At its MAX conference in Miami, running November 10–12, Adobe unveiled a sweeping AI offensive that cuts in two directions at once: deeper automation inside its flagship creative applications, and a deliberate opening of those applications to AI agents built by other companies. The centerpiece is a local MCP server that exposes capabilities of Photoshop, Illustrator, Premiere and After Effects to external models such as ChatGPT, Claude and Gemini.
- External agents get native access: A local MCP integration lets third-party agents work on native project files, preserving layers, vectors and sequences as editable elements.
- Firefly goes proactive: Adobe’s creative agent becomes always-on, gathering context and executing quick actions such as removing a background or generating a video thumbnail.
- New editing muscle: Photoshop gains relighting and generative style sliders, Premiere adds professional color grading and object removal in video, and Lightroom introduces blink correction and blemish removal.
An MCP bridge into the professional toolchain
The Model Context Protocol is the common connector that allows an AI agent to talk to an application. Adobe’s implementation runs locally, meaning an external agent can read a brief, make changes inside Photoshop or Illustrator, and hand back a result without flattening the project into a static image. According to Adobe, layers, vectors and sequences remain editable after the agent finishes, so a professional can review or correct the work afterward. The company demonstrated examples including Claude preparing an illustration for screen printing and ChatGPT generating variations of a poster for several cities.
The feature will arrive in private beta in early 2027, so it will not be available to all users from day one. Adobe is also positioning its own assistant, Adobe Creative Agent, as a creative collaborator that accepts instructions in natural language. It is now in beta for Premiere, Illustrator, InDesign, After Effects and Frame.io, and can handle multi-step tasks such as applying effects across an entire sequence or adjusting color and exposure on multiple clips. Two controls keep the user in charge: Plan Mode, which lets you review, edit and approve a proposed plan before execution, and a Model Selector for choosing the right model per task.
“We are ushering in a new era of creativity where creative professionals and makers of all experience levels have the tools, models, and agents they need to bring their most ambitious ideas to life.”
That statement came from Adobe president and future CEO Anil Chakravarthy, who added that the goal is work reflecting each creator’s unique taste, style and vision.
Firefly evolves from generator to studio
Firefly is no longer just an image generator. It now blends Adobe’s own models with third-party engines from ByteDance, Google, Kling, OpenAI and ElevenLabs, aiming to spare creators from hopping between applications to pick the best tool for an image, video or voiceover. New additions include Firefly Video for moving from story structure to final cut, Firefly Design in beta for combining illustration, animation, 3D and print on a single canvas, and Firefly Image with precision editing for changing one part of a frame without affecting the rest. Quick Actions, Firefly Graph in beta and Firefly Builder in private beta round out the set, the last aimed at prototyping without code.
For marketing teams, Adobe is integrating CX Enterprise Coworker into GenStudio, its content supply chain platform. The collaborator lets teams manage campaign planning, creation and activation through a conversational interface while coordinating the necessary agents and workflows. The package also includes Adobe Brand Intelligence and Intelligent Reflow, which adapt designs to new sizes and layouts and will reach Photoshop, Illustrator and InDesign. Workfront gains Ideation Space to turn ideas and data into a structured brief, while Firefly 3D Digital Twin creates product replicas so labels or materials can be changed without repeating a photoshoot.
The strategic gamble behind the open door
Adobe acknowledges it is moving forward without fully settled boundaries. Deepa Subramaniam, a company vice president, told reporters that “we’re still figuring out where, when and how to draw the line,” adding that Adobe has “no need or desire to limit what the MCP endpoint can expose.” Not everyone shares that enthusiasm: Figma co-founder Dylan Field said in October that MCP is not a perfect standard and mainly benefits the large AI labs.
The MCP server is arguably the real story here, and it carries a calculated risk. Opening Photoshop, Illustrator, Premiere and After Effects to ChatGPT, Claude or Gemini acknowledges an uncomfortable truth: more people will want to edit by talking to a chatbot rather than by launching a desktop application. Adobe would rather be the tool those agents operate than be left out of the conversation, and that is a smart defensive play. Yet the same openness blurs the product’s identity. If value shifts to the agent issuing orders, Adobe risks becoming invisible plumbing beneath an experience others build. A vice president admitting the line is not yet drawn is not a slip; it is the honest confession of a company that chose to jump before it had the map. The logic holds, but the leap comes with vertigo.
Sources & References