Jonathan Anderson told a story at Adobe MAX this week that stayed with me. Anderson, now the creative director at Dior, said his mother showed him a wedding video and told him how amazing the bride looked. He had to tell her it wasn’t real.
It got a laugh, but it’s the question sitting underneath almost everything Adobe announced this week in Miami. As more of the work gets done by machines, how does anyone know who or what made it? And when an agent is doing the work inside your tools, who’s actually driving?
Letting other agents in
The biggest announcement wasn’t a feature. Adobe is opening Photoshop, Illustrator, Premiere, and After Effects to outside agents through MCP integrations, previewed this week and due for Creative Cloud customers in early 2027. On stage, models from OpenAI and Anthropic operated Photoshop and Illustrator directly, adapting a poster for different platforms and preparing artwork for screen printing. Because the integrations run locally, the work stays in native files, with layers and sequences still editable by hand.
In August I wrote that Adobe’s ChatGPT plugin made it a layer underneath someone else’s conversation. This goes further. The plugin let assistants call Adobe tools from the outside. MCP lets them work inside the flagship apps themselves. It’s the strongest answer anyone in the creative market has given to the lock-in question, and it tells you Adobe would rather do the work behind your agent than insist you come to Adobe first.
Adobe’s own agent is spreading too. The Adobe Creative Agent now shows up as an AI Assistant, in beta, across Premiere, Illustrator, InDesign, After Effects, and Frame.io. Firefly is becoming what Adobe calls an AI creative studio, with Firefly Design and Firefly Graph in beta and Firefly Builder in private beta. Builder turns an idea into a working prototype of a site or app, which puts Adobe squarely in territory Figma has been working hard to claim. Always-on agents that handle recurring work in the background, like organizing footage from a long shoot, are due in beta early next year.
Underneath all of it is the idea I’ve been calling the Universal Context Layer. Adobe published a piece this week on what it calls the context layer for AI-powered marketing, and its brand intelligence work is about capturing what a brand knows so agents can apply it without being told every time. Box made that argument for files last week. Adobe is making it for brands.
Choice as a position
What came through most clearly was tone. Shantanu Narayen, who hands the CEO role to Anil Chakravarthy on December 1 and stays on as executive chair, used his farewell to talk about choice. Chakravarthy followed by saying some creators want nothing to do with AI, and that Adobe will keep building for them, alongside the people who are excited to use it. That’s an unusual thing to say out loud in the middle of an AI launch, and I think it was Adobe reading the room correctly.
Adobe’s Jason Levine was even more candid at the second keynote. He said Adobe’s line about technology amplifying human creativity is marketing, and that it’s also true. You don’t hear that kind of self-awareness from many stages.
Where the time goes
Adobe’s own research, published this week, says a lot about where the work is heading. In its brand intelligence study, 74% of approvers said they’ve signed off on off-brand content. Leaders estimated that 9% of public-facing content is off-brand, while the people closer to the work put it at 27%. Adobe’s explanation is that review capacity is being outpaced by content demand, and that up to 24% of approvals need a judgment call only a senior creative or brand leader can make.
That’s the Friction Tax in Adobe’s own numbers. When an agent can produce a thousand assets, the making gets cheaper, and the checking doesn’t. The scarce resource becomes the handful of people who know what the brand actually is. Adobe’s answer is to capture that judgment so agents can apply it, and its Firefly Design demo ran an automated brand check across a whole campaign. So, that helps. The question every buyer should ask is who checks the agent’s check.
Provenance, said quietly
Here’s what surprised me, though. Adobe helped start the Content Credentials effort and has the most complete provenance story in this market. Yet I didn’t hear Content Credentials mentioned in the main keynote.
The story was there, just not on stage. The same day, Adobe’s business blog announced that Content Authenticity for Enterprise is rolling out across Creative Cloud and Experience Cloud. Credentials record when and how generative AI was used, automatically, and they’re backed by invisible watermarking and fingerprinting so they survive when metadata gets stripped. Adobe also noted that Google is bringing Content Credentials verification to Gemini, Search and Chrome, and that OpenAI has announced C2PA conformance for generated images.
That matters more than ever three weeks out from December 2, when the EU’s machine-readable marking requirement reaches AI systems that were already on the market. Canva told me on the record last month that its new Craft and Brush models put no marks on what they make. Adobe sits at the other end of that spectrum. I’d have liked to hear it say so from the main stage.
The harder question is one nobody has fully answered yet, Adobe included. Credentials can record that AI was used. When an outside agent works inside Photoshop alongside a person, it’s much harder to show which steps the agent took and which the person did. That’s the authorship question I’ve been asking every vendor this fall, and it gets bigger the moment agents can drive the tools directly.
Anderson had the right frame for all of this. He said people should be the conductor of the technology and bring their own taste to it. Adobe built a lot this week to let agents do more of the playing. What’s still being worked out is how you tell, afterward, who conducted what.
