Today’s doodle is the face I made during the talk. I keep hearing the same narratives and not enough about the technical aspects of implementation.
I attended the Henry Stewart talk today by Aprimo on Agentic AI workflows for digital asset management systems. They promoted different kinds of Agentic AIs and shared example ways on how different DAMS could use agents to get through lots of data in less time, with less human labor. For example, a Critic Agent would do quality check review to give feedback on other Agentic AIs content quality; Librarian Agent that auto-create metadata from an uploaded asset or tags subjects.
I’ve always been interested in separating the marketing lingo from system capabilities and functionalities. A lot of very technical subject webinars or demos tend to not have the technical experts give the talk. I see this a lot at conferences or during client meetings when I attend a vendor demo for a new system or tool.
Discussion without technical explanations and framing without history creates confusion and spreads misinformation, especially for new grads or professionals who don’t have the systems knowledge to know what’s sales-talk and what’s contradictory to the history of how systems have evolved in the content management field.
From the talk, I had three core questions:
- How do we differentiate Agentic AI from how databases/DAMS have auto-populated metadata upon file upload or auto-changed file sizes since the 90s?
- What kind of prep-work do DAM managers need to do to prepare data for Agentic AI to be able to use it?
- Would DAM mangers need to enter metadata profiles or schema documentation into the system for Librarian Agents to be effective?
Because the realities in most organizations managing data:
- Data are inconsistent, missing, outdated, and don’t follow the same metadata schemas. Schemas are not typically defined in the system because that requires configuration (which most places never do upon getting a system). Agentic AI isn’t useful if it doesn’t have clean, consistent, structured data.
- Any database system (CMS, DAM, MAM, ILS) has these things: auto-uploads metadata (e.g. title, name, file type); can OCR and pull out abstracts; auto-creates different file types/file format versions per asset; has QC workflows based on rules you configure. The proposed Agentic AIs for these tasks do not offer anything new.
- Copyrights and fair use and provenance research are important parts to DAM labor. Agentic AI like a Compliance Agent who checks legal/brand compliance don’t have the subjectivity to determine fair use of copyright.
Need help comprehending how digital asset management systems works? I’m available for consultancies and freelance data services: https://www.decolfutures.com/services