The frontier models shift weekly, and the tools built on top of them shift faster. For the engineers responsible for deploying AI in production, standing still means falling behind. That accelerating cycle is reshaping how enterprises approach AI infrastructure, explained Daemon Behr, field CTO of the Americas at Nutanix, in a video interview with The Forecast, recorded in April 2026 at the .NEXT event in Chicago.
"The combination of traditional AI and generative AI is what we're calling modern AI. It's not a bubble that's popped per se. It's an evolution to expand to generative AI."
Teams no longer pick a single model and commit, Behr explained. They must orchestrate the use of frontier and open models, GPU resources and autonomous agents, all while keeping changes focused on governance and deterministic outputs. He described the discipline emerging to manage this complexity as agentic engineering, which he sees becoming a core competency for organizations adopting agentic AI.
"The orchestration of all those different components, the vector databases, the data sources, the various agents that you create, the frameworks for that, the infrastructure that it's actually running on as well, that's all part of agentic engineering."
The short answer is orchestration. Behr describes a discipline that spans everything from assigning GPU resources to specific models to load balancing endpoints across cloud and bare metal.
"With agentic engineering, you're able to say these GPU resources are assigned over to this model," Behr said. "You can increase the number of instances, you can load balance your endpoints. You can use some endpoints that are running in the cloud or on-prem, on bare metal."
The orchestration extends beyond compute. It includes vector databases, data sources, the agents themselves, the frameworks that govern them and the infrastructure they run on. Behr calls all of that agentic engineering.
Frontier models change capabilities on a weekly basis. For organizations that need deterministic outputs, that volatility is a problem. Behr said enterprises want to lock certain models down rather than ride every update.
The solution is an AI gateway that connects to frontier models when needed while also supporting open models running on premises, locked to specific versions. That gives teams control over tokens, spend, performance, availability and governance over their Model Context Protocol servers.
"Having that greater control over where you're running your inference is a game changer," Behr said, "and it's something that we're able to do really well."
Behr practices what he builds. He runs his own team of AI agents to keep pace with an industry that reinvents itself weekly.
He uses a fork of an open source research agent, which he described as more secure, to research and curate information daily. The agents send him messages over Telegram and generate reports he studies alongside videos from roughly 50 creators.
He also deploys a swarm of agents to evaluate new software. When a project gains traction on GitHub, Behr tasks an agent team with downloading it, deploying it in a sandbox and reporting back on how it works, what its deficiencies are and whether it is worth adopting.
"I was doing that on my own, and then I found out that I didn't have enough time to do that," Behr said. "So that's where I thought, well, I could actually lean on the technology that I'm studying and learning to do that for me."
Behr frames the current moment as an evolution rather than a bubble. He calls the combination of traditional AI and generative AI "modern AI," an expansion rather than a burst.
“Things are happening in a much shorter time span,” he said. “You might be at the peak point of expectations, and that could pop the next day. But then a whole new thing pops up, so the cycle of innovation is a lot more frequent. The bubbles are popping on a very regular basis, but the bubbles are smaller.”
The technical limitations of the past are slipping away, he said.
“The constraints that have to do with creativity: What do you think is possible? What can you strive for, and how can you learn things along the way?” Behr pauses, then posits that this has created “an innovator’s paradise.”
Related:
Video transcript:
Daemon Behr: The fact that there's so many different large language models that are out there have different capabilities, whether it's multimodal or reasoning or so on, and the frontier models change on a weekly basis in terms of capability. But that impacts organizations as well because maybe they don't necessarily want to have those changes occur on a regular basis. They want to be able to lock certain models down to get deterministic outputs from their environment. So what we're able to do is we have our AI gateway that allows you to connect to those frontier models when needed, but you also have the capability to have open models that are running on-prem that you can lock to specific versions. So you have the capability of managing your tokens, managing your spend, have control over the environment, the performance, the availability, and also governance over your MCP servers as well. So having that greater control over where you're running your inference is a game changer and it's something that we're able to do really well.
Ken Kaplan: Teams have to build those capabilities. What do you call that activity or that area?
Daemon Behr: That's agentic engineering.
Ken Kaplan: So just describe that world a little bit. What is an agentic engineer doing?
Daemon Behr: So with agentic engineering, you're able to say these GPU resources are assigned over to this model. I want to scale with X number of users and then you're able to do that. You can increase the number of instances, you can load balance your endpoints, you can use some endpoints that are running in the cloud or on-prem, on bare metal. There's many different options on how you actually do the inferencing and the orchestration of all those different components. The vector databases, the data sources, the various agents that you create, the frameworks for that, the infrastructure that it's actually running on as well. That's all part of agentic engineering.
Ken Kaplan: How do you stay plugged in and moving ahead with so much rapid development around AI and how people are using it and what's next?
Daemon Behr: So I actually have a bunch of agents myself that I use. I use Nemo Cloth, it's a fork of open cloth that's a little bit more secure, in order to do research for me and curate my information on a daily basis. And then it actually sends me messages over Telegram and creates reports which I then study that as well as various YouTube videos from probably about 50 different people on a regular basis. And from that, I'm able to understand the changes that are happening in the industry. I also have a swarm of agents that I use to deploy new and innovative software that I come across on GitHub. So if something is rising up at the GitHub stars, I'll tell one of my agent teams to go download that, to deploy that in sandbox, tell me how it works, tell me the deficiencies, and then tell me if it's valuable to me. If it is, then I will actually adopt it. If not, then it'll tell me that I can just kill it.
Ken Kaplan: How did you get to that mindset and to actually do that?
Daemon Behr: I was doing that on my own and then I found out that I didn't have enough time to do that. So that's where I thought, well I could actually lean on the technology that I'm studying and learning to do that for me.
So the combination of traditional AI and generative AI is what we're calling modern AI. So it's not a bubble that's popped per se, it's an evolution to expand to generative AI. Now in terms of things that you could call bubbles and that are popping, it's not the same way that we'd have the like the dot-com bubble or busts like that where you have I guess a souring of expectations or the the trowel of disillusionment per se. That's happening on a much more shorter time span. So what may be at its peak point in terms of expectations, that could pop the next day and then you have a whole new thing which is propped up. So the rotation of innovation is a lot more frequent. So the bubbles are popping on a very regular basis but the bubbles are smaller.
Jason Lopez is executive producer of Tech Barometer, the podcast outlet for The Forecast. He’s the founder of Connected Social Media. Previously, he was executive producer at PodTech and a reporter at NPR.
Ken Kaplan is Editor in Chief for The Forecast by Nutanix. Find him on X @kenekaplan and LinkedIn.
© 2026 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.