Agentic AI sparks need for more orchestration, oversight
Orchestration all the way down: Experts say multiple levels of autonomous coordination and oversight will guide the near- and mid-term evolution of AI and computing.
ORLANDO, Fla. -- The near-term evolution of AI and more distant prospects such as quantum computing are expected to drive an IT future in which CIOs will need to coordinate deeper layers of technology resources.
Indeed, orchestration -- in various guises -- emerged as one of the key themes at Gartner IT Symposium/Xpo 2024 this week. Several speakers at the annual event discussed how emerging technologies must work together to yield the best results.
A key example is agentic AI, a form of artificial intelligence in which systems can make autonomous decisions and take actions on their own. The technology, which remains in the early stages of development, could represent the next wave of AI.
The rise of the meta-agent
Sri Elaprolu, director of the AWS Generative AI Innovation Center, said agentic AI involves taking a user's command, understanding its intent and breaking it into smaller tasks that individual AI agents can execute in concert. For example, multiple agents would need to communicate with each other to perform the tasks involved in arranging an international business trip and aggregating them into travel options for the user, according to Elaprolu, who discussed AI deployment trends in a session at the conference.
"Where we see this heading is meta-agents or multi-agents," he said. "This is where you are not only devising agents but orchestrating between agents. Agent orchestration is starting to take up quite a bit of interest from customers, and you can see this growing rapidly in the coming months."
Jensen Huang, president and CEO at Nvidia, also pointed to the rise of AI agents and the need to coordinate them during his keynote session at the symposium.
"We are going to have a very large population of digital workers," Huang said, citing sales, supply chain management and AI engineering agents as examples. Digital workers, as is the case with human employees, will need to be able to find each other and work together to accomplish their intended missions, he added.
Agentic AI, which can run with little or no human intervention, could get future oversight in the form of guardian agents.
Digital guardians to patrol AI agents
However, the prospect of businesses launching millions of AI agents rules out human administration, noted Daryl Plummer, an analyst at Gartner. Enter an agent oversight tier Plummer referred to as "guardian agents."
"It's become very clear that guardian agents are going to be necessary, because humans-in-the-loop will collapse on itself as an equation," he said, referring to an approach in which people train AI tools and then monitor their work.
By 2028, 40% of CIOs will call for the use of guardian agents to autonomously track and oversee AI agents, Gartner predicted. Cybersecurity risk in particular will motivate them: The explosive growth in agents will increase the attack surface that internal and external threat actors can exploit in organizations, Plummer said. The consulting and market research firm forecasted that 25% of enterprise data breaches will be attributed to AI agent abuse by 2028. Such abuse might involve subverting an agent to funnel money into a fraudulent account, Plummer noted.
Orchestrating future tech advances
The more distant shape of computing will also feature a need for orchestration, according to Frank Buytendijk, another Gartner analyst who spoke on the future of computing at the symposium. The major development over the next five to 10 years will not be, as many people would argue, quantum computing but a mix of that and other technologies, he said.
It's become very clear that guardian agents are going to be necessary, because humans-in-the loop will collapse on itself as an equation.
Daryl PlummerAnalyst, Gartner
Buytendijk called this coming era "the paradigm of everything." It will likely include advances in both conventional and quantum computing as well as potential breakthroughs in fields such as neuromorphic computing and photonic computing. The former aims to model computing components on the human brain, while photonic, or optical, computing uses light to perform digital computations.
"We are dealing with a bunch of computing mechanisms," Buytendijk said, looking ahead. "It's not really about quantum. We think it's about orchestration."
Orchestration will sort out which disparate style of computing should do a particular job, based on characteristics such as cost, speed, accuracy and energy efficiency, he said.
"The tech providers that own the orchestration own the future of computing," Buytendijk said.
John Moore is a writer for TechTarget Editorial covering the CIO role, economic trends and the IT services industry.
