AI changed the economics of the mechanism-of-action work I do, and I say that as someone who spends most of his week in the edit. Work that once meant a long, expensive 3D animation pipeline I can now draft, revise and localize in a fraction of the time. That speed matters most where iteration used to be the enemy: briefing investigators, explaining a therapy to a room of stakeholders, or producing the same explainer in six languages without six full builds. For a lot of early-stage and internal work, being able to iterate cheaply is the whole point.
The help is real, but it is bounded, and I have learned exactly where the boundary sits. AI is excellent at accelerating a mechanism I already understand and can specify clearly. It becomes unreliable the moment scientific judgment is required: deciding which step of a pathway to foreground, how far I can simplify before I become wrong, or where a visual metaphor quietly implies a causation the data does not support. Those calls are what make an explainer trustworthy, and they still belong to people who understand the biology.
The failure mode is subtle and it is always the same. An AI-generated film can look convincing while being scientifically off in a way a lay viewer will never catch but a KOL will spot in seconds. A receptor that behaves wrongly, or a mechanism shown as settled when it is still contested. In front of the exact audience the film is meant to impress, that is not a small blemish. On more than one project I have had to slow a team down and send a sequence back to the science before it went anywhere near a congress screen.
So I keep AI in the pipeline and out of the driver's seat. Used well, it clears the grunt work and buys my team more room to argue about the hard, human decisions, which is where the quality comes from. As a shortcut around scientific review, though, it produces something cheap and wrong in front of the people who matter most. I reach for it where it earns its place and hold the line where accuracy is the entire deliverable.
Working with Medical Affairs teams has only sharpened that instinct. In this field a beautiful film that is subtly wrong is worse than no film at all, because credibility is the one thing you cannot reshoot. My job is to use every tool that makes the work faster without ever letting the science slip, and holding that balance, not the technology itself, is what I am really being trusted to do.