When most people talk about artificial intelligence (AI), they usually mean Generative AI (GenAI). While this has brought a great deal of flexibility and freedom to video, audio and text creation, it is also responsible for a substantial amount of ill-conceived or poor-quality material. This has not only sullied the reputation of AI but also overshadowed the potential of a more practical form of the technology, agentic AI.
More of a background technology than GenAI—which has also been around a lot longer, the modern iteration being formalized between 2014 and 2017—agentic AI is less generally known, even though it began to appear around three years ago. The crucial difference between the two is that while GenAI reacts to prompts or triggers to create content, agentic AI is intended to autonomously establish and control a function or situation.
AUTONOMOUS WORKFLOWS
In the context of broadcasting, agentic AI is regarded as key to the development of applications—known as “agents”—to run specific technical tasks, such as monitoring, dealing with system faults or failures, and manage a variety of functions across production and distribution chains.
"Unlike GenAI, which creates content on demand, agentic AI orchestrates entire content workflows autonomously," explains Stephanie Lone, global leader of solutions architecture, Media and Entertainment, Games and Sport at AWS. "This means systems can coordinate complex multi-step processes without constant human intervention at each decision point."
The basis for this, Lone adds, is multiple "specialized agents" working together proactively to natural language requests. This combination—and simultaneous operation of applications— forms the basis of systems from other developers, including Witbe's Agentic SDK, which was launched in February this year. Intended as a "test automation framework" to run and scale monitoring, testing and validation over video workflows, Witbe chief executive Mathieu Planche describes it "a suite of different agents" designed to automate test processes.
"Test automation used to be a script that people had to develop," he says. "We have moved away from this manual process into something that is, while not human-less because we still have [a] human to control what the Agentic SDK is doing, but it's a process that is largely simplified."
The first of the agents in the suite is Test Designer, which was developed to understand the testing strategy of the video service being analyzed and then write test cases based on that. "Agentic AI technologies can observe inputs, plan… actions and do most of the heavy lifting in different workflows that are otherwise taken care of by people," Planche adds.
As with previous methods of automation, the developers of agentic AI systems are quick to emphasize that the aim is not to replace human operators. Fares Birke, principal practice lead for Applied AI at Qvest, comments that the goal is to enable people to "work better, faster and with fewer interruptions" by automating tasks that are either repetitive or require a lot of coordination.
"Agentic AI is about achieving outcomes, while GenAI is primarily about creating outputs – text, audio, video—in response to prompts," Birke says. "This makes agentic AI far more relevant for complex media environments where speed, reliability, compliance and coordination matter as much as creativity."
‘ENRICHED METADATA’
Agentic AI-based systems are now proliferating in the broadcast sector for a variety of applications. These include: searches of archives, transcripts and live feeds; managing ingest, logging, review and delivery; triggering MAM, newsroom, post-production and distribution systems; and ensuring operational and editorial rules. This can be seen in the number of recent system launches, including ThinkAnalytics' ThinkMetadataAI. Launched at IBC 2025, with a U.S. debut at last month’s NAB Show, it uses agentic AI to automatically create "enriched metadata" for content catalogs.
ThinkAnalytics co-founder and chief technology officer Peter Docherty comments that over the past 20 years the company had focused on deep metadata enrichment for personalization and viewer engagement but in the last three to four years has moved into targeted advertising.
"A lot of agentic processes enable automation and operational efficiencies that weren't there before," he says. "That's obviously quite different from GenAI because agentic AI helps streamline operations. What we've done with our Metadata AI is combine GenAI with Agentic and other capabilities to automate processes that were previously manual. This enables huge automation at scale but with intelligence behind the decision-making."
At the NAB Show last month TVU Networks showcased “TV Cortex,” a service that uses Agentic AI to aggregate and manage the audio and video content used to create news stories. TV Cortex has the power to act autonomously and purposefully to achieve an outcome independently of our interaction with it. In theory, operating this way may involve less “human shading” of events and more factual coverage.
According to TVU Networks President Paul Shen, TV Cortex is intended to augment human productivity, not to replace it. If used effectively, the system frees up producers and editors to focus more on the creative aspects of news production, like crafting compelling stories and providing more human context.
“TV Cortex leverages agentic AI, deploying a hierarchy of master and subordinate agents that dynamically scale to handle the volume and complexity of incoming stories,” Shen said. “These agents automate the retrieval, categorization and scoring of stories based on factors such as recency, relevance, market size and customizable editorial priorities.”
KEEPING QUALITY—AND THE TRUTH IN CHECK
A crucial area of broadcast and streaming distribution that has undergone a high level of automation and streamlining in recent years is compliance and Quality of Service (QoS) monitoring. AI—and agentic in particular—is now offering the potential to take this further through new systems such as Bitmovin's Observability, which was introduced last October.
"We're leveraging AI in different areas of our product range, including the Observability product for looking at sessions, playback, QoS metrics, root cause analysis and error debugging," explains Jacob Arends, product manager for playback at Bitmovin. "AI is really good at summarizing data, for example, so we are using that in our Observability products to make it easier for customers to understand complex datasets [and] what actually went on in a session."
In the last year or so, doubts have arisen surrounding GenAI chatbots, especially with systems “hallucinating”—basically making things up— or just agreeing with a question. Because compliance monitoring is a crucial failsafe in the broadcast distribution chain, there can be no such doubts; any automation technology must be 100% trustworthy.
"Our Agentic SDK is not creating bugs within our customers' platforms," states Witbe's Mathieu Planche. "They detect errors and flag them for people to review and analyze. Then people decide if they are escalating them internally or if they just discard it."
As for the future, Planche acknowledges that agentic AI technology for broadcast is still "new for everyone" and that media control rooms, staffed by operators, are "not going away right now."
At AWS, Lone observes that the potential for Agentic AI in media production "is massive" but says the broadcast sector does have to be careful with it. "The industry still needs to establish strong guardrails around data governance, transparency, content provenance and responsible automation practices," she comments. "Once these are firmly in place, companies can have more confidence in AI-driven decisions and feel comfortable deploying the technology on a broad scale."
Standards and oversight are absolutely necessary for such an autonomous technology as agentic AI could be. But, right now, the broadcast sector appears to be weighing up whether there is already enough automation in key processes or if these new agents are the key to more efficient and fault-free operations.
