Someone put AI in the plan. The slide is tidy. The budget line has a number. The vendor list is long. What it should actually make is still an open tab.
That is how a lot of the last two years have been spent: buying a tool, then hunting for a problem that makes the purchase look wise. The output arrives quickly. The idea does not.
Speed is not the interesting part
Generative tools are good at filling a calendar. Variations, crops, captions, animatics, first drafts, the tenth layout nobody needed. If the brief was already a category default, the machine will give you a more efficient default. You will look busy. The brand will not have moved.
The interesting question is not “can we generate this”. It is “what should exist that we could not afford to make by hand, and still has to meet the old standard”. New tools. Old standards. That order matters. Reverse it and you get novelty with no memory.
Put it in the pipeline, not on the pedestal
We use AI the way a studio uses a better lamp. It should help you see the work, not become the work. A production pipeline that can explore, reject, and finish faster is useful. A pitch that treats the model as the idea is a stall with extra slides.
The useful outputs are specific. A generative pipeline that keeps art direction consistent across a hundred assets. Synthetic media that you can actually put in front of people. A prototype you can hold, tap, or walk around, instead of a deck about the future. If a person cannot use it on Tuesday, it is still a conversation about tools.
One node off-model
Pattern-finding is what these systems are for. The job is to notice the node that does not belong — the option the model would not have suggested, the constraint that makes the rest of the set make sense. That is still a human decision. If you cannot tell which output is the departure, you have not directed anything. You have sampled.
If AI is in the budget and nobody has agreed what it should produce, bring us the budget line. We will start with what it is for, then build the pipeline that can make that, to a standard that would have been embarrassing to miss before the tool existed.