What DAM and content vendors are saying about AI readiness

The DAM will continue to be the System of Record for digital content. This will be even more important as AI continues to disrupt the content supply chain. AI built without governance is at best ineffective and at worst a liability. What AI will enable is the increase of users to the DAM but through AI Assistants like Claude or Copilot. They will be able to access the DAM without ever interacting with the interface by pulling content from the DAM to kickstart campaigns or creating content based on brand guidelines.

Brands need to have their system of record in place. The content you have in the DAM needs to have the right taxonomy and metadata structure, fit to your business and users need to have the correct user permissions so they can only access what they need – whether in the DAM or outside.

AI makes the right metadata more important. Metadata is an important part of maintaining that system of record for your content but what has changed is that metadata now needs to be built for machines rather than keyword searches. And information like what exactly is in the image becomes less important (especially with things like NLS) but information like usage and rights restrictions and creation context (was this created by creatives or AI) becomes more important.

Agents are moving DAM from a passive repository into an active layer of the content supply chain. The shift is from “search and retrieve” to “configure and delegate”. Instead of a person finding an asset and doing the work by hand, an agent handles the enrichment, transformation, localization, or compliance check, and a human reviews the outcome. Customers are approaching this in phases:

• Early/Mid-adoption: Testing single-purpose agents (alt-text generation, tagging, duplicate detection) — low-risk, high-volume tasks where errors are cheap to catch.
• Ahead of the curve: configuring agents to run workflows using their own taxonomy and brand rules with humans leading the decision-making, for example adapting on-brand content from the Bynder DAM in figma, checking that it’s brand compliant with the brand compliant agent, enriching the asset with business and product-specific metadata as it’s uploaded into the DAM and transforming that to fit channels, marketplaces, and geographies. Customers want agents that are native to their DAM’s context, connected to their taxonomy, metadata, brand guidelines.

Controls need to cover permissioning (who can configure/approve), provenance (C2PA-style signed metadata on what’s AI-touched), and brand-specific guardrails so agents follow your rules. Human-led should be the operating model, especially with EU AI Act obligations now in force.

Value today: accessibility (alt-text at scale), metadata enrichment, localization, and asset transformation/repurposing. What’s holding others back: distrust in output accuracy, messy underlying taxonomy, and unclear accountability when an agent gets it wrong.

We support all of the above. Bynder preserves third-party markings supporting C2PA requirements, organizes and filters AI content, and can label AI created content. We have generative AI capabilities in our Studio product as well as Transformation Agents to run transformations at scale and three other Agent categories (Governance, Brand Compliance, and Enrichment). The latest AI capability is with our MCP server which allows the access of Bynder through AI assistants. Users would be able to retrieve assets via AI Search and interact with brand guidelines via tools like Claude and Copilot.

DAM will play a critical role in the enterprise AI ecosystem by providing the trusted, structured and governed content that AI systems need to work effectively. As organizations use AI to create, enrich, search and distribute content, DAM can act as the foundation for managing the assets, metadata, permissions and workflows that surround that content.

The role of DAM is therefore evolving from a content repository to an intelligent content layer, connecting people, content, AI and business systems. By combining strong metadata, automation, workflows and governance, DAM can help organizations make their content more accessible to AI while maintaining control, security and consistency.

AI-driven workflows only work when they can act on structured, accessible, real-time data about content. Asset-related truth is spread across systems: the CMS knows the content, the PIM knows the products, the rights database knows licensing and copyright. AI can’t reason across that if it lives in silos. The DAM is the natural place where that asset intelligence is consolidated, governed and made actionable – which is why we in Fotoware believe DAM is moving from being a departmental image library to being a company wide system of action – making it strategically relevant to the enterprise AI ecosystem.

Our best advise would be to not start with technology. AI readiness has everything to do with clear answers to what, why and how. That’s easy to lose sight of when the technology moves this fast and the possibilities are this exciting.

Fotoware believe five things need to be in place, and in our conversations we see several skip the first two.

1. Start with the why. Before anything else, the organisation needs a clear, shared understanding of where it wants to get value from AI and what that value actually is. Not “we need an AI strategy” – but which workflows, for which people, with which outcome. Without that, AI becomes random acts of innovation: activity that looks impressive but changes little and gives little long-term value. Throughout our product development, we leverage all technology, including AI, following a human-centric approach and starting with the problem: Person → Process → Technology → Validate: understand who is involved and what outcome they need; design the workflow steps, roles, decision points and governance; then apply AI and automation exactly where they meaningfully enhance the flow; then test with real assets and real users before scaling.
2. Competence and control points – the “how”: There’s a lot of talk about human-led AI, but being human-led is an operational capability that must be invested in. It requires people who know how to set AI up, use it well, and check its work. If AI is to act based on content and data, a clear and detailed permission model must be in place. Access controls must be defined per user, per asset, per data value and per action. It’s no longer enough to group every user into one of a handful of generic access levels. Rather, DAM admins must ensure that access is tailored based on the user’s task at hand and also set restrictions on the data presented along with the given assets. By doing this, the organization enables itself to judge whether an AI output is right, giving the organization control. Adoption and compliance have to go hand in hand.
3. Accessible data. If critical information sits in systems or formats that can’t expose it through APIs, no amount of integration capability makes it usable. Automation requires machine-readable inputs; if data can’t be accessed programmatically, AI can not use it in automation.
4. High-quality and structured metadata: AI value scales with the quality of the data supporting it. As AI accelerates content growth and system complexity, we must direct our efforts toward strengthening metadata quality and structure. AI-driven workflows and automated decision-making rely on predictable, high-quality metadata: well-defined fields, consistent naming, controlled vocabularies. When metadata varies between systems, or similar concepts are labelled differently across departments, automation stops – the system doesn’t know what to match or how to act. Structured data creates the semantic clarity automation depends on.
5. Provenance recorded at ingestion and kept intact: In addition to implementing AI, you also need to be able to show documentation of the files that are free from AI manipulation to ensure they remain trusted. Additionally, tracing a file’s source and lifecycle is a crucial part of the DAM’s role within the organization, and is central to ensure that your files are handled with their complete value intact, and that system users know which files to use and for which purposes.

The difference is whether the metadata is tied to a decision someone needs to make. Useful AI metadata is written into fields that a workflow or a search actually uses, in a vocabulary the organization has agreed on, and it can be reviewed. Noise is a long list of generic descriptors written into a free-text field that nobody queries.

Most auto-tagging disappoints for the same reason. It describes what is in the picture, accurately, and nobody needed that. A tag saying “sky, outdoor, person” is correct and useless. A tag identifying the product line, the campaign, the case file or the consent that applies is something a workflow can act on. Generic accuracy is not the goal. Workflow relevance is.

That means automatically generated tags should be mapped to a controlled vocabulary that matches the category the asset belongs to, and be specific to the collection they are used for rather than the output of generic object recognition. In practice this often requires a trainable AI solution that produces a customized set of tags based on prior learning, instead of a model trained on the world at large. It also means the tags need to land in a defined field rather than a general keyword pile. Metadata in a free-text field is searchable at best. Metadata in a specific, structured field can be evaluated by a rule, matched across systems and used to route work. The same tag is useful or noise depending on where it is written.

AI metadata also needs to be auditable and correctable. A person has to be able to see what the model wrote, fix it, and have that fix stick. This is the practical meaning of human oversight: AI proposes at volume, people stay accountable for the record. It follows that AI-generated values should be distinguishable from human-verified ones, so that the organization can treat them accordingly and knows later which is which. Without that, volume without review produces confident, consistent, wrong metadata at scale, which is harder to correct than having none at all.

AI has made generating metadata almost free, so quantity is no longer the constraint. Judgement is. The organizations getting value from auto-tagging are the ones that decided what they needed the metadata to do before they turned it on.

Looking ahead, this matters more, not less. AI is moving beyond generating metadata toward acting on it: recommending actions, routing content and data, triggering workflows, suggesting approvals and optimizing publishing decisions. Our strategy in Fotoware points toward AI delivering agentic intelligence throughout, with human oversight. Every one of those actions is only as good as the metadata underneath it, which is why the quality bar on generated metadata is a question about trust rather than tidiness.

AI makes metadata dramatically more important.

The assumption is that because AI can generate tags and descriptions, metadata discipline becomes optional. The opposite is true. Metadata is the multiplier of AI value: it’s both the input AI reasons over and the output that makes AI’s work trustworthy and reusable. Many initially assume AI will eliminate the need for metadata because AI can “understand” images, videos, and documents automatically. In reality, enterprise AI increases the value of metadata because AI needs context, governance, and trusted information to operate safely and accurately.

Two things change. First, volume raises the stakes. AI accelerates both content creation and variation, so poor metadata no longer causes a small findability problem – it causes automation to fail. Second, metadata becomes legally load-bearing. AI-related regulation is expressed through metadata: usage rights, provenance, transformation history, consent. If that metadata is wrong, incomplete or missing, teams can unknowingly publish content that violates regulation and expose the organisation to reputational, legal or financial risk.

So, AI changes how metadata gets created – enrichment, auto-tagging, captions, OCR, translation, custom models trained on your own content – but it raises the bar on metadata governance. That’s why we in Fotoware are investing in metadata quality and metadata-driven workflows as a foundation for everything else.

We see a progression based on results, learnings and values created. This is true both in our own product development and in our customer conversations. These progressions mean we are moving step-by-step from simple automation → to intelligent automation (business logic combining data from many systems into decisions) → to agentic intelligence (systems that actively propose solutions and take decisions).

Our vision in Fotoware is an AI-enabled content workflow automation ecosystem. The end game is agentic intelligence throughout, where agents act on governed, structured data inside defined workflows while ensuring human validation points throughout. An agent is only as trustworthy as the metadata and business rules it acts on.

The near-term entry points for agents in content operations:

• Decision support in workflows – instead of manually checking license, embargo date and intended use across three systems, business rules evaluate the conditions and return one clear, always-up-to-date answer: can I publish this?
• Compliance guidance – a compliance view that doesn’t just show status, but tells you what is compliant, what’s at risk, what requires action, what should be deleted or restricted, and guides you through the next steps.
• Workflow optimization – AI-powered insights that suggest ways to improve, create and streamline workflows based on real usage patterns.
• Distribution – automatically generating and populating shares from business rules, so distribution follows the same governance logic as the rest of the workflow.

Our position is the one we apply to agents generally: AI should act on governed, structured data inside defined workflows, not roam freely across an unstructured library. AI is only as trustworthy as the metadata and the business rules it acts on. So the controls that matter are the ones that define what AI is allowed to do, and who confirms that it did it correctly.

Define the boundaries before you automate. Decide which decisions AI is allowed to make on its own, which ones it may prepare but not execute, and which ones stay with a person. This is an organisational decision, not a system setting. In our experience it is the step most often skipped, and it is the one that determines whether the rest holds.

Grade the controls by consequence. Requiring human sign-off on everything does not scale, and in practice it turns into approval without real review. Tie the level of control to what happens if the decision is wrong. Enrichment that affects search can run automatically. Publishing, external sharing, deletion and anything with legal or privacy exposure needs a person who confirms it.

Make the rules machine-readable. A policy in a document does not stop an automated workflow. Rights, licence terms, consent status, embargo dates and approved usage have to exist as structured fields the workflow evaluates before it acts.

Name who is accountable. Every automated decision needs an owner who validates that the output is correct and who can intervene. Human-led AI is an operating model with named responsibilities.

Be able to reconstruct what happened. Preventive controls will not catch everything. Traceability of every access, change and action, together with verifiable provenance for the asset itself, is what lets you find the error, correct it and demonstrate what took place. This comes from our work in law enforcement, where it is non-negotiable, and it is increasingly expected everywhere.

One point underneath all of this. These rules are expressed through metadata: usage rights, provenance, transformation history, consent. That makes the DAM the place where this control either works or fails. If the metadata is wrong, incomplete or missing, teams can unknowingly publish content that breaches regulation. A person might notice. An automated workflow will not.

Today the value sits in making large volumes of content findable, usable and compliant.

Enrichment at scale. Customers use auto-tagging, object tagging, people and face detection, image descriptions, OCR and brand and logo recognition to describe content that would otherwise arrive undescribed. The point is not that AI describes better than a person does. It is that the realistic alternative was no metadata at all, because nobody had time. A media house ingesting thousands of images a week, or a police force processing body-worn camera footage, cannot staff manual description at that volume.

Findability. Generated descriptions, extracted text and translated metadata let people locate assets they did not know existed. For organizations working across languages and markets, metadata translation removes a practical barrier to using a shared library.

Privacy and rights work. Face and people detection supports privacy filtering. Brand and logo recognition supports editorial and rights decisions. Both are compliance tasks that previously required someone to review every image or frame.

Decisions, not only descriptions. The most valuable pattern we see is AI and automation combined with business rules, so the question a person actually has gets answered. Instead of checking licence terms in one system, embargo date in another and intended use in a third, the workflow evaluates the conditions and returns one clear, current answer: can I publish this.

Two things separate the customers getting value from those who are not. The first is whether the metadata is structured enough for AI output to be acted on rather than only stored. The second is whether the organization decided what it wanted the AI to do before switching it on.

Our direction from here is to move further up that chain. We have next up on the roadmap natural-language asset discovery for complex queries and are working on duplicate detection to keep collections clean, AI-powered content modification, flexible metadata translation where users choose fields and languages, workflow optimization insights based on real usage patterns, and custom AI models that learn from user input and suggest tags specific to a customer’s own content. We are also investing in compliance automation and content authenticity, so that as AI takes on more of the work, the record of what was done to an asset stays intact.

Our perspective in Fotoware is that there are two barriers: the technical and the organizational barrier. They are not equally hard to solve.

The technical barrier is data readiness. Asset-related information still lives in multiple places, often unstructured, so organizations struggle to deliver a unified, real-time view that automation can act on, and that becomes a barrier to scaling intelligent workflows. AI applied to a fragmented, inconsistently tagged library returns fragmented, inconsistent results, which then confirms everyone’s suspicion that AI doesn’t work here.

The organisational barrier is bigger, and it is the one we would highlight. There are three parts to it.

First, many organisations have not settled why they are using AI and where they expect the value. Without that, AI work becomes scattered activity that is hard to prioritise, hard to measure and hard to defend when budgets are reviewed.

Second, competence. Human-led AI requires people who can set AI up, use it, and check its output, plus defined control points and clear accountability for what an automated workflow decides. Where that competence is missing, organisations tend to either hold back entirely or adopt without oversight. Neither produces value.

The third part is governance uncertainty. Teams in regulated or reputation-sensitive sectors hold back not because AI can’t do the task, but because they can’t yet prove what it did. That’s precisely why DAMs needs provenance tracing in place, if vendors are to expect wide-range industry adoption.

Data problems can be fixed with a project plan. Clarity of purpose, competence and governance strategies take longer to build, which is why we think they are the real constraint. The gap between organisations that address this and those that do not will grow quickly.

Yes. Fotoware is a member of the Content Authenticity Initiative (CAI) and we support C2PA in full. Our Content Authenticity capability gives organizations complete end-to-end support for the C2PA standard inside the Digital Asset Management platform, documenting and proving the provenance of every image and video from the moment of capture through to publication. We will continue to invest in these capabilities.

Yes, and we apply them where they improve a workflow outcome.

In use today: AI-generated image descriptions and captions that enrich metadata and improve accessibility, generated descriptive tags that power search and discoverability, and metadata translation of tags, descriptions and OCR text into multiple languages. These run as part of normal ingestion and enrichment rather than as a separate AI tool users have to think about.

We are investing in content modification through AI-powered integrations such as background removal, colour enhancement, face blurring and editing, to enable flexible asset reuse. This is particularly relevant in retail and marketing, where the same asset needs to work across many channels. We are also investing in natural-language search that handles complex queries, and in custom AI models that learn from user input and suggest tags specific to a customer’s own content rather than generic descriptors.

How we decide what to build matters more than the list. We develop generative capabilities sequentially, using results, learnings and the value created to determine the next step, and we start with the problem rather than the technology: understand who is involved and what outcome they need, design the workflow, then apply AI where it meaningfully improves that workflow, then validate with real assets and real users before scaling. Software has to deliver value, not only functionality.

One point we would add for readers of this guide. Generation is only half of the responsibility. As generative tools put content creation in everyone’s hands, a DAM vendor’s obligation is as much about governing generated content, its provenance, rights and usage, as about producing it. That is why our work on generative capabilities runs alongside our work on content authenticity and compliance automation rather than separately from it.

Not autonomous agents in production today, and we are deliberate about the sequence. What is in the product is the layer agents depend on. Dynamic Variables combine data from connected systems with a customer’s own business rules and return an answer rather than a set of fields: licence type, embargo date and intended use evaluated together into one current value that says whether an asset can be published. Conditional workflows in Flow mark compliance state automatically when assets meet defined criteria. Configurable notifications route the right information to the content owner, the marketing team or the compliance officer. Automated share creation generates and populates shares from business rules, so distribution follows the same governance logic as the rest of the workflow. Flow analytics shows which workflows are running and which rules and actions are used most.

That is decision-making driven by rules and data, with a person accountable for the outcome. It is the step before agency, and it is the step that determines whether agency is safe.

Our stated direction is to move from simple automation, to intelligent automation, to agentic intelligence where the system proposes solutions and takes decisions, with human validation points throughout. The near-term entry points we are working toward are decision support in workflows, compliance guidance that recommends next steps rather than only reporting status, workflow optimization based on real usage patterns, and rule-driven distribution.

Our position on why the sequence matters: an agent is only as trustworthy as the metadata and the business rules it acts on. Turning an agent loose on an unstructured library with inconsistent metadata does not produce value, it produces liability at machine speed. Our customers work in newsrooms, law enforcement, public institutions and regulated industries, where an automated decision that cannot be explained or traced is a bigger problem than a task that took longer. We would rather ship agency that holds up in those settings than ship it first.

Content Authenticity, now launched. Strictly speaking this is not an AI capability, but it is our most important release in the AI context. Built on the C2PA standard as a member of the Content Authenticity Initiative, it records an asset’s provenance in a cryptographically bound Content Credential: where the file came from, what has been done to it through its lifecycle, and whether AI was involved in creating or modifying it. As generated and AI-modified content becomes ordinary, that last point is what lets an organization tell the difference between a retouched background and a fabricated scene, and document the difference to others.

AI capabilities in daily use. Auto-tagging and AI-generated image descriptions enrich content at ingestion, so large volumes arrive described rather than undescribed. OCR extracts embedded text and stores it as searchable metadata. People and face detection support privacy filtering and people-based search. Metadata translation makes tags, descriptions and extracted text usable across languages and markets. These run inside normal workflows rather than as a separate AI tool.

We are investing next in natural-language asset discovery, duplicate detection, AI-powered content modification, flexible metadata translation, workflow optimization insights, and custom AI models trained on a customer’s own content.

Compliance automation. Conditional workflows in Flow use business-logic-driven variables to mark an asset’s compliance state automatically when it meets defined criteria and approved field combinations, which turns rules that used to live in a document into something the workflow enforces.

The layer agents depend on. Much of our recent investment has gone into the foundation that makes automated decisions trustworthy rather than into agents themselves. Dynamic Variables combine data from connected systems with a customer’s own business rules and return an answer rather than a set of fields, so that a workflow can determine whether an asset may be published. Automated share creation applies the same governance logic to distribution. Flow analytics shows which workflows and rules are actually being used.

Alongside this we are building out the connectable DAM: API-first unified APIs, a modular UI extension framework, SDKs and widgets, and more plug-and-play connectors, so asset data can move between systems in real time. All these investments take us toward our vision of an AI-enabled content workflow automation ecosystem. The end game is agentic intelligence throughout, where agents act on governed, structured data inside defined workflows, with human validation points along the way.

DAM will increasingly supply the context that enterprise AI needs to work responsibly with content. An AI tool may be able to find an asset or generate a layout, but it also needs to know which asset is approved, which brand and market it belongs to, how it may be used, and which guidelines apply.

We see this as a fundamental shift in what a DAM is for. Brand knowledge lives in the DAM. Brand intelligence travels everywhere. This becomes your system of record for brand standards, guidelines, policies, assets, and context, while allowing that intelligence to flow into the tools and AI systems your teams already use.

By turning assets, guidelines, metadata, permissions, rights, and rules into structured, machine-readable knowledge, the DAM gives AI the governed context it needs to make accurate, on-brand decisions.

For enterprise brands, this is becoming more and more vital. Enterprise brands come with complex workflows and multi-layered ecosystems, so brand intelligence needs to reach the tools and AI systems where teams already work.

That changes the role of the DAM team too. Success is no longer only about whether a person can find and reuse an asset. It increasingly includes whether another system can select the right content, understand its constraints, and act on it appropriately.

We’ve moved past making a DAM work within a single AI tool, the real shift has been making the DAM an AI-readable intelligence layer between your brand and every single model, tool, team, and agent that acts on it.

AI readiness starts with brand foundations.

Before AI can create real value, you need:

• One clear source of truth for brand assets and knowledge
• Structured, machine-readable brand data
• A taxonomy and metadata model that reflects how your business actually runs
• Defined permissions, rights, and usage rules
• Clear ownership across content and metadata
• Reliable workflows for approvals, legal compliance, and asset lifecycles
• Structured context, so AI can tell the difference between what exists and what is actually right to use

Remember: AI does not fix fragmented brand foundations. It amplifies them. Disorganized assets, disconnected systems, manual processes, and unstructured brand knowledge only become bigger problems once AI scales them up. (This matters even more with agentic AI. If an agent acts on outdated, incomplete, or poorly governed brand knowledge, it automates the wrong decision just as fast as the right one.)

With the foundations in place, testing can start. Start with one workflow and test whether the right context and controls reach the final outcome. Assign owners to the underlying knowledge, define where human review belongs, and measure outcomes and correction effort. The pilot should test the readiness of the connected system, but also the quality of the AI output — ensuring you’re building workflows that can scale.

Our view is that metadata earns its value by improving a decision. It should help someone find the right asset, select a permitted version, route a review, or understand how content is being used. A long list of plausible tags adds little if it does not reflect how the organization actually works.

Useful AI-generated metadata connects three layers:

• Content understanding: What is visible or expressed in the asset, such as its subject, composition, or text.
• Business structure: How the organization categorizes products, campaigns, markets, audiences, and asset types.
• Operational context: Where the asset came from, what it belongs to, which usage restrictions apply, and its approval status.

The distinction between what AI can infer and what it needs to retrieve is essential. A model can suggest a description or classify photography against a defined taxonomy. License terms, ownership, and approval status need to come from authoritative records. A confident prediction should never become permission to use an asset.

Teams should define the fields that matter, provide clear classification guidance, and test the results against real searches and workflows. The measure of success is whether the metadata helps people and machines make better decisions with less correction and uncertainty.

Much more important, but its role is changing.

AI reduces the need for people to manually describe every asset, but it doesn’t remove the need for structured context.

Descriptive metadata can increasingly be inferred automatically — things like visual search, semantic search, facial recognition, and image analysis can happen without manual tagging. But not everything can.

An AI agent cannot look at an image and know whether it is approved, licensed, restricted to a specific region, cleared for a particular channel, or the latest version. That information must come from a structured brand and business context. And as metadata influences more and more actions, an error can travel further and further. That can look like a poor search result, but it also can be the wrong asset in a campaign. So it’s vital that DAM teams are meticulous with definitions, trusted sources, and correction processes. Metadata quality should be judged by the decisions it supports, not simply the number of fields populated.

Agents turn the DAM from a passive repository into an active operator.

Today, dam systems help people find, organize, and distribute content. Agents take on the repetitive operational work around that content: reviewing metadata, organizing libraries, structuring guidelines, checking assets against brand rules, generating templates, and triggering workflows.

The key distinction: agents need a trusted system underneath them.

At Frontify, an agent does not invent brand context. It reasons over structured brand knowledge that already lives in the platform: assets, guidelines, brand essentials, metadata, rights, permissions, and business rules.

That is our natural advantage. Frontify AI capabilities run directly on proprietary brand knowledge, rather than operating as an isolated tool detached from the source of truth.

We are already building this across the platform: our dam agent protects metadata health; our guidelines agent helps teams collaborate with machines to structure brand guidance; and our brand checker evaluates whether assets match brand standards.

The brand assistant expands that vision, connecting to external tools, understands market context, and lets users manage brand operations in natural language.

Agents will not replace the DAM. They are the execution layer built on top of a governed brand intelligence foundation.

We see two broad patterns. Teams beginning to explore AI often want practical value without having to build their own solutions. Frontify’s built-in AI capabilities, including the Brand Assistant and, through early access, the DAM agent and Brand Checker, provide a starting point for finding guidance, improving metadata, and reviewing content against brand standards.

That access can also extend across the organization. By connecting Frontify through MCP to tools such as ChatGPT, Claude and Copilot, teams can ask questions, retrieve brand guidance, and find relevant assets within the environments they already use. This brings trusted brand context closer to everyday decisions.

More fluent teams are starting to build independent tools and agents on top of Frontify, drawing on approved assets, brand guidance, and metadata to support their own ways of working. Frontify remains the system of record, supplying the trusted context that powers those experiences.

As AI begins acting on content, governance needs to follow the work. Organizations need confidence that the rules governing an asset remain effective wherever it is retrieved, adapted, or published.

That starts with controlling both access and action. Permission to view an asset should not automatically grant permission to modify it, publish it, or send it to another AI service. Teams need clear boundaries around:

• Who or what can access content
• Which purposes, markets, audiences, and channels it is approved for
• Whether it can be adapted or used for AI training
• When human approval is required
• Which actions must be recorded and reviewed

To function properly, these rules need to be current, maintained by a clear owner, and expressed in a form connected systems can evaluate. Where a decision depends on confidential information, the system should apply the restriction without exposing details the requester is not authorized to see.

Accountability also requires a record of what happened: which asset and version were used, which rules applied, and what action was taken. Teams need a way to stop a workflow, investigate an unexpected result, and correct changes.

For us, this makes governance a fundamental part of the infrastructure supporting AI. Connections such as MCP can make brand knowledge available across tools, but connectivity alone does not guarantee that every downstream system will enforce its rules. Those controls must be designed and verified across the complete workflow, including how external AI services process the content they receive

The most practical value comes from reducing the effort between having a task and being able to move it forward. For DAM and brand teams, that often means finding suitable content, understanding how to use it, and preparing it for the next person.

Search and discovery are a good example. People can describe what they need in their own language, making existing assets easier to find and reuse. The Brand Assistant and the Frontify MCP addresses a similar problem: someone creating a presentation or campaign can ask for relevant guidance without searching through documentation or waiting for the brand team to respond.

Our automations help teams handle routine work, such as updating asset titles, tags, and workflow status. Teams can add AI actions to generate descriptions, analyze images, or translate metadata as part of those workflows. This reduces repetitive administration and helps prepare assets for their next stage of use.

Our upcoming DAM agent builds on this by helping teams review metadata health and identify gaps or inconsistencies. The aim is to make regular DAM audits and health checks easier to carry out, with less manual inspection and clearer priorities for improvement.

We’re also seeing customers bring brand context and assets directly into their creative tools. One example is using guidelines to inform image-generation prompts. Another is bringing approved assets from Frontify into Figma in the size and format needed for the design. Frontify’s asset repository and media delivery capabilities support that process, reducing the need to download, resize, and prepare files manually.

Across these examples, the measure of value is what becomes easier for the team: fewer searches, fewer routine questions, less repetitive administration, and less rework. Those are the outcomes worth measuring as adoption grows.

Our view is that the biggest barrier is confidence in the complete workflow. Although AI adoption is happening throughout enterprise organization, it’s often fragmented and ad hoc. In order to see holistic adoptions, teams need to trust that AI can support everyday operations reliably before they give it greater responsibility.

That confidence is difficult to build when brand information is patchy, guidelines lack structure, and ownership is unclear. If people still have to check every asset, permission, and output manually, much of the benefit disappears. The work moves faster in one place, but the burden of verification remains elsewhere.

Teams need confidence that AI is drawing on current guidance, selecting approved assets, and operating within the right permissions and usage conditions. They also need to know when it will ask for help and who is accountable for the result.

Building that confidence starts with reliable foundations, connected tools, and clear responsibilities. With those in place, teams can learn where AI earns greater autonomy and where human judgment remains essential.

Yes. Agents are central to our product strategy, and we are developing and delivering agentic experiences across the platform. We also believe the terminology matters. As we explain in our perspective on agent washing, an agent should be able to reason, adapt, and take action toward a goal within defined boundaries. Different tasks benefit from different forms of AI, and we aim to make those distinctions clear.

Our DAM agent focuses on metadata health. Our development work on the Guidelines Agent focuses on helping teams build and organize brand documentation from existing context. Alongside these, capabilities such as the Brand Assistant and Brand Checker help teams access brand knowledge and evaluate content against their standards, with different levels of autonomy suited to the work.

Through Frontify MCP, external agents can also access structured brand knowledge and assets within the scope of the connected user’s permissions. This allows customers to bring brand context into their own tools and workflows.

Frontify provides the governed brand intelligence that makes agents useful, while our native agents increasingly help teams act on that intelligence

AI will shift DAM from a passive storage and findability tool into the trusted orchestration hub – from a system of record to a system of action. The DAM will be the foundational layer for the content ecosystem, ensuring that the right content can be accessed at the right time by the right people, that content is always on brand,

Meaningful AI value depends less on model capability than on organizational foundations being in place first — clean, structured, and governed data. The value derived from AI is only as good as the systems, processes, and data it is modelling from. Defined governance guardrails and human-in-the-loop approval workflows with clear audit trails are still required.

AI makes metadata more important, not less — generative and retrieval systems are only as accurate as the structured context they can draw on, so as content volume grows exponentially with AI-generated variants, the tagging, rights, and taxonomy layer becomes the difference between an AI system that retrieves the right on-brand asset and one that does not. AI can generate metadata at scale, but it’s only valuable when it’s governed, consistent, and mapped to the same structured vocabulary humans and other systems already trust, rather than a flood of auto-generated tags no one validated.

Agents are emerging as the layer that acts on top of DAM rather than just searching it — instead of a person manually finding an asset, checking its rights, and dropping it into a campaign, an agent can be tasked to pull approved, on-brand assets for a specific campaign or channel, localize or resize them, check licensing status, and route the output for approval, effectively automating the repetitive middle of the content supply chain; customers are largely in an early, cautious-experimentation phase. The near-term version of this is “agent drafts, human approves”. Eventually it will be fully autonomous.

As AI begins working with and acting on content, organizations need controls at several layers:

• Machine-readable rights and licensing data (usage restrictions, expiration dates, channel and market limitations) so AI systems can enforce compliance automatically
• Clear provenance and audit trails documenting what’s AI-generated versus human-created
• Accountability audit trails – who approved what, and when (as accountability becomes harder to trace once generation is automated)
• Version and permission controls that prevent an agent from acting on outdated, restricted, or unapproved assets

Meaningful value today is realized in narrow, well-bounded applications where the task is repetitive: AI-powered search and auto-tagging that cuts the time spent hunting for assets, automated metadata generation that fills gaps in legacy libraries, campaign asset localization and resizing across channels and markets, and AI-assisted content variant generation for testing and personalization.

Yes, MediaValet natively supports C2PA. Launching agentic AI in Q4

DAM will become an increasingly important context and control layer for enterprise AI.

AI models are getting very good at understanding what is inside an image, video, document, or audio file. But that is only part of what an enterprise needs. AI also needs to understand whether an asset is approved, what rights apply, where it can be used, what it relates to, and what actions are permitted. That is where DAM becomes more valuable, not less. Its role expands from storing and managing content to providing the trusted business context around that content. This becomes especially important as more AI operates outside the DAM itself. Marketing agents, creative tools, commerce platforms, and enterprise copilots will all need governed access to content. The DAM can provide the context and controls that help those systems use content correctly. AI can understand content. DAM helps AI understand how the business can use it.

AI readiness should start with the use case, not the technology. A company may be ready to use AI for search or transcription long before it is ready to let an agent publish or transform content without review. The more authority AI is given, the more context and governance it needs. At a minimum, teams should understand what content AI can access, whether the metadata and rights information it depends on can be trusted, what permissions apply, and where human review is still required. The goal is not to make every part of the DAM perfect before using AI. It is to make sure AI has enough reliable context to perform a specific task well.

The goal should never be to generate more metadata. It should be to generate metadata that makes content easier to find, govern, measure, or use. Generic AI can describe hundreds of things in an image, but most of that may have little business value. Useful enrichment connects what AI sees to the language the business actually uses, such as products, brands, campaigns, markets, talent, rights, or content types. Confidence thresholds, controlled vocabularies, validation rules, and human review also matter. If an AI generated tag does not improve a decision or an action, it is probably just noise.

More important, but its role changes. AI can increasingly infer what is inside an asset through computer vision, OCR, transcription, embeddings, and multimodal models. That reduces the need for people to manually describe every visible detail. What AI cannot infer as easily is business context. It may recognize a person in an image, but metadata can tell it that the person is approved talent, the image is part of a specific campaign, the usage rights expire next month, and the asset is only approved for certain markets. AI can increasingly determine what an asset is. Metadata helps determine what that asset means to the business and what can be done with it.

Agents are important because software is becoming an active participant in content operations, not just a tool used by people. Today, a person typically finds an asset, evaluates whether it is appropriate, understands the surrounding context, and decides what to do next. Agents will increasingly handle parts of that process themselves. For that to work, agents need more than access to files. They need rights, permissions, approval status, provenance, relationships, and business rules, along with clear limits on what they are allowed to do. Some agents will operate inside the DAM. Others will sit in marketing, commerce, creative, or enterprise AI platforms and call on the DAM when they need trusted content or context. The more important question is becoming less “Does your DAM have agents?” and more “Can your DAM safely support agents that are finding, evaluating, transforming, and activating content?”

Most customers are starting with focused, repeatable tasks where the business rules are already understood and the value is easy to measure. That includes metadata enrichment, natural language discovery, rights checks, content validation, transformations, localization, and identifying what needs attention before content can move forward. The conversation tends to evolve quickly. It starts with “Can AI help with this task?” and moves toward “Under what conditions can we trust it to handle this without someone reviewing every step?” That is where agents become more than a productivity tool and start becoming part of the operating model.

The more AI is allowed to do, the more important governance becomes. AI should operate within the same basic controls that apply to people: permissions, rights, approved sources, business rules, and clear boundaries around what it can change, create, publish, or distribute. Teams also need visibility into what the AI accessed, which model or tool it used, what it changed, and when human review was required. Content provenance will also become increasingly important as generated and transformed content enters the asset lifecycle. The objective should not be maximum autonomy. It should be useful autonomy within known boundaries

The strongest use cases are moving beyond basic tagging into faster discovery and action. Conversational and multimodal search let people ask natural language questions across large content libraries rather than relying on exact metadata or folder structures. Agents can combine that intent with filters, metadata, permissions, rights, and other DAM context to return results that are not only relevant, but appropriate for that user and use case. The next step is action. Customers are using AI to enrich metadata, apply filters, identify rights or compliance issues, prepare content for localization, generate variations, and trigger the appropriate next step based on the context surrounding the asset. That is where the value becomes more significant: AI is no longer just helping people find content. It is helping them determine what they can do with it and move the work forward.

The biggest barrier is confidence. Most companies already have access to powerful AI models. What holds them back is uncertainty around the content those models can access, whether rights and permissions will be respected, how outputs will be reviewed, how data is handled, and whether actions can be audited. The companies that move beyond experimentation will be the ones that make governance, visibility, and clear operating boundaries part of the implementation from the start.

Yes. Orange Logic supports C2PA compliant metadata and content provenance capabilities to help customers retain information about content origin and modification throughout the asset lifecycle.

Yes. Orange Logic supports generative AI for content creation, transformation, enrichment, and other content operations. Our approach is open rather than tied to a single provider. Customers can use different models and services based on the task, their business requirements, and their AI strategy.

Yes. Orange Logic Agent Studio enables customers to build, deploy, and govern agents that can work with enterprise content and its surrounding context, including metadata, rights, permissions, business rules, and content state. Agents can use different approved models and tools, operate individually or together, and remain within defined enterprise controls. Our view is simple: the intelligence comes from the model. The confidence to act comes from the context and governance around the content.

DAM is moving from a place where content is stored to the layer that gives AI context. Enterprise AI is only as good as what it can reach and trust. The DAM already holds the approved assets, the metadata, the rights and the relationships between products, content, campaigns, and markets. That’s the context AI agents need. A DAM like Sharedien built on composable, API-first (MACH) principles that exposes its content through an MCP server becomes something any AI (Copilot, Claude, ChatGPT or a custom agent) can query and act on directly. The DAM stops being one more silo AI has to work around and becomes the trusted content backbone AI runs on. Safe, Governed and scalable. We also need to further prepare ourselves for the fact that 60-70% (maybe even 80%) of the content, moving forward, will be consumed by agents and not human beings. That requires a different approach towards machine readability and general availability.

Speaking on a meta level, the data and data quality need to be in order. Meaning the ones who invested heavily either during COVID or during their digital transformation in clean and good data processes have an advantage here. Those are the companies that are running away with it in the AI age.

Structured, connected metadata. You need a data model that shows how assets relate to products, campaigns, markets and rights, not just flat keyword fields. We believe Sharedien is uniquely positioned to shine here given that we have our open knowledge graph-like entity data model that connects not only assets but also assets to products to campaigns to literally any data entity that a customer has.

Open access. APIs and a standard interface such as MCP, so AI tools can reach content without custom integration projects.

Governance built into the data. Rights, expiry dates and usage restrictions stored as machine-readable fields that an agent can check, not buried in PDFs.

Useful metadata comes from your business vocabulary: your taxonomy, product IDs, brand terms and campaign names. Noise is generic labels like “person, outdoor, blue” that nobody searches for. Useful AI tagging is guided by the organization’s own taxonomy and reference data (for example, retrieval-augmented tagging), comes with confidence scores, and keeps a human in the loop for the fields that matter. The question to ask isn’t “how many tags did AI add?” It’s “did findability and reuse go up?”

More important, though the job changes. Humans will do less manual tagging. But metadata becomes the way AI understands what an asset is, whether it may be used, and how it relates to everything else. In an agentic world, metadata tells the machine what it’s allowed to do. Poor metadata used to mean a bad search result. With agents, it can mean a wrong, off-brand or non-compliant action at scale. That is why we, at Sharedien, don’t believe in standardized AI meta-tagging: we believe it should be based on the customer’s data and that data and its relations should within your Content Hub your DAM. That is also why our AI tagging assistant is Rag-based.

Agents will do the repetitive work around content: intake and quality checks, enrichment, localization variants, rights checks, distribution to channels, and answering “do we have an approved image of X for market Y?” We see two patterns. Some agents work inside the DAM, automating workflows. Increasingly, external agents in tools like Copilot or Claude reach into the DAM through MCP to find, check and use approved content where people already work. We believe this will further separate modern architecture from dated architecture, as an MCP server is the ingredient that can make an external agent really work with a DAM.

Customers: Most are in the exploration and pilot phase. The first question is usually “can our AI assistant find our approved content?” The second is “can it act on it safely?” Customers who connect their DAM via MCP move from idea to working pilot quickly, because they don’t need a new integration project. We also have enterprise customers, large retailers, manufacturing and consumer goods companies who are doing more already and leveraging our graph-like entity-based data model to use their 50+ agents to ask deep questions about the data. As product assets, any type of content, and any type of campaign information are linked together in a smart way

• Agents get the same role-based permissions as the humans they act for. They should have no super-user access yet.
• Rights, licenses and expiry dates are stored as machine-readable fields and enforced at the API or MCP layer, not just shown in the UI.
• Every AI action is logged and auditable.
• Human approval gates for publishing and for anything leaving the organization.
• Clear labeling and provenance for AI-generated or AI-modified assets.
• Data sovereignty: knowing where content is processed and which models see it. This matters a lot to European enterprises.

Sharedien offers proprietary AI services for years. On top of that we allow our enterprise customers to connect their approved LLMs in an open way. We become an operating system for the AI use cases that has the context, the data, and can do any AI use case in a safe and governed way.

You can find a few of the real-world use cases Sharedien has already supported for years:

• Smart, natural-language search that understands intent, not just keywords.
• AI-assisted tagging and enrichment against the customer’s own taxonomy.
• Plus Meta data extraction services, where you can send assets to an llm and get metadata back
• Automated quality and brand-compliance checks at upload.
• Generative image editing for variants, formats and background changes, which cuts agency round-trips.
• AI-generated briefings and summaries that help teams start campaigns faster.
• Connecting the DAM to enterprise AI assistants via MCP, so approved content shows up inside the tools employees already use.

Editor’s note: Sharedien noted a PDF of roughly ten additional customer use-case stories accompanied their response.

The data foundation, not the AI itself. Content is spread across systems, metadata is inconsistent, rights information isn’t machine-readable, and legacy DAMs are closed and can’t expose content to AI. Close behind are governance uncertainty (“what is AI allowed to do with our assets?”) and a lack of internal ownership. The technology is ready. Most content estates aren’t yet. Forget cultural issues because it’s generally not allowed within the organization, or it is not clear what is allowed to use. It’s, beside the data issue, also an issue of not having clear AI guidelines defined.

Yes natively in the platform

Yes. Examples: the Gen AI Image Editor, Smart Search, Smart Briefing, AI Quality Check and AI-assisted (RAG-based) tagging. All of these work on the customer’s governed content.

Our brand new RAG-based tagging assistant is featured in the PDF we have attached.

One that is live already the RAG based Tagging Assistant. Plus Sharedien provides an MCP server, so customers can connect AI agents and assistants (such as Claude, Copilot or custom agents) to their DAM content, metadata and rules, within their existing permissions.

The DAM becomes the governed layer where content is made understandable, trustworthy, and usable before anyone, or anything, acts on it. This is a bigger job than storing finished files. The DAM now has to supply the content, as well as everything that gives it meaning: metadata, relationships, permissions, rights, approval status, and provenance. Without this, AI is guessing.

The mental model we find most useful is that the DAM sits between two AIs. Inside the enterprise, AI tags, enriches, searches, translates, and adapts content. Outside the enterprise, AI-powered discovery interprets and recommends what brands publish. The DAM is the bridge between them. It prepares approved content so it can move safely into websites, commerce, campaigns, and agentic workflows with its context intact, and it does that in both directions.

That also turns the DAM from a one-way library into part of a loop. Content is governed at the source, activated into experiences, measured, and improved. The opportunity isn’t “AI features in the DAM.” It’s the DAM as the trusted content foundation the rest of enterprise AI depends on.

Five fundamentals decide whether AI helps or amplifies issues in the DAM:

1. A governed source of truth. People and systems need to know which asset is current, approved, and authoritative. If there are five copies and no answer, AI will confidently pick the wrong one.
2. A content model AI can read. Metadata, taxonomy, and relationships should describe what an asset is for, its purpose, audience, product, market, channel, not just that it’s a JPEG.
3. Rights and policy as machine-readable controls. Usage rights, region, expiry, consent, and approval state have to travel with the asset, not live in a spreadsheet someone forgets to check.
4. Connected access. AI has to reach the approved source through secure integrations and APIs, rather than working off copied files scattered across repositories.
5. Human accountability. Owners, review thresholds, audit trails, and escalation paths need to exist before an agent is allowed to change or publish anything.

A good readiness test is to take one valuable, repeatable workflow and ask three things: can the system supply the right asset and context, can it keep out what must not be used, and can a human understand and undo whatever happened? If any answer is no, adding AI widens the gap instead of closing it.

More important, and it changes who does the work. Models are good at proposing the descriptions, keywords, entities, transcripts and classifications that people used to write by hand, which takes real manual burden off teams. What that doesn’t do is remove the need for a metadata strategy. If anything it raises the stakes on the schema, taxonomy and governance underneath, because whatever logic you’ve defined is now being applied across your whole library at machine speed rather than one asset at a time.

The line between useful metadata and noise is simple: useful metadata is tied to a decision. It helps someone or something find the right content, understand what it means, apply the correct rights, pick the right market or channel, or connect the asset to a product, claim, or campaign. Noise is generic labelling produced with no reference to a controlled vocabulary or business context, tags for the sake of tags.

So the quality bar should be fitness for use, not tag count. Good metadata is accurate, specific, explainable, and mapped to a taxonomy you own. And the level of human review should follow the risk. A visual description is usually safe to automate. Rights status and regulated claims are not, they need authoritative source data and real approval. The goal was never more metadata. It’s enough trusted context for the next person or machine to act correctly.

Agents earn their place wherever a workflow spans several steps, systems, or decisions. In the DAM, that starts with bounded tasks: natural-language and visual search, enrichment, duplicate detection, metadata updates, rendition creation, content routing. Across content operations, agents can help carry work from brief to asset to review to localization to activation.

The real question isn’t what an agent can do, but what it’s allowed to decide. An agent can prepare work, recommend a next step, or execute something low-risk on its own. But changing rights, approving a regulated claim, or publishing externally should stay behind explicit permissions and human approval. An agent should work inside the DAM’s controls, not around them.

The customers who succeed here start narrow, with clear inputs, an accountable owner, and a measurable outcome, faster findability, fewer duplicate creations, shorter review cycles. Autonomy expands only as the quality of context and controls proves it’s earned. Anyone starting with “let the agent publish” has the sequence backwards.

Governance has to become executable. A policy in a PDF or a rights spreadsheet cannot control an agent. The controls have to live in the content system and be enforced in the workflow itself:

• Identity and least-privilege access for people, services, and agents alike.
• Asset-level permissions, rights profiles, licence terms, region and channel restrictions, consent, expiry, and approval state.
• Provenance and content credentials that trace origin and AI involvement, including C2PA where it applies.
• Human approval gates set by action, audience, and risk, with the strictest controls on public, regulated, and brand-sensitive content.
• Full auditability: what an agent accessed, what it changed, which version or model was involved, who approved it, and where it ended up.
• Clear retention, rollback, and incident-response paths for when something goes wrong.

The principle is that AI operates inside the same governance model as the content, with tighter controls where its speed and scale raise the stakes. Transparency and governance features support compliance readiness, but we’re always clear that no technology guarantees compliance on its own.

The clearest value comes from use cases that improve work teams already do, rather than attempts to redesign the whole content organization at once:

• AI-assisted tagging, enrichment, and natural-language or visual search that get people to the right approved asset faster.
• Transcription, summarization, and language support that make rich media easier to find and adapt.
• Rights-aware retrieval, where restricted or expired assets are excluded or flagged before anyone reuses them.
• Brand-grounded generation, where approved brand context and review are built into how content gets made and varied.
• Content-gap and AI-visibility work that finds where a brand is missing or misrepresented in AI answers, then routes teams back to governed source content to fix what’s published.
• Agent-assisted asset operations: securely searching assets, retrieving details, updating metadata, preparing renditions and links.

What holds organizations back is often the state of the content foundation underneath them; repositories that don’t talk to each other, metadata that’s inconsistent from team to team, rights that live in a spreadsheet instead of the DAM, and ownership no one’s quite sure of. AI doesn’t create the content-debt problem – it just makes it impossible to keep ignoring.

SitecoreAI DAM is built around a governed content foundation for both human and agentic work. Today that covers AI-assisted tagging, enrichment, search and visual discovery; configurable metadata and taxonomies; rights, versions, permissions, approvals and auditability; creative and downstream integrations; and provenance visibility, including reading C2PA metadata already attached to an asset.

We’re extending that foundation in phases. One capability embeds the DAM’s unique asset ID into the C2PA manifest of newly generated renditions, so a downstream asset traces back to the exact approved version it came from, useful the moment a branded image turns up somewhere it shouldn’t. Another signals AI involvement directly in the interface, marking content that was generated or edited with AI. And we’re opening secure access for compatible AI agents to find and work with governed DAM assets through an MCP server.

The piece we’re most focused on is closing the loop between what a brand has approved and what AI actually says about it. The DAM holds the governed foundation. Knowledge Studio turns selected sources into traceable Knowledge Documents, where every fact has an approved source behind it. Scrunch then observes what AI assistants say about the brand in the wild and checks each claim against that approved knowledge, sorting it into one of three buckets: the claim matches what you’ve approved, conflicts with it, or falls in a gap you haven’t answered yet. Each points to a different action, a conflict is a brand-accuracy issue to investigate, a gap is usually knowledge you haven’t created, and a match is reassurance you’re being represented correctly. A team then corrects the source through the approval process they already use and watches whether the public answer changes.

No one controls what an AI model says. What this gives an enterprise is a governed way to improve the answers it can influence, with a dated record of what changed and when.