Cisco GSX FY26 Recap: What’s Next at Cisco GSX FY27

No re-introductions, no warm-up small talk, straight into the thing you were both actually thinking about. That happened to us more times than we counted at Cisco GSX FY26, where we ran booth PS03 in Las Vegas, and it's the whole reason we keep showing up.

It's also why the conversations went somewhere useful. When you've known someone for a few years, you get past the pitch and into what actually worked, what didn't, and what they wish had happened six months sooner.

The Question That Kept Coming Back

One question came up more than any other, and it came up everywhere: how do you actually start with AI?

Not which model. Not which vendor. Just, practically, where do you begin.

Our Field CTO, Ben Haddox, heard it again a few weeks later at Cisco TechX Nebraska in Omaha, on stage with Cisco talking through AI outcomes and architecture with the local tech community. Same question, different room, different industry mix. That told us something. This is not a Cisco-conference talking point. It is what people are genuinely stuck on.

Here's the short version of what we tell them.

Don't start with the model. Start with a question your operations team already has to answer under pressure, and work out whether your systems can answer it today. Something like: when this service degrades, what caused it, and what else is about to break? Most enterprises find that the data needed to answer it exists, but it's scattered across network, infrastructure, application, and security tools that don't share a common view of what's connected to what.

That gap is the real starting point. An AI layer on top of disconnected data will produce fluent answers that nobody can act on, because the model is guessing at relationships instead of reading them. Establish the relationships first, then put AI on top of something it can reason about. It is less exciting than picking a model, and it is the step that determines whether anything downstream works.

"The appetite for AI is real, and the real test was never the demo. It's what happens when AI meets day-to-day operations."

Ben HaddoxField CTO, OnStak

What That Question Turned Into: AI Assurance

There was a second version of the same problem in almost every hallway conversation. Teams had AI running. It was working fine, technically. But nobody could tell them how to prove it was safe to keep running.

That question followed us home from Vegas and shaped most of what we built this year. We call it the Yes Layer.

In plain terms, it watches an AI system while it runs, so the evidence that it's behaving correctly is already there when someone asks for it, instead of something a team scrambles to assemble after the fact. The reason that matters comes down to who's asking. It's rarely the engineers who built the thing. It's a risk committee deciding whether a model can touch production. It's a CISO who has to sign an attestation. It's an internal auditor asking what the system did last quarter and why, or a regulator asking the same question with less patience. Those groups don't want a demo. They want a record.

Without that record, the pattern is predictable: the pilot works, everyone likes it, and then it sits in front of a review board for two quarters while somebody reconstructs the evidence by hand. The technology was never the blocker.

This is the kind of product we most like building, because we didn't come up with it in a planning session. Customers described the problem to us in a hallway, repeatedly, until we understood it.

Why Cisco Sellers Keep Bringing Us In

OnStak has been a Cisco engineering partner for over a decade. The accreditations are more of a side effect of that than something we set out to chase.

The part that actually matters to a seller is simpler. While many organizations engage when there's an opportunity to close, we engage when there's a challenge to solve. We'll invest in the hard problem before there's a transaction attached to it, because our engineers and Cisco's engineers have spent years working side by side and that's how the relationship has always run.

What We're Bringing to GSX FY27

Three things we'll be ready to talk through:

There is one more thing, launching officially later this year, in the cross-domain correlation space. It is the piece underneath a lot of what is described above, and it is the reason we can make the "establish the relationships first" argument with a straight face. We would rather show you than write about it, so come find us.

Mostly, though, we are looking forward to the hallway conversations. Those are still the best part.