Adobe Goes AI Agent: What Changes with MCP Support?
機械翻訳 / Machine-translated

機械翻訳 / Machine-translated
@aifriends
AI Friends(https://aifriends.jp)のクロスポスト公式アカウント。AIツールの紹介・使い方・できることを、中学生でもわかるやさしい日本語で届けます。
Wouldn't it be convenient if you could just say "remove the background and make a thumbnail" and Photoshop would do it? On November 10, 2026, Adobe took a major step in that direction. Let's walk through what changes with the new AI agents and MCP support.
At Adobe MAX, held in Miami, Adobe added AI capabilities across its flagship software all at once. According to TechCrunch's coverage, the lineup includes Photoshop, Premiere, Illustrator, and InDesign.
The announcement rests on two pillars. The first is Creative Agent, an AI agent built by Adobe itself. An agent is an AI that receives instructions and carries out multiple tasks on its own.
The second is support for MCP, which allows external AIs to control Adobe software. MCP (Model Context Protocol) is a shared connection standard that links AIs with apps — think of it like a universal USB port.
According to an article covering Adobe's announcements, Creative Agent handles multi-step tasks from plain-language instructions.
Built on this foundation, Firefly AI Assistant is in beta. It includes brand-aligned settings, a "Plan Mode" that shows you a plan before executing, and the ability to choose which AI model to use.
AI assistants are also coming to Illustrator, Premiere, InDesign, After Effects, and Frame.io — also in beta.
According to TechCrunch, the agent is the type that proactively stays active at all times. It tracks the user's workflow and quickly handles small tasks like background removal or video thumbnail creation.
"Always-on agents" for repetitive work like asset organization and quality checks are still at the preview stage.
Consider an e-commerce staffer who processes 100 product images every week. The idea is that the agent handles file naming and image quality checks in the background.
One of the most closely watched announcements is support for local MCP. External AIs like Claude, ChatGPT, and Gemini will be able to control Adobe software running on your local PC.
The supported apps are Photoshop, Illustrator, Premiere, and After Effects. According to AI Weekly's article, layers, vectors, and sequences remain editable. Adobe's Deepa Subramaniam explained that a human can step in and finish the work at any time.
Examples given for Premiere include applying effects to all clips, tracking masked subjects, and adjusting color and exposure across an entire sequence.
That said, MCP is in preview. The private beta is slated to begin in early 2027. This is not something you can use today.
Adobe already has integrations with Claude and ChatGPT. MCP expands on that, opening more features to external AIs.
According to Digital Trends and others, here are the new features by app:
Firefly Builder is in a closed beta. TechCrunch reports it as a competitor to Figma.
For example, a photographer retouching wedding photos might save significant effort recreating group shots where everyone has their eyes open.
Adobe isn't relying solely on its own AI. According to TechCrunch, Firefly supports models from Google, Runway, ElevenLabs, Black Forest Labs, OpenAI, Kling, ByteDance, and others.
Two image models from the acquired company Topaz Labs are also being released: "Wonder 4" for upscaling and noise reduction, and "ImageHDR" for converting standard images to HDR.
In other words, Adobe is positioning itself not just as a potential target for AI disruption, but also as a tool that AIs themselves will use.
The materials reviewed for this article did not include details on availability timing or Japanese-language support in Japan. Pricing was also not addressed in the articles covered.
Even so, the implications are significant. Design agencies and video production companies often perform similar tasks hundreds of times. Delegating that to AI frees people up to focus on planning and final decision-making.
Consider a small regional production company that needs to reformat a single video into vertical, horizontal, and square versions. A realistic workflow would have the agent handle the groundwork while a human finishes the job.
Adobe's Resh Sidhu made a comment to the effect that even if you can generate 1,000 concepts, it's still up to humans to decide what's worth making.
According to AI Weekly, some details remain unsettled. The names of supported external agents have not yet been announced. It's also unclear what happens to Content Credentials — the system that records a work's provenance — when an AI other than Firefly has done the editing.
Additionally, according to IT Brief's article, the local version comes first, with a cloud version to follow at an unspecified date.
On how much of its functionality Adobe will open up externally, Subramaniam said the company is "still thinking through where, when, and how to draw the line." It seems that user feedback will play a large role in shaping those decisions.
It's a shared standard for connecting AIs with apps. When an app supports it, AIs like Claude can call on that app's features.
Local MCP is in preview, with a private beta planned for early 2027. Some features, like Firefly AI Assistant, are already available as beta releases.
According to Adobe, layers and sequences remain editable. The design allows humans to step in and make corrections afterward.
The articles reviewed for this piece did not include pricing information. You'll need to wait for official announcements from Adobe.
Adobe envisions AI handling the groundwork while humans do the finishing. Their stance is that judgment and creative sensibility remain human work.
For now, try out the already-available Firefly AI Assistant and existing integrations, and start thinking about which parts of your workflow you could hand off.
This article is a cross-post from AI Friends.