After Effects now has its own AI Assistant, in public beta since Adobe’s September 8 announcement. It’s an agent-style helper that takes plain-language instructions and carries them out inside your project, and it’s aimed as much at veteran motion designers as at newcomers.
What changed
Adobe says the assistant takes on “tedious, technical and multi-step tasks.” Across Adobe’s post and the coverage from CG Channel and CineD, the listed jobs include reorganizing large projects, repairing technical issues, writing expressions without you coding them, inspecting compositions and footage, summarizing a composition a collaborator handed you, and generating images or video with Firefly.
The key detail is scope: CineD reports that it works across your entire project, not just the composition you have open. You can also choose whether it asks permission before acting or goes ahead on its own.
It isn’t in the regular release. After Effects 26.5, the September release, doesn’t include it. You need the separate After Effects (Beta) app from the Creative Cloud desktop app. It follows similar assistants that came to Photoshop and Premiere Pro in June, according to CG Channel.
Why it matters for designers
Expressions are the classic wall in After Effects. Plenty of motion designers know exactly what they want (wiggle this, but only on the x axis, and ease it out over the last second) and still lose twenty minutes to syntax. Describing the behavior and getting working code is a real time saver, and you can read what it wrote to learn from it.
The project-wide scope matters too. Inheriting someone else’s file with vague layer names and nested precomps is one of the most tedious parts of motion work. An assistant that can survey the whole thing and tidy it up is useful before you’ve even started animating.
What to watch out for
This is beta software and Adobe says as much. CineD notes that performance and reliability vary depending on the task. Don’t run it on the only copy of a delivery file, and start with the “ask permission” mode so you can see each step before it happens.
Read expressions before trusting them. Code that looks right can still break when a layer is renamed or a comp is duplicated, and the assistant won’t always know how you plan to reuse a rig.
Cost is an open question. The sources we checked don’t state how generations are limited in the beta or what pricing will look like at full release, so don’t build a client workflow around it yet.
