#031: The AI Foundation No One Talks About

Over the past year, we’ve been working with AI across Creative Force.

Alongside our product efforts, some of the most revealing lessons have come from what we’re building inside Dust, a platform that lets teams create internal AI agents.

We’ve used it to support summarizing Chorus calls, preparing onboarding handovers, pulling insights from CS conversations, and generating internal reports.

It’s not about replacing anyone’s judgment; it’s about helping team members get to the right information faster, without digging through transcripts, tickets, or CRM data on their own.

Most of the time, the agents work well. Responses come quickly. The structure makes sense. Occasionally, they even surface things we might have missed.

But they don’t always get it right. The context can be off, irrelevant details might get flagged, or the output just misses what people actually needed.

It’s rarely an issue with the agents themselves. More often, it’s the foundation: fragmented data, unclear inputs, or small inconsistencies that throw things off.

That’s the pattern I keep coming back to. In my experience, AI doesn’t break at the point of generation. It breaks upstream, when the foundation isn’t ready.

In this issue, I want to look at what that groundwork really involves, and why it rests on three operational pillars.

The Three Pillars of an AI-Ready Studio

In my conversations with studio leaders and through our own work building Dreem, I’ve come to believe that real AI adoption rests on three operational pillars:

1. Data Integrity,

2. Integrations, and

3. Workflows.

Each one solves a different failure point we’ve seen firsthand, and together, they make AI viable at scale.

1. Data Integrity

If the input’s broken, the output will be too.

Even the best models can’t fix bad inputs.

This is the invisible weight of content operations. The AI doesn’t know the product the way your team does; it can only infer from what it’s given.

If a size variant is mislabeled or a color tag is missing, the styling logic quietly breaks. This is the shift from we know what’s wrong when we see it to the system knows it before we do.

Studios don’t need more data. They need trustworthy data: accurate, structured, and actively maintained.

That includes things like clear product metadata, consistent tagging, and reliable links between assets and product records.

And even when the data is clean, AI can only make sense of it if it’s accessible—which brings us to the second pillar.

2. Integrations

Making AI Collaborative, Not Siloed.

If data integrity is about quality, integrations are about access.

Your AI tools need context, like product details from your PIM, imagery from your DAM, and campaign logic from your CMS.

Without real-time connections across these systems, AI becomes just another disconnected utility: powerful in theory, but hard to trust in practice.

When I talk about integrations, I don’t just mean APIs. I mean interoperability: systems that speak the same language, update together, and share meaning.

This is the shift from standalone tools to systems that make decisions together.

Imagine a styling agent that generates outfit suggestions. If it can’t check inventory, it might use out-of-stock items.

Or, if it can’t access seasonal rules or brand restrictions, it might create combinations that don’t align with campaign goals.

So, now we’ve got structured data and systems that talk to each other. But there’s one last ingredient: deciding how AI fits into the actual work your team does every day.

3. Workflows

Turning Automation into Production Logic.

If integrations are about systems talking to each other, workflows are about how people use those systems… and work together.

Still, even when AI is accurate and well-connected, it needs to fit into the way your team works. It needs a human in the loop.

What happens after it generates something? Who reviews it? Where does feedback go? What if it’s wrong?

This is the shift from instinct to structure.

Studios have always relied on tacit knowledge, like stylists who “just know” what’s on-brand, or editors who “feel” when something’s off.

AI can’t replicate that instinct, of course. Nor should it, for that matter. But it can follow rules if you give it rules to follow.

That means defining:

• Who reviews and approves outputs

• When AI steps in, and when it steps back

• How feedback flows into the next generation

• What “good” looks like, and how it's enforced

In early Dreem tests, we didn’t just need styling outputs. We needed a way to approve them, revise them, and feed that feedback back into the system.

Without that loop, the AI got faster, but not better. In the end, it doesn’t fail on its own. It fails when the environment isn’t ready for it.

Structure Before Scale

The most surprising thing about working with AI isn’t what it can do. It’s how easily it breaks when the basics aren’t in place.

In Dreem, we’ve seen this firsthand. Some of the most advanced features we’ve built haven’t run into model limitations.

They’ve run into workflow gaps, missing metadata, or systems that couldn’t talk to each other. Not technical failures. Structural ones.

The lesson is simple: before you scale AI, you have to structure for it. That means clean data. Real integrations. Clear workflows. The unglamorous stuff.

The stuff that doesn’t get applause in a LinkedIn post, but determines whether AI is something your team actually trusts and uses every day.

If there’s one thing I’ve come to believe, it’s this:

AI isn’t the hard part anymore. Readiness is.

And the studios getting this right? They’re not chasing what’s new. They’re investing in what’s necessary.