Broadcasters have spent recent years asking what generative and agentic AI could do for their business. In 2026, the questions have become more intentional: can it cope with a live newsroom, a sports production or an archive whose rights and metadata were built up over decades? The mood is more practical now. Teams still want AI to help them deliver speed and creative range, but the tools must share context and fit established workflows. A clear point of human control matters too. A striking demo is easy; a dependable production system is much harder.
Adoption across wider enterprise markets is moving quickly. IDC reported in June that 50% of organisations were deploying AI agents in production across several business areas, with another 27% running them in at least one. Important advances are happening at the heart of the media industry too.
The hard part starts after the demo
A model can summarise a clip or tag an archive. A separate service can automatically generate highlights from a sports match or press conference. Each may work well on its own; challenges come when they need to work together across an entire newsroom, or across the many different organisations now involved in a single story.
Take a live newsroom as an example, where several agentic AI tools work across the same running story. Which source is authoritative? How are rights, corrections and editorial decisions carried from one system to the next? What should happen when two agents disagree? And who is accountable for the result? These are production questions, rather than abstract debates about standards.
They decide whether AI saves a producer time or gives that producer another system to police. The broadcasters we speak with want reliable connections between tools, clear governance, and a record of how decisions were made. Above all, they need to retain creative control, editorial judgment, and human-in-the-loop frameworks to intervene before output reaches an audience.
These demands form common strands through the IBC2026 Accelerator cohort, which comprises eight projects and a dedicated incubator. The projects bring media organisations and technology partners together for a six-month development sprint, ending in live proof-of-concept demonstrations at IBC2026.
Newsrooms need a shared understanding of the story
News production shows why a collection of smart AI assistants is not enough. A planning system, media asset manager, rundown and graphics platform may all hold part of a developing story. If each AI tool reads only its own fragment, staff still have to repeat context and correct misunderstandings.
The SMART STORIES incubator is tackling that problem across the news chain. Project champions, which include AP, the BBC, NBCUniversal, Channel 4, ITN, ITV, Sky, Washington Post, Reuters and Al Jazeera, are working on an open standard for story-context interoperability, with a Story Object Model designed to carry structured editorial knowledge between tools.
The aim is quite concrete: give agents a shared account of what a story is, what has changed and which editorial rules apply. The team plans to test a reference implementation in multi-vendor scenarios, including breaking news and planned coverage, while keeping human editorial control in the workflow. The long-term goal is simple to state, even if it is hard to build: when a story changes, that update should reach every tool automatically, instead of a producer re-typing the same correction into five different systems by hand.
The FRAMES Accelerator backed by champions including the BBC, EBU, ITV, RAI and MovieLabs addresses a related problem earlier in the production process. It is exploring a common semantic model and knowledge graph that can connect scripts, archives, newly captured material and specialist teams. Creators remain in the loop, while agents handle the groundwork: finding relevant footage by meaning rather than exact keywords, labelling clips automatically, and improving the quality of older material.
Personalisation moves into the live chain
For years, much of media personalisation happened at the edge of a service, through recommendations, thumbnails and home-page rows. The next step reaches into the live production and delivery chain.
For example, ITV, AWS, TCS and Camb.ai are developing an AI for Live Sports initiative exploring how several AI capabilities can be coordinated from production through to consumption. Its proposed “meCast” approach would adapt live content for individual viewers, drawing on real-time content understanding, highlight creation, localisation and audience engagement while working within editorial and rights constraints.
This raises a tougher engineering challenge than producing several fixed versions of a stream. The workflow has to understand an event as it unfolds, decide which changes are useful, make them at speed and preserve the integrity of the source. The Delta Protocol project backed by champions including DAZN, the EBU, RAI, the IET and SMPTE, with Google Cloud, Neoviews and Eluvio among its participants goes further upstream, testing whether live media can move from frame-by-frame delivery towards a semantic model in which meaningful changes are transmitted and content is composed for the viewer at the destination.
The best AI has a defined job and a clear owner
Some of the strongest deployed examples give AI a narrow role and good data. They also provide an obvious hand-off to a person and clear viewer benefit. Formula E and Google Cloud’s Strategy Agent is a great example: it turns live telemetry and timing data into near-real-time race insights. The resulting graphic goes to the broadcast director for approval before transmission. That last step matters. The agent deals with a volume and speed of data that would be hard for a person to process during a race, while the editorial decision stays in the control room. The value is easy to explain: viewers get a clearer account of the strategy unfolding on track.
The Premier League Companion powered by Copilot, is another strong example, it uses context-aware prompts and agentic analysis to give fans personalised access to live insights, stories and 30 seasons of league history. We also see accessibility as a huge area of innovation. To name just one example, G&L Systemhaus brought Signapse, Norsk/id3as and Akamai together to build a broadcast-grade AI-powered sign-language service for live and on-demand streaming, addressing a gap that human interpreters alone cannot cover at current content volumes.
These projects tackle very different needs, yet each starts with a defined user problem. All three are finalists in this year’s IBC2026 Innovation Awards.
Collaboration is always part of the answer
No broadcaster can solve the industry’s biggest AI questions in isolation. A new wave of generative and agentic AI tools are now crossing an ever broader array of newsroom, production, archive, cloud and distribution systems, often with different vendors at each stage.
It’s never been more important for media companies, technology providers, educational institutions and standards bodies to spark debate across industries, collaborate among competitors, and challenge their assumptions together. We hope to see many new solutions presented, questions asked, and perspectives changed at IBC2026.
