Caylent CTO On AWS AI Besting Microsoft And Google; VMware Migration ‘Boon’ To AWS

Randall Hunt, CTO of award-winning AWS partner Caylent, explains how AWS is besting Microsoft and Google in AI, migrating VMware customers to the cloud, and Caylent’s unique AI portfolio and successful use cases.

Red-hot Amazon Web Services partner Caylent is winning AI deals and launching new innovation at a blistering pace with Amazon Bedrock and AgentCore being a key market differentiation versus Microsoft and Google Cloud.

“[Microsoft is] playing catch-up now with model choice and model selection, and they also never develop the muscle of fine-tuning and optimization of these AI models,” said Randall Hunt, chief technology officer for the award-winning AWS partner Caylent.

“Because we’re able to build on top of Bedrock AgentCore, we are able to take advantage of any model that we want and leverage the best primitives at the same time,” Hunt said.

The Los Angeles-based AWS Premier Tier Consulting Partner built its Caylent Accelerate portfolio that leverages AI-powered delivery to transform the way customers modernize databases and migrate to AWS.

[Related: AWS CEO On $38B OpenAI Deal, ChatGPT, Nvidia GPUs And ‘Powerful Reminder’ Of AWS’ ‘Trust’]

The Caylent Accelerate for Cloud Migration offering specifically aims to cut VMware costs and customer migrations to AWS to half the time with AI-assisted assessments that uncover the most efficient migration paths.

“A huge portion of VMware customers are actively looking at alternatives. I think it’s 73 percent of VMware customers are pursuing alternatives at this point in time,” Hunt said. “I know it’s been a boon for our business, and I hope it continues. We are finding ways of accelerating the move from VMware into AWS, or the move from on-prem into AWS.”

AWS awarded Caylent its Global Partner of the Year Award in both Migration Consulting, as well as GenAI Industry Solution in 2024.

Bedrock AgentCore ‘Nailed It’ And Trek10 Acquisition

Hunt worked at AWS for years as a top solutions architect and software engineer. He was also a software engineer at the likes of SpaceX and MongoDB.

Hunt said Bedrock AgentCore—which is AWS’ agentic platform for building, deploying, and operating AI agents securely at scale—has been a game changer in 2025.

“Bedrock Agentcore is like taking a step back and thinking, ‘Hey, knowing what we know now, after years of working with generative AI, how would we make the perfect service for this?’ And they nailed it. They really did.”

In a move to further its AI and AWS reach, Caylent last month acquired fellow AWS Premier Tier Services company Trek10 to expand its portfolio into managed services, potentially reach new customers overseas and strengthen its ability to deliver end-to-end AWS services.

In an interview with CRN, Hunt takes a deep dive on why AWS’ AI portfolio and strategy is besting Microsoft and Google Cloud, successfully migrating VMware by Broadcom customers to AWS, and AI deals now in production.

Do you think AWS is leading in the AI era versus Microsoft and Google?

Yes. AWS has the best primitives—storage, compute, networking—in the entire cloud industry.

If you look at things like EC2, VPC [virtual private cloud], Amazon S3, Lambda, ECS [elastic container service], EKS [elastic kubernetes service]—hands down, best primitives and best things to build AI on.

Also, because we’re able to build on top of Bedrock AgentCore, we are able to take advantage of any model that we want and leverage the best primitives at the same time.

So S3 has vector storage now, which is just a tremendous capability for any form of RAG.

Since November of 2022, everyone has been trying and chasing and figuring out what exactly they needed to build. Companies took different shots. They took different approaches. There wasn’t always alignment on what things should look like from a developer SDK perspective, from a design perspective—people weren’t sure exactly what to build.

With AgentCore now, I feel like I can use whatever I want.

I can even use the OpenAI models in AgentCore if I want. I’m not limited to just Bedrock models.

I can go out and use anything, but I still get the best execution primitives, the best query primitives—all of those things.

Why is AWS AgentCore such a market differentiation versus Microsoft and Google?

Agentcore and Bedrock are drivers for customers who are already looking and building agents and AI services.

From a competitive standpoint, Microsoft went all in with OpenAI.

So they were focused in the beginning only on OpenAI models and building with those.

So they’re playing catch up now with model choice and model selection, and they also never develop the muscle of fine tuning and optimization of these models.

They’re developing it now. But the advantage that AWS has with like Sagemaker, we’ve been fine tuning foundation models since 2022 with Sagemaker Jumpstart. You can take all the open-source models and you can even then run those in Bedrock.

So that’s what’s got me really jazzed about building agents and building things on AWS.

I get to continue to use the primitives that are phenomenal, and now I have a framework and a system that removes a lot of the undifferentiated heavy lifting of things like: session management, chat management, code interpreter, browser use—like all of these tools that I don’t really want to have to manage are just built into AgentCore.

How is Caylent Accelerate for Cloud Migration helping you migrate VMware by Broadcom customers over to AWS?

We were motivated to create this offering primarily because of Broadcom’s acquisition of VMware.

A huge portion of VMware customers are actively looking at alternatives. I think it’s 73 percent of VMware customers are pursuing alternatives at this point in time.

I know it’s been a boon for our business, and I hope it continues.

We are finding ways of accelerating the move from VMware into AWS, or the move from on-prem into AWS.

So that involves taking the output of these tools like [Microsoft] Hyper-V and others, and pushing them through a series of models—so not just using one model blindly, but dynamically constructing the prompts in the context and getting the best price performance we can on some of these conversions.

Then spinning up in AWS intelligently and in a tiered fashion.

What are you hearing from VMware customers?

The driver of the conversation is around the commercial terms and changes that Broadcom is making.

I mean, [Broadcom] is intentionally shedding VMware customers. AWS is a fantastic place for those customers to land.

It is a nice cherry on top to say, ‘Also now you get to take advantage of all of these AI features that you previously had to call from on-prem to go across the network and all these other things. Now you have less latency, AI and all this other good stuff.’

Explain to me a successful Caylent AI use case with a customer?

We worked with customers like Unbounce. We helped them build and select different models based on the various use cases that they had.

That’s one of the advantages of Bedrock: we can take the best model available for a given use case, and we can programmatically determine the best solution.

We often use a framework like DSPy, which is an open-source framework from Stanford, to test all of these models really rapidly. We can construct the perfect prompt for each model because they don’t all take the exact same formats.

We start out what with what I would describe as a “vibe check.” So we test the thing we’re trying to ask the model to do: is it even possible? And then we build robustness with evals. So what that means is we introduce different contexts, and we build resilience and robustness around the implementation by making the prompt able to respond to varied contexts.

After that, we do a little bit of economic optimization.

So if this is a real-time use case, then we need to use real time inference—we need to do streaming, we need to do other things to optimize it. If it’s a multiturn conversation, we can use Bedrock Prompt Caching and we can get really nice results there, because subsequent invocations of the model don’t require us to pay the same dollar-per-token-cost.

If it’s not a real time use case, we can use Bedrock Batch, which then is 50 percent off across the board.

So overall, first we check if it works and if it’s possible. Next we build out the evals, the resilience, the reliability, the capability, and introduce differing context. And then we optimize the economics. That’s our process across all of these different customers.

---

Talk about Caylent’s Bedrock Battleground offering that lets customers quickly compare, test and select the most suitable LLMs or prompts?

We built an open-source tool called Bedrock Battleground. It’s on GitHub.

The reason we built that is, every four months since November of 2022, the price performance of the AI models has improved at least 2X. So it’s faster than Moore’s Law.

We’ve advised our customers to build a platform to leverage generative AI and not to bet on one particular model. Because they keep leapfrogging each other and you always want to be able to take advantage of the latest capabilities. So we’ve built a lot of stuff on Bedrock.

We mostly use Bedrock AgentCore at this point.

So that’s one of the things that’s changed is, our runtimes used to be in ECS and Lambda. And now, a lot of our runtime is in Bedrock AgentCore.

We think that service has serious legs.

When Bedrock initially came out, it was an interface for prompting models, and then they added all these features on. But I don’t think the world knew exactly how to go and build these systems yet. We were all discovering it at the same time.

Bedrock AgentCore is like taking a step back and thinking, ‘Hey, knowing what we know now, after years of working with generative AI, how would we make the perfect service for this?’ And they nailed it. They really did.