But what underpins this technology? Why are we now witnessing the rapid growth of agentic AI?

Several key concepts are currently being utilized by businesses and other stakeholders.

One is the notion that AI can be specialized to perform various tasks or operations. This is evident with Claude, and agents that can operate a computer just like humans do.

There’s also the concept of system distillation and ensemble learning, where models interact with each other.

On the data side, companies must decide how to deploy systems and where to store the data, as well as how to aggregate it for use.

At Imagination in Action in April, my colleague Daniela Rus, director of the MIT CSAIL lab, interviewed several professionals about their thoughts.

Cindy Howson from Thoughtspot, Kevin Shatzkamer of Google Cloud, formerly of Microsoft, and Anshul Ramachandran from Windsurf participated.

Regarding significant potential, Howson mentioned that the groundwork was already laid, and introduced the concept of the "Internet of AI" as a new paradigm.

Shatzkamer spoke about productive AI and its capabilities, while noting that although much of the technology is available, it’s “not yet fully integrated.”

Ramachandran discussed how generative models are becoming adept at specialization and the proliferation of agentic systems.

“Even as we encounter certain physical limitations in the real world,” he said, “it will open up new frontiers in models, power, and technology in general, enabling a new kind of application and ways of thinking.”

Barriers to Evolution

In terms of current business limitations, Howson mentioned the challenge of obtaining clean, consistent data, and transitioning from structured data to semi-structured data, such as data stored in PDFs.

“I think many companies have clean, consistent structured data,” she said. “When we talk about semi-structured data, think about the PDFs on your network drives – which employee manual is the correct version? It’s anyone's guess. … I think some of the data management practices we’ve applied to structured data, we haven’t applied to semi-structured, but I believe the technology is ready; it’s more the people, mindsets, and processes that are less prepared.”

She also pointed out that 81% of people struggle with basic literacy.

Big Capabilities

The panel also discussed how systems are becoming smarter.

Ramachandran talked about multiset retrieval, and how systems can conduct searches in the manner humans do, with one search leading to another, to improve accuracy and generate rich results.

Shatzkamer discussed long-term memory and context windows, as well as research reasoning capabilities.

He also mentioned the future significance of quantum computing and supervised fine-tuning.

“Look at where quantum computing stands on the near horizon,” he said. “I think that's going to be a game changer in terms of AI processing capabilities, right? I think right now we're in a world of more, bigger, faster, and we keep on trying to build as much infrastructure as possible to meet the demand. And I think we'll see that trend continue for the foreseeable future.”

As for supervised fine-tuning, he stated:

“As much as we've talked about supervised learning … in this new supervised fine-tuning world, (you) can build smaller models with human involvement in a much more meaningful way.”

Ramachandran suggested that generative AI is reaching critical mass, with interesting data that doesn’t always require massive LLMs. He provided examples of user behavior statistics that can unlock numerous actionable insights for nearly any type of business, highlighting that you don’t need a large data center or many Nvidia GPUs to conduct such research.

Shatzkamer commented that the open-source community played a crucial role in maturing all of this. Howson spoke about the decentralization of the cloud and the resulting "hybrid world."

Question on Emerging Tech

When Rus asked about the single most intriguing emerging technology, the panelists responded as follows.

Howson brought up agentic analytics.

Shatzkamer discussed operational metrics for efficiency.

Ramachandran said he’s most fascinated by robotics and physical AI.

All of this has significant implications for our new AI landscape. Stay tuned as we observe much of this continuing to evolve throughout the year.

The above is the detailed content of Building Blocks Of Agentic Systems: What Does It Look Like?. For more information, please follow other related articles on the PHP Chinese website!