Agentic AI is an extension of generative AI; it uses a large language model (LLM) as a ‘brain.’ These autonomous agents can organize themselves to operate systems, manage applications and make decisions. While this type of artificial intelligence is still nascent, these AI agents would theoretically be able to communicate with each other dynamically and flexibly, and not only via predefined interactions.
This creates a more flexible, human-like thinking paradigm that can use context, past experiences and reasoning to respond to novel situations.
There are several different design patterns made possible by this type of artificial intelligence. This includes agent orchestration, which includes one lead agent that organizes and distributes tasks to a team of sub-agents. Autonomous tool use, which gives an LLM the ability to call up external tools like databases, is also crucial for agentic AI to work properly, by allowing it to find and access the necessary information and perform actions.
Agentic AI and other advanced AIs could help solve the ‘GenAI paradox’: While companies across almost all sectors have moved fast to adopt generative AI, almost none of them are seeing bottom-line impact.
Unlike relatively simple LLMs, which focus on producing content, integrating agentic AI would reshape organizations on a much deeper level. Agentic AI would force organizations to fundamentally rethink how information is organized, how tasks are executed, and how processes are designed—all in order to properly develop an intelligent agent that could manage the system and solve problems autonomously.
For communications providers, agentic AI could be utilized for network management and optimization. Not only could an intelligent agent predict network capacity and traffic, but it could actively direct traffic flow, ultimately ensuring better performance across the entire system. This has clear benefits for both the operator and the end user, including fewer network disruptions and increased reliability and speed.
Agentic Management of the Satcom Environment
Using agentic AI could help an operator “get total sellable capacity from a network,” said Brian Barritt, CTO of Aalyria. It helps answer the critical question for operators: “how do you quickly arrive at close to optimal solutions?”
Satellite networks are massive and complex, and mapping a route across a network that is under heavy traffic is a difficult task. It must take into consideration the massive variety of paths that can potentially be taken across inter-satellite links, to ground stations, across multiple orbital regimes, and between moving spacecraft.
“The set of possible configurations of these systems are massive." -Brian Barritt, Aalyria
“The set of possible configurations of these systems are massive,” Barritt said. “There can be millions or tens of millions of candidate beams that you could choose to bring into existence and assign to a particular channel and bandwidth.” By incorporating agentic AI into the operation of a system, traffic flow and ideal paths can be calculated with greater accuracy, taking a burden off the operator and improving the overall capability of the network.
Jake Saunders, a director at ABI Research, emphasized the role that agentic AI could play in the OSS, or operation support systems, of satellite networks.
“This is essentially a fabric which looks after the whole network,” he said. “When a base station or a switching center or a data center goes down, they can reroute the traffic.” This can include routing tasks with additional layers of complexity as well, such as priority routing and guaranteeing spectrum access to certain customers.
The biggest challenge for many satellite operators is providing the end user with the latency that they need. “The biggest challenge we face with satellite constellations, even in LEO, is latency,” Saunders said. “If you’re geostationary, or even in LEO, your latency is greater than for a 5G network on the ground, where it’s on the order of 10 milliseconds or less.”
While most users wouldn’t notice a difference in 10 milliseconds versus 15, certain systems might—and a priority monitoring system or intelligent data traffic system could reduce that latency to speeds competitive with terrestrial. For a use case like first response in the case of an emergency, that difference in system capability and speed can be crucial.
Despite these significant advantages, agentic AI is severely limited by time and computational power constraints.
“These things can find provably optimal solutions, but the time they take to run grows exponentially with the size of the complexity of the problem,” Barritt said. One solution is partitioning the network and asking the AI to solve many small problems instead of one large one. The operator can also train the AI on certain assumptions, such as that a satellite will only ever serve certain locations that are directly beneath it. This, combined with traditional algorithms, allow the AI to efficiently solve the problem.
“That’s one approach,” Barritt said. “But you give up some flexibility and freedom [by] partitioning the problem like that.” Another option is to use different types of artificial intelligence.
For example, Aalyria is working with deep learning and reinforcement learning AIs for optimization. These types of AI are not based on an LLM, and thus cannot be properly called agentic AI, but they can be trained to “very quickly arrive at solutions that are near provably optimal, and to do it very quickly,” Barritt said.
Managing the Satcom-Telecom Integration
Communication providers, whether terrestrial or space-based, are extremely conservative. Operators often prioritize keeping the system within “tight confines of autonomous capability,” Saunders said. But as telecom and satcom are brought closer to true integration, greater processing speed and computing capability will become crucial to bring together these disparate systems. This need could push them to embrace agentic AI sooner, especially as telecom still sees little return in investment from the billions of dollars that have been poured into 5G.
Agentic AI promises to increase the speed and efficiency of network connectivity across and between satellite and terrestrial networks.
The landscape of 5G and 6G NTN development is where this convergence is taking place. Just as humans once had to manually configure networks before the development and implementation of modern network technology drastically increased the speed of communication, agentic AI promises to increase the speed and efficiency of network connectivity across and between satellite and terrestrial networks. Terrestrial and non-terrestrial operators can share locations of antennas and base stations, allowing the AI to plan and route the entire network.
But this technological advancement doesn’t come without significant security concerns; telecom is still very hesitant about allowing unsupervised AI-enabled automation.
The Cost of Complexity and Scale
Satellite operators interested in agentic AI must consider many factors, such as complexity costs, the resources required for scaling, and the increased number of vulnerabilities created by autonomous system management.
With the introduction of autonomous agents into the system, the computational expenses of the system as a whole rise as well.
As a system becomes more complex, it incurs increasing complexity costs—more areas of risk, which in turn create more potential for chaos. With the introduction of autonomous agents into the system, the computational expenses of the system as a whole rise as well. This is also a reflection of the problem of scale in these global networks. AI of all types is a resource-intensive behavior. To properly scale, the system must be able to handle the increased load and still provide reliable, trustworthy performance to the end user.
“Both the telecom and satellite industries have a risk-averse culture,” said Barritt. “The customer wants to understand, in some detail, what’s happening under the hood of their system.”
There is also a strong concern for ensuring the customer has a seamless, uninterrupted service, no matter what is happening ‘under the hood’ of these complex networks. “When it comes to telecom, the greatest concern is making sure that one, their customers are never inconvenienced, and two, that there’s no major cascading disaster,” Saunders said.
A cascading failure on a network as large as a satcom or telecom network could quickly become devastating. Security for these systems is absolutely paramount—and one of the reasons that operators are so hesitant about incorporating unproven, or even proven, AI. Autonomous agents evolve and adapt over time. While this provides needed flexibility, it also means that developments can be unexpected and expose vulnerabilities at any time.
Other security risks are possible as well. Questions of accountability quickly arise—who is responsible for decisions made by an intelligent, autonomous agent? And agentic AI, which is based on a language model and is easily commanded via text, could theoretically begin communicating in formats that humans can’t read or understand.
But these possibilities are extreme, and for now, agentic AI remains a nascent technology that has the potential to optimize satcom technologies to their greatest potential.
Explore More:
How to Train Your Teleport: AI Fuels Intelligent Ground Systems
Podcast: Advanced Digital Ecosystems, Autonomy and Space Superiority
Is AI the Key to Greater Situational Awareness in GEO?
Analysys Mason: Artificial Intelligence is More Than Just Generative AI
