I've watched brand after brand argue over this while the decision that actually mattered went unmade: which dependencies they could live with, and what those would cost them the day they hit the ceiling.
That was the thread running through our third roundtable in July, a private breakfast for senior marketing leaders at the Ochre restaurant in the National Gallery. No slides or pitches, just an honest conversation about how AI is affecting the way brands think about capability and control.
Three sessions in, the arc has held. Our first roundtable asked who was in control. The second, what breaks when production runs at AI speed. This one asked the structural question underneath both: when AI changes what you can do in-house, how do you rebuild the model, and where do you accept that you are relying on someone else to hold it up?
Ten themes came out of the room. Here are a few that have stayed with me:
Vendor lock-in that builds so quietly you only notice it when leaving has become a project in its own right.
Shadow AI that has moved from tools to outputs, with unapproved assets reaching real audiences before anyone senior has seen them.
The knowledge AI needs to run well, still sitting in people's heads rather than written down anywhere it can be trusted.
We have written all ten themes up, each with a practical response. Most of them remain unsolved across the industry. The question we kept circling back to. Have you mapped your dependencies, or are you waiting to discover them?
If you are a senior marketer running AI-driven production across multiple markets, this is worth a read.