time! But AI is great for helping my kids with their homework when I’m too tired! These reasoning models are like virtual tutors. Better than just giving the answer, they walk through the problem step-by-step. It’s a lifesaver after a long workday.

Final Thoughts: AI That Works, Not Just AI That Wows

Jose’s insights offer a refreshing view of what meaningful AI looks like in the world of networking. The key themes are clear: Start with use cases, ground solutions in data, test with real customers, and focus on outcomes - not just innovation for innovation’s sake.

From improved connectivity and firmware hygiene, to predictive analytics and semantic search for support cases, Aruba is embedding AI throughout the networking lifecycle. And perhaps most importantly, it’s doing so in a way that empowers human operators rather than replacing them.

As Jose put it: “AI isn’t coming for your job - people who know how to use AI are.”