Enterprise technology is witnessing an interesting confluence, with developer empowerment on one side and artificial intelligence hype on the other. While gen AI promises to democratize the toolset needed for creativity and ingenuity, it falls to platform leaders such as Google Cloud to flesh out the pathways to realize this expansive tech future with AI agents.

Google’s Jason Davenport discusses the hype vs. reality of AI agents with theCUBE.

In a world where technical expertise can emerge from both structured paths and chance encounters, Google LLC is helping stoke that passion, abstracting the technicalities of app development to “help people make cool stuff,” according to Jason Davenport (pictured, right), technical lead for DevRel at Google.

“You think about where we were three years into this whole big push for LLMs and generative AI; you could argue it started back in 2018-ish with some of the things that were published with GPT and transformers,” he said. “But you think now, with agents, we can get code to do things for us on our behalf and execute those goals. We’re starting to be kind of in that territory.”

Davenport and Aja Hammerly (left), director of DevX AI at Google Cloud, spoke with theCUBE’s Savannah Peterson for the “Google Cloud: Passport to Containers” interview series, during an exclusive broadcast on theCUBE. They discussed AI agents as the next iteration of a decades-long effort to make computers work for developers. (* Disclosure below.)

Understanding AI agents

While the specific buzzword “AI agents” is new, the core principle has existed for decades. From early chatbots to microservices, the goal has always been to delegate work to software — to put computers to work executing complex tasks. Today’s AI agents might have more powerful tools, such as large language models and application programming interfaces, but at their core, they’re “code with a job,” according to Hammerly.

“I spend a lot of time explaining that the hype and the reality, there’s a difference, but a lot of it is just around terminology and that the skills we already have are useful still,” she said. “We have a lot of the basic skills we need to work with these tools. There’s not that much more to learn.”

Many of the skills developers already have — system orchestration, deployment practices and foundational coding — remain highly relevant. Much of the intimidation around AI comes from terminology rather than entirely new concepts. Words such as “transformers” or “retrieval-augmented generation” can sound intimidating, but often they describe patterns developers have seen before, just with updated names.

While generative AI capabilities have advanced significantly, the fundamental goals of business remain the same: Attract customers, grow relationships and deliver value. AI agents aren’t rewriting those goals — they’re new tools for achieving them, according to Davenport.

“One of the things that I found in my workflow, even thinking about expectations, is just using a large language model for the initial question and plan,” he said. “I find that where we’re at now with even thinking models, that has helped me reset my expectations about what I could also get out of an LLM.”

Skills, mindsets and the road ahead

The core mindset with which to approach AI adoption is as a tool, not a magic wand. Success in AI is heavily dependent on the antecedent instruction set. Working with AI is effectively a skill — such as coding or architecture — that improves with practice.

Iteration, failure and curiosity are essential parts of the process. Asking an AI tool why it produced a certain output can be just as valuable as asking it to try again. Over time, you build an understanding of how to “set the tools up to win,” according to Hammerly.

“It’s kind of cool that the tools will help you work with the tools,” she said. “If you care about how it happens, you’d better be very specific. And if you don’t care about how it happens, just that it happens at an accuracy level, or correctness, style and everything that you like, don’t be overly specific.”

Far from rendering data strategies obsolete, AI will make first-party data even more valuable. Businesses that deeply understand their customers will deliver more personalized, high-quality experiences — an advantage that comes from data they alone possess. Harnessing this data fully, however, requires hands-on refinement. Tools such as Google’s AI Studio or Gemini CLI let developers experiment, make mistakes and learn how to interact with AI effectively, Davenport added.

“One of the benefits is just starting to ask it for tasks,” he said. “It’s a journey to programming yourself and programming code or all these things, but the way that you have to do it is get in and start to figure out how these things work. Sometimes it doesn’t work, and failure is an acceptable modality in these things.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of the “Google Cloud: Passport to Containers” interview series:

(* Disclosure: TheCUBE is a paid media partner for the “Google Cloud: Passport to Containers” series. Neither Google Cloud, the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE