Industry Insights: Integrating AI into workflows and production systems

As artificial intelligence becomes more embedded in broadcast operations, the challenge is no longer simply adding new tools. It is integrating intelligence into the systems media organizations already rely on every day.

In this second installment of our three-part Industry Insights roundtable, vendors examine how AI is being woven into newsroom systems, production environments and media supply chains rather than layered on top as separate platforms.

The discussion explores how APIs, orchestration tools and intelligent asset management are enabling AI to operate inside existing workflows.

Participants also address the operational barriers that still slow adoption, including legacy infrastructure, inconsistent metadata and cultural resistance inside production teams. Together, the conversation highlights the technical and organizational conditions that determine whether AI becomes a useful operational layer or remains a disconnected experiment.

Key takeaways from this Industry Insights roundtable

• Integration over overlays: AI deployments succeed when capabilities are embedded directly into existing systems such as NRCS, MAM and automation platforms rather than introduced as separate tools.
• Legacy systems challenge: Many integration challenges stem from inconsistent metadata structures, fragmented archives and legacy infrastructure that were not designed to support real-time AI outputs.
• Metadata foundations: Clean, governed metadata is widely viewed as essential for AI reliability, influencing everything from search and automation to compliance and monetization workflows.
• Asset management role: Intelligent asset management systems provide the operational backbone that allows AI to scale across production, post, archive and distribution workflows.
• Trust and workflow design: Adoption often stalls when AI tools disrupt established production processes or lack clear review mechanisms, making transparency and consistent performance critical for operational trust.

How is AI being integrated into existing production and newsroom systems rather than layered on top?

Yang Cai, CEO and president, VisualOn: Successful deployments integrate AI directly into existing encoding, packaging, and delivery pipelines through APIs and workflow orchestration tools. This enables AI to function as a decision layer within established systems, rather than as a disconnected overlay that adds operational friction.

Josh Rosen, VP, production and content, North America, Gravity Media Group: The deployments that really stick treat AI output as system-native metadata that writes directly back into MAMs, NRCS platforms, automation systems, or asset databases. When AI output becomes part of the normal operational fabric, adoption rises significantly. When it lives in a separate interface, usage drops quickly under production pressure. For example, once the AI tools and automation were integrated directly into the MAM at one of our large Silicon Valley sites, accurate tag density increased by 40%+ and successful AI back-end content searches increased by nearly 65%.

Scott Goldman, general manager, U.S., Qibb: AI is being integrated directly into NRCS and MAM environments through APIs and connectors, so teams don’t need to adopt separate AI tools. Producers and journalists see AI-generated metadata, clip suggestions, and search results inside the systems they already use. This embedded approach reduces friction and drives real adoption across newsroom workflows.

Ken Kobayashi, business manager, remote cameras, Sony Electronics: One way AI is being integrated into existing production and newsroom systems is through the use of a camera’s automation features. The main camera will continue to be controlled by an operator, but subsequent cameras or even prompter cameras can be driven by PTZ cameras featuring AI capabilities. PTZ cameras are already widely installed in the studio or newsroom, but are mostly used for fixed angle shots.

Advertisement

Sam Peterson, COO, Bitcentral: The most effective AI is exposed directly within the systems producers already use, rather than introduced as a separate toolset. Transcription, tagging, and clip suggestions are surfaced through existing workflows, minimizing disruption and training overhead. This approach accelerates adoption while preserving established editorial processes.

Danny Hollingsworth, director, post production product marketing, Avid: New capabilities are increasingly embedded directly into creative tools and content platforms already in use. For example, the MediaCentral Rules Engine is embedded into the Avid ecosystem ensuring auditable automation logic. This includes task led automation that can move content through defined steps without requiring manual handoffs between systems.

Zoe Liu, CEO and founder, Visionular: In our experience, successful deployments integrate directly into encoder SDKs or cloud encoding workflows, use standard interfaces compatible with existing MAM and OTT systems and operate invisibly within established automation chains. AI-driven encoding optimizes internal encoding decisions while remaining fully compliant with video coding standards. The generated bitstreams remain strictly standard-aligned, ensuring full compatibility with existing decoders and players, without requiring any special playback infrastructure.

Simon Wocka, regional sales manager, Pixitmedia: The successful implementations integrate AI into ingest, archive, search, and data movement through APIs and platform-level connections. This enhances existing MAMs, creative tools, and automation systems rather than replacing them. Customers want their workflows enabled, not disrupted, and they want AI to make existing systems work harder rather than ripping everything out and starting again.

What challenges arise when AI tools intersect with legacy NRCS, MAM or automation platforms?

Josh Rosen, VP, production and content, North America, Gravity Media Group: The biggest challenges are schema mismatches, inconsistent taxonomies, and fragile handoffs between legacy systems and newer AI services. Even when AI output is accurate, it can fail operationally if metadata fields don’t align or automation rules weren’t designed for richer data. Most issues turn out to be plumbing and standards problems rather than model limitations.

Scott Goldman, general manager, U.S., Qibb: The biggest blockers come from inconsistent metadata, fragmented archives, and aging infrastructure — not the AI models themselves. Legacy systems often lack clean naming conventions, timecode consistency, or structured fields needed for reliable automation. AI can help bridge gaps, but long-term success still depends on improving data quality and modernizing workflows.

Sam Peterson, COO, Bitcentral: Legacy systems were not designed for AI-generated outputs, which creates challenges around compatibility, validation, and editorial trust. Newsrooms often need to adjust workflows to ensure AI-driven metadata and clips meet broadcast standards. Starting with pilot projects in high-volume, lower-risk areas helps teams prove value before expanding AI deeper into core production systems.

Danny Hollingsworth, director, post production product marketing, Avid: Inconsistent metadata structures and asset identities limit the effectiveness of automation and intelligent orchestration. Many legacy systems were not designed to support ongoing enrichment or rule based task coordination. This creates pressure to unify content and metadata layers before more advanced automation can be applied reliably.

Miguel Coutinho, head, NDI: A lot of older systems expect updates in batches, not as a constant stream. When AI starts generating live scene data, those assumptions can cause lag or dropped information. Making the two worlds talk to each other in real time is usually the harder job than the AI itself.

How important is metadata quality in determining whether AI initiatives succeed or fail?

Yang Cai, CEO and president, VisualOn: Metadata quality is foundational, as AI outputs are only as reliable as the inputs they analyze. Inconsistent or incomplete metadata quickly erodes confidence in automated decisions, especially in large-scale video operations.

Josh Rosen, VP, production and content, North America, Gravity Media Group: Metadata quality is foundational because it determines whether people trust search, retrieval, and downstream automation. If tags are inconsistent or noisy, teams quickly revert to manual methods and institutional memory. The less visible work — taxonomy design, standards, and cleanup — often delivers the highest long-term ROI.

Clara Aler, head of marketing, Knox Media Hub: It’s fundamental. In M&E, metadata is what drives automation. Poor metadata directly undermines AI-driven decisions, breaks workflows or results in an inaccurate dataset to train AI models. Successful AI deployments rely on a centralized, governed metadata backbone to ensure consistency across the entire supply chain.

Advertisement

Scott Goldman, general manager, U.S., Qibb: Metadata quality is one of the strongest predictors of whether AI delivers value; poor metadata inevitably leads to poor AI outcomes. The real impact of automation is turning unstructured media into consistent, searchable intelligence. With clean, enriched metadata, AI becomes far more accurate and scalable.

Jacob Arends, senior product manager, playback and AI, Bitmovin: Metadata is critically important in media workflows, and having accurate, meaningful metadata is essential for the successful deployment of many AI-driven tools. For example, AI tools that support contextual monetization and targeted advertising, as well as personalized playback and discovery, rely on a foundation of high-quality metadata. Data integrity based on AI workflows is not a straightforward challenge and so reliability and scalability are priority concerns for any enterprise level deployment.

Sam Peterson, COO, Bitcentral: Metadata quality is foundational. AI can’t surface the right content if the underlying data is inconsistent or incomplete. Poor metadata does not just limit AI’s effectiveness, it exposes existing discovery problems faster. Clean, structured metadata enables AI to deliver real operational gains instead of creating more review work.

Rich Welsh, president, SMPTE: This is realistically the same value metadata holds in general. The better quality and depth of your metadata, the better any automation system will perform. Agentic workflows are an obvious area where metadata can drive significantly better outcomes, and metadata quality can make or break the usefulness of this approach.

Danny Hollingsworth, director, post production product marketing, Avid: Reliable metadata is essential for accurate search, task automation, and content routing. Automatically generated data quickly loses value if it cannot be aligned with existing asset records and business rules. Strong metadata practices allow assisted and automated workflows to scale across teams.

Steph Lone, global leader, solutions architecture, media and entertainment, games, and sports, Amazon Web Services: Metadata drives AI, so high-quality analysis is vital for success in downstream applications. Content owners often have petabytes of video and assets to manage, whether for post production, delivery, or archival purposes. Implemented properly, AI can generate quality metadata, which can then be used in agentic workflows for extracting technical specifications, summaries, and both chapter- and shot-level details; or creating media operations agents for specific tasks, such as celebrity detection or quality control. Agentic AI systems can also reduce costs for features such as ad break detection, skipping credits and semantic search.

Phil Petitpont, co-founder and CEO, Moments Lab: The fast rise of agentic AI in media workflows makes metadata quality more important than ever. Any AI-powered tool is only as good as the data it has to work with. The good news is that multimodal AI is able to comprehensively index video, breaking it into scenes and enriching it with the detailed metadata needed for agentic AI to pinpoint exact clips to accelerate story production.

Miguel Coutinho, head, NDI: Metadata is extremely important in two ways. First, existing metadata associated with a file that has been archived can help the model learn what the file is about. On the other end, when auto-tagging a file with metadata, it’s important the AI model is following a clear standard. Secondly, in a live production, metadata can help keep everything in sync which is extremely useful if an AI model is doing auto-switching, or stitching different sources together.

Geoff Stedman, chief marketing officer, SDVI: Many AI initiatives are focused on improving metadata quality, particularly for archives that lack the thorough descriptive information needed to improve content discoverability and enhance viewing experiences. The success or failure of these initiatives is largely based on how accurate the AI tools are at extracting descriptive information (and associated timecode) so that the metadata is usable and searchable. For some use cases, such as (semi)-automated caption creation, accuracy is critical in order to satisfy regulatory guidelines, but in others, having any data at all is vastly superior to having none, even if it’s not 100% accurate.

Simon Wocka, regional sales manager, Pixitmedia: It’s absolutely critical. If the underlying metadata is inconsistent or poor quality, the value of AI drops off very quickly. But when customers invest in getting their metadata foundations right, everything else becomes easier and more scalable. You can’t build reliable AI workflows on unreliable data.

Advertisement

What role does intelligent asset management play in enabling broader AI use?

Aitor Falcó, sales manager, Knox Media Hub: Intelligent, cloud-native MAM is the enabler. It provides a single operational view of content, metadata, rights, and status across the supply chain. This allows AI to move beyond analysis into execution — triggering workflows, enforcing rules, and coordinating processes across ingestion, processing, compliance, and distribution.

Scott Goldman, general manager, U.S., Qibb: Intelligent asset management is the foundation that allows AI to scale beyond isolated use cases. When assets are consistently tagged and searchable, and rights-aware, teams can quickly find, reuse, and move content across production, digital, and monetization workflows. Without this layer, AI remains limited to point solutions rather than supporting end‑to‑end operational impact.

Derek Barrilleaux, CEO, Projective: I recently heard the line from a colleague: “You can’t plug AI into chaos.” That completely aligns with our philosophy: Get your house in order, establish a sensible foundation, and then you can start to get value from AI. With a framework in place for your post production, now you can leverage AI on that framework and actually provide value to the organization.

Rich Welsh, president, SMPTE: Content intelligence is one of the most mature areas using AI and we’re seeing enormous upsides to understanding asset libraries better using analysis tools. This is not only enabling better search and retrieval, but extends into new monetization of catalogue and huge opportunities in hyper-localization and personalization. Ad revenue generally follows on from this, and I can see a time where content is entirely personalized simply because this will be a requirement from advertisers.

Danny Hollingsworth, director, post production product marketing, Avid: Intelligent asset management provides a shared framework for media, metadata, and rights information. This enables guided and automated workflows that can operate across production, post, and archive. It allows automation to act as a connective layer rather than a point solution.

Simon Wocka, regional sales manager, Pixitmedia: Intelligent asset management provides the foundation that allows AI to scale beyond isolated projects. You need unified visibility across storage tiers, controlled data movement, and consistent discovery. Without that foundation, AI value stays fragmented. You might have some nice capabilities in pockets, but they never compound across the organization.

Where has AI adoption stalled due to workflow complexity or cultural resistance?
How are organizations deciding which AI tools are worth operational trust?

Michael Chan, VP, delivery operations in corporate, Accedo: Workflow complexity and legacy system integration represents the highest barrier to scaling AI beyond isolated pilots. Legacy platforms, databases and management systems were designed for human-editable content workflows. Integrating modern AI tools requires a modern interface that existing systems lack; upgrading them or replacing them is a capital intensive decision.

Josh Rosen, VP, production and content, North America, Gravity Media Group: Adoption tends to stall when teams are asked to adopt AI tools without rethinking how work actually flows through production and post. Editorial resistance is common when AI touches accuracy, context, or brand voice without clear review loops or accountability. Under real deadlines, anything that adds steps or uncertainty is quickly bypassed.

Rich Welsh, president, SMPTE: We’re seeing some interesting challenges around perceived efficiency gains versus reality in general. There have been recent indicators that expected gains are triggering workforce layoffs. Those gains are either not realized, or hiring to support new AI workflows is totally offsetting the savings made elsewhere. There is obviously the question of mass layoffs in general but realistically AI isn’t having that impact in media at the moment.

Danny Hollingsworth, director, post production product marketing, Avid: Trust is built through repeatable results, transparency, and a clear understanding of how automated or assisted actions are triggered. Organizations favor systems that guide users through tasks while keeping editorial decisions with humans. Tools that demonstrate consistent performance at scale are more likely to be trusted in live production.