From vault to vision: What SXSW reminded me about the work that still matters

Digitization is the beginning, not the end.

What Violeta shared about Paramount's archive work illustrated something I've believed for a long time: digitizing an asset and making it usable are two different things. The film goes into the cloud. Now what?

The answer Paramount built was an AI-powered search tool that uses vector embeddings to make millions of minutes of footage semantically searchable. Not searchable by keyword, by metadata tag, or by whatever someone remembered to type into a catalog field years ago. Searchable by meaning. What made that possible was a chain of partnerships: Iron Mountain handled the digitization, AWS provided the cloud storage, and Paramount built the search capability on top of that foundation. It's a good illustration of how this kind of work actually gets done — no single organization does it alone.

The practical result was significant. A 240-minute compilation project that previously took a team 32 hours to complete was done in eight. That's not a marginal improvement. That's the difference between a task that bottlenecks a workflow and one that doesn't. The creative teams embraced it so quickly that Paramount eventually integrated the tool directly into Adobe Premiere and Photoshop, so users never had to leave their creative session to find what they needed.

That's the shift I find most exciting, and most important to understand clearly. The archive stops being a place where things are stored and becomes a resource that creators actually reach for — rediscovering material and finding creative ideas inspired by content they didn't even know they had.

AI is a toolkit, not a replacement for humans.

One of the most important things Carin said during the panel, and I want to repeat it because it gets lost in a lot of AI conversations, is that none of this replaces creators. It extends what creators can do.

The Sunset Boulevard restoration is a good example. We went back to that film because the best surviving audio element had accumulated significant hiss — a result of both the element's age and being multiple generations away from the original recording. We'd done our best with a previous restoration, but we knew the audio still wasn't right. When I saw what Peter Jackson's company, Park Road Post, had done with the audio separation work on Get Back, separating previously inaudible voices from ambient studio noise, I called them.

What came back was not just a cleaned-up track. They separated the audio into discrete stems: music, dialogue, and effects. And critically, they didn't discard the noise. They delivered the noise floor as its own separate track, because there might be sonic details within it that the final mixer would want to preserve as part of the film's original character. That's an extraordinarily thoughtful approach, and it won a major technology award in 2025. But it was driven by human judgment at every step. The AI did the separation. People made the decisions about what that separation should accomplish and why.

That's the right model. Use AI for the tasks where it creates massive efficiency gains — and to do things that couldn't be done before: searching, cataloging, metadata enrichment, audio separation, OCR on handwritten labels. Reserve human expertise for the decisions that require it.

The archive is now at the center of the AI conversation.

What to do with this

If you're responsible for a physical or media archive of any size, here are the questions worth asking right now:

What do you actually have, and what condition is it in? Not a rough estimate. A real inventory, ideally conducted before the material has passed the point of intervention.

Is your digitized content findable by the people who need to use it? Having digital files in a system that only a few specialists know how to navigate is not meaningfully different from having the material on a shelf.

Are your metadata and your physical and digital asset records integrated in a single view? The absence of that unified view is one of the most consistent problems we see, and one of the most solvable. In fact, Iron Mountain has a great solution for this called Iron Mountain InSight® DXP - so if you’re curious about how we can help, please reach out.

And finally:

It's time to listen to the archivist in your organization when they say that time is running out.

The conversation we had at SXSW was energizing, and the technology is genuinely remarkable. But the thing that stays with me from years of doing this work is simpler than the technology. Every film deserves its opening night again. The work we do, the painstaking, labor-intensive, sometimes unglamorous work of preservation and digitization and cataloging, is what makes that possible. And an archive always exceeds its original intent. With the tools that now allow for new discovery, new creativity, and new insights, AI isn't just helping us work faster and better — it's revealing value that was always there, waiting to be found. Urgency isn't strong enough of a word.

Andrea Kalas is VP, Media & Archival Services at Iron Mountain. She has spent her career at the intersection of film preservation, archive technology, and the moving image industry, including earlier roles at Paramount Pictures, the British Film Institute, UCLA Film & Television Archive, and DreamWorks.