For broadcasters and service providers, implementing AI effectively means understanding the difference between automation, orchestration, and agents, and building in governance and trust from the start.
The New Operating Layer
AI in media and entertainment has evolved dramatically over the past few years, and that accelerated rate of change is not showing any sign of slowing down. It is not just a set of discrete tools deployed on specific tasks anymore, but effectively an operating layer that runs across entire operations. It routes content, monitors quality, makes routine decisions, and crucially learns from its own outputs. It is not replacing the people doing the work, rather it is running continuously underneath them.
The distinction matters, because the catch-all term “AI” currently covers three quite different things that many industry discussions treat as interchangeable. These also reflect the progress that the technology has made in operator deployments.
Automation is rule-based and was at the forefront of the AI industry rollout. It does exactly what you configure it to do, but does it reliably and at scale. Examples include automated metadata tagging, scheduled compliance checks, and flagging known violation patterns.
A step beyond that, orchestration coordinates multiple automated systems so they hand work to each other, turning a collection of tools into a connected workflow rather than a row of silos.
And agents go even further: they make contextual decisions within defined parameters, responding to conditions rather than simply executing instructions.
While representing a definite evolution of the technology, the three co-exist in the wild. The organizations making real progress are the ones who know which of these three they are deploying at any given point, and where their capabilities are the strongest.
Building from the ground up
Most broadcasters start their AI deployments in the same way, with a problem they need to fix. Maybe archive retrieval is slow, or QC misses issues that only become visible on specific devices. Perhaps dubbing backlogs are delaying international releases, or churn is rising and no one is entirely sure why.
The first wave of AI deployment addressed these friction points directly, providing smarter metadata, automated compliance checking, churn prediction, load forecasting, and more. These implementations worked because the problem was defined and the metrics were obvious.
More recently, AI has pushed quicker and further upstream than many expected, with agentic systems now handling script breakdowns, storyboard generation, and production scheduling before even a camera rolls.
It is when these separate capabilities start connecting that significant benefits occur. The newsroom, production, advertising, and distribution teams stop passing work between silos and start sharing a common logical layer. Content moves where it is needed rather than where someone's manual process sends it.
This is happening in the real world now too. AI can, for instance, orchestrate contextual advertising by analyzing video frame by frame to place dynamic insertions with real precision rather than rough approximation. And crucially, outputs feed back in, which means that the system improves as it runs in a constant positive feedback loop.
Sky Italia's delivery platform shows what this looks like in practice and shows what the concept of the AI operating layer can achieve. Its AI routes video data across the network in response to real-time conditions, maintaining buffer-free 4K streams for millions of viewers while anticipating demand spikes before they become visible as problems. The operational benefits include reduced egress costs, lower storage overhead, and less manual intervention. What’s more they compound over time.
The Operational Reality
All the industry is being shaped by AI deployments, but three areas in particular have seen the sharpest transformation.
Operations teams have been using automated monitoring, rule engines, alarms, and analytics for years. AI has added better anomaly detection, detection of complex patterns, prioritization and fast root cause analysis. That enables it to surface issues more quickly, whether that be a graphics trigger misfiring at the wrong moment, an audio channel dropping, or a compliance violation that would otherwise have surfaced only as a viewer complaint.
During one recent global sports broadcast, for example, AI caught graphic rendering errors on mobile devices and switched automatically to a backup encoder. No viewer noticed, which is the ideal scenario for any technical mishap. This is now the standard against which manual monitoring is measured.
Localization has long been where attempts to scale break first. The sheer volume of processes involved, from translation to subtitling, compliance edits, and platform-specific packaging, has consistently outpaced what human teams can deliver at competitive cost. AI has absorbed much of that load, handling audio description and SDH subtitling with a consistency that would be difficult to sustain manually. More recent dubbing models are becoming ever-more sophisticated and are capable of preserving emotional nuance rather than just mechanically converting dialogue. Netflix is one that reports this has had measurable benefits in terms of increased completion rates for global titles being watched with subtitles.
However, sports broadcasting may be where AI's speed advantage is currently most visible. Highlight packages that once took hours to produce are now assembled in near real-time by systems that understand what counts as a significant moment in a given sport, rather than applying generic edit logic. NBC Universal's personalized Olympic recaps which were generated at impressive scale for individual viewers, show what becomes possible when AI orchestration is combined with audience data.
Notably, the same systems generating those clips also forecast audience surges and adjust cloud resources accordingly. This is precisely what an AI operating layer looks like in practice.
The problem of trust
None of this scales cleanly unless the content can be trusted. The risks of synthetic media are real and growing, whether that be fabricated promotional materials, manipulated clips, or AI-generated presenters delivering false information. And once synthetic or contaminated content enters a workflow, it becomes progressively harder to detect and more costly to fix.
The practical answer is to treat authenticity as a core requirement when building out workflows, not as an afterthought. A functioning trust stack has three primary components. Digital watermarking first creates durable identifiers that persist through all production and post production processes. Provenance frameworks built on open standards, the C2PA protocol in particular, create cryptographically signed records of an asset's origin and any transformations it has undergone; France Télévisions and ARD have both implemented daily use of these protocols for VOD content. And hardware-backed authentication, implemented in C2PA-enabled cameras, confirms at the point of capture that material comes from a verified, real world source.
Adoption is uneven and there are many challenges. Metadata can be stripped, social media distribution remains a significant gap, and key management is genuinely complex. These are real constraints against adoption, but the alternative is operating at scale without this integrity layer in place. Increasingly that carries unacceptable levels of brand and legal exposure that will only grow as AI becomes more central to production.
Where to begin
The broadcasters making the most progress with AI deployments share a common approach: they have all started with a specific, measurable problem rather than an abstract AI strategy.
The most effective strategy therefore is to pick one or two areas for ramping up AI use where friction is highest and the metrics are clear and measurable, such as time-to-air, versioning throughput, or detection-to-resolution times on QC failures. And we say ‘ramping up’ because a workflow audit will often reveal that AI is already operating in parts of the business. It is these areas that are the logical places to formalize and extend.
Governance needs to be taken into account early, not because AI is inherently risky, but because agentic systems need clearly documented human-override paths before they can reliably operate at scale. Any and all AI deployments should include provenance and authentication requirements from the start, so integrity is designed in rather than retrofitted. And inbuilt checks and balances, the ability of editorial and technical teams to evaluate, shape, and challenge AI outputs, is at least as important as the technology itself.
The near future
Netflix's acquisition of Interpositive AI is one of several signals that leading media companies are beginning to treat AI as core infrastructure rather than an additional suite of tools that sit alongside legacy systems.
The shift that matters over the next three to five years is not simply about AI doing more tasks, it is about AI making more decisions. We are going to increasingly see a movement away from systems that simply execute instructions as with the early AI deployments surrounding automation, to agentic systems that apply judgment themselves within defined bounds. This changes the nature of the operation itself, not just its efficiency, and creates new workflows rather than simply faster ones.
Broadcasters and service providers who understand the difference, and build their governance, their trust stack, and their team capabilities around it, are the ones who will be setting the pace and gaining the most advantages from the new AI operating layer. The rest will be catching up.
