What the show floor covered, and what was on everyone’s mind
One question ran through IBC 2026: Who is accountable when an AI system does something nobody approved?
For three years this industry asked about the model. Translation quality. Speed against real time. By September those were settled enough that the floor had moved on to harder ones: who approves agent access, where the data actually sits, and who keeps watching an agent after it goes live.
At IBC 2026 the industry caught up to where our client work has been for the past year. Accountability is now the constraint on scaling AI. The demos drawing crowds covered metadata tagging, live translation and content discovery, work production teams did by hand eighteen months ago. The tools work now. Nobody has settled who answers for them.
There Is a Lack of Trust in the Agents Being Deployed
Agents were the clearest case. Almost every rogue-agent story traces back to one gap. Very few teams had written down what an agent was allowed to do before they gave it access.
The worry is grounded. Earlier this year:
- Agents at OpenAI took unsanctioned action during an internal security test
- Anthropic found its own agents had reached the live systems of three companies, after a test environment that was supposed to be isolated wasn’t
- A separate government security test saw its own agents do the same, unprompted
Deloitte found nearly 74% of companies expect to be running agents at moderate scale by 2027. Only 21% have mature governance for agentic risk.
That gap is harder to see than a technology gap. It is also far less exciting to fund.
Model Context Protocol has emerged as the answer. MCP is a standard for defining what an agent can see and touch before it gets the chance to overstep. It forces a team to decide in writing what an agent is allowed to reach, rather than assuming good behavior and hoping for the best. That constraint is the reason to adopt it.
Teams that skip that and wait for a better model will get a faster version of the same unscoped access.
We ran into this early. WizeMedia Suite, which we demoed at NAB 2025, had multi-modal agents running against live production workflows. That meant the scoping question had to be answered in code rather than in a policy document. That work sits under our High-Impact Agentic Media Accelerators now. Role-based access control and MCP support are part of the build, so what an agent can reach gets decided at design time.
Key takeaway: Scope gets decided before deployment, or it gets decided after an incident.
Control Over Content Became Critical
The accountability question showed up again around infrastructure. Cloud let broadcasters run production from almost anywhere. It also scattered the answer to a question every operations lead needs ready: where does a given piece of content live right now, and who can reach it?
For a few years it got deferred in favor of speed. At IBC 2026 it was being asked out loud.
FIFA’s decentralized 2026 World Cup production and LaLiga’s sovereignty push are the same move. Keep the reach, take back control of where the content sits. There are numbers behind that choice. 23% of enterprises that moved to public cloud have already moved workloads back, mostly over cost and control rather than performance.
In our M&E work, the pattern is consistent. The teams that handle this well can answer the location question before something breaks. Everyone else improvises afterward. That is what our Cloud FinOps for Media practice exists for, and why WizeMedia Suite keeps asset location visible across video, audio and metadata workflows instead of treating it as an infrastructure detail.
Key takeaway: Flexibility on infrastructure holds up when a team can say where its content lives and who can reach it before something breaks.
Interoperability Finally Shipped
Scope on the agent and location on the content only hold if the systems enforcing them can talk to each other.
The European Broadcasting Union’s Media eXchange Layer is why competing vendors showed working interoperability on the floor this year instead of a roadmap slide. Grass Valley and Lawo were among them. So were Matrox and EVS. MXL took an Innovation Award.
It got there by being built implement-first. Engineers from different companies wrote code against each other from day one instead of agreeing on a specification and hoping separate implementations matched later.
The EBU’s other push was the Dynamic Media Facility, and this was the year it stopped being an architecture diagram.
- Qvest ran the largest multi-vendor DMF showcase on the floor, with orchestration, media processing, audio and monitoring from different vendors in one live environment
- Grass Valley demonstrated DMF principles across a camera-to-consumer workflow, bridging existing SDI and ST 2110 kit into software-defined operations
- Lawo showed processing allocated across studios, central facilities and remote production on demand
All three describe the same thing. The facility stops being a fixed set of boxes and becomes a pool of capacity that gets assigned to a production and released afterwards. That raises the question running through the rest of the show. When capacity moves on demand, who approves what runs where? At Wizeline, we actively transition clients from fixed kit to these software-defined operations through our Broadcast ^ AI and Media Supply Chain Optimization services.
Key takeaway: Buyers at IBC 2026 could evaluate interoperability as working code running across competing vendors.
The Work Changed Shape
Serious buyer conversations included “what does this cost to run” and “who on my team can actually operate it once it is live”.
The engineer who used to execute the task now spends their time deciding whether the agent executed it correctly. That is a harder skill to teach than the one it replaced.
As Grass Valley CTO Ian Fletcher put it, “complexity doesn’t go away, it just moves around.” It has moved into the judgment sitting above the task. Most organizations have not built a way to train or reward that judgment yet.
Key takeaway: The skill that matters now is judging the agent’s work, and few organizations train or reward it.
Is Token Economics Next?
This one did not make the main stage. It came up at Devoncroft, and in the halls and parties around IBC, which is usually where next year’s agenda starts. On that evidence, yes.
It is the accountability question again, in financial form. Most enterprise AI setups send every request to the best available model with the full context attached. A two-line summary and a board-level synthesis get the same treatment. Three things follow:
- Cost grows faster than value
- Spend can’t be attributed to a team, a role or a workflow, so nobody can manage it
- Retries and inefficient retrieval inflate the bill quietly
Average monthly enterprise AI spend hit $85.5K in 2025, up 36% from $63K the year before.
Media carries more of this than most sectors, because the workloads are large and constant. Transcription, translation, metadata generation, clip search. All of it runs across whole archives and every live feed. A routing decision that looks trivial on one asset is a line item on a million.
In our work, this is a governance problem before it is a procurement one. What matters is deciding, per request, which model the work deserves, how much context it needs, and what token budget it runs under, with cost logged against a team. The number to manage is cost per successful task.
Key takeaway: Token spend becomes manageable once it can be attributed to a team and a task.
Governance Is What Lets You Scale
Across agents, content, interoperability, people and cost, the pattern held. The organizations ahead of this decided accountability in advance. Everyone else is retrofitting governance onto systems already in production, which is slower, more expensive and more public when it goes wrong.
Governance for autonomous media systems is what makes scaling AI in production survivable. The teams that get that are shipping while their competitors are still writing policy.
Ready to review your AI governance and media supply chain? Book a discovery session with our M&E specialists.
