Adobe turns its AI agent loose on Photoshop, Premiere and Firefly

Adobe has used its MAX event in Miami to push artificial intelligence deeper into nearly every product it sells, including Photoshop, Premiere, Illustrator, InDesign and Firefly. The most consequential announcement is a Model Context Protocol (MCP) release that lets external chatbots such as ChatGPT, Claude and Gemini call features inside Adobe's Creative Cloud applications.

The move matters because it turns Adobe's software from a set of applications into a set of services that other companies' assistants can drive. Adobe already offered integrations with ChatGPT and Claude for a limited set of product functions. The new local MCP exposes a far wider range of endpoints, so third-party agents could in principle edit an image, adjust a video or build a layout on the user's behalf.

Adobe has not decided where to stop. Deepa Subramaniam told TechCrunch: "We are still figuring out where and when and how to draw the line. What we're putting out there is a number of endpoints and capabilities across Illustrator, Photoshop, After Effects, and Premiere. We're just going to see how people use it, what they're trying to do, and keep showing more."

She added that Adobe has no intention of narrowing the surface. "We don't have any need or desire to sort of limit what the MCP endpoint can expose. What we really want to do is to support that creative workflow, how you choose to do it," said Subramaniam.

What did Adobe add to Firefly?

Firefly, Adobe's own creative hub, received the largest share of new tools. Builder, a prototyping tool now in private beta, lets users assemble designs that present ideas — a direct move onto Figma's home ground. Firefly also gained a tool for producing on-brand graphs and an image editing capability that changes one part of a picture while leaving other objects untouched.

A new AI video tool in Firefly assembles a finished video from clips, still images and sound effects. Adobe is also making its agent proactive and always on, so it gathers more context about the user's work and remains available for any action. Quick actions such as background removal or generating a video thumbnail are reduced to a single step.

Which features reach Photoshop and Premiere?

Photoshop gains a relighting feature and sliders for prompt-based generative styles. Premiere receives a new color grading system aimed at professional editors plus an object removal tool for video. Lightroom can now correct blinking eyes and remove blemishes. Adobe framed these as a continuation of the AI features it has already shipped across Photoshop, Express, Premiere and Illustrator.

On the model side, Adobe already hosts third-party models from Google, Runway, ElevenLabs, Black Forest Labs, OpenAI, Kling and ByteDance. Its recent acquisition Topaz Labs is contributing two image models: Wonder 4, for upscaling and denoising, and ImageHDR, which converts standard images into HDR.

Why does the MCP bet carry risk?

Opening Creative Cloud to outside assistants is a strategic wager, and not everyone in the industry shares Adobe's enthusiasm. Last month, Figma co-founder Dylan Field said at an event in India that MCP is not a perfect standard and benefits frontier labs more than it benefits toolmakers.

That criticism cuts at the heart of Adobe's plan. If chatbots become the primary interface for creative work, the application that owns the user relationship today may find itself reduced to a supplier of endpoints for someone else's assistant. Adobe is betting the opposite: that exposing its capabilities widely will keep its tools at the center of creative workflows, whichever interface the user chooses.

The company says user feedback will determine how much of Creative Cloud the MCP eventually exposes, which suggests the scope of today's release is a starting point rather than a finished boundary.