Adobe plans to let third-party AI agents control functions in its Creative Cloud applications through the Model Context Protocol (MCP), starting with local endpoints and adding cloud access later.
Users would be able to work with their preferred agent, while Adobe would decide which functions, models and algorithms to make accessible through the endpoints.
The plans sit alongside the Adobe Creative Agent, available through AI assistants within Adobe applications, and connectors that bring some of its tools into external AI services. Users can move between conversational interfaces and direct application controls.
Agent access
"At the end of the day, the simplest way to get exposed to what we're doing with our agentic strategy is to use the Adobe Creative Agent in the applications. That has all the reasoning layer, the intelligence layer to understand your project, your format, what the application you're in can do, and increasingly has context of other applications, so you can just get the job done," said Deepa Subramaniam, vice president of product marketing for creative professionals, Adobe.
Subramaniam identified Claude, ChatGPT and Gemini as services where Adobe's connector tools let users begin creative tasks before moving into its applications. Presentations showed the company's own agent working within Firefly, Premiere, Photoshop and Illustrator.
With planned MCP access, an external agent could operate Adobe software alongside other applications on the user's desktop. Subramaniam cited rotoscoping and image hole filling as examples of the capabilities Adobe could expose. She did not provide a complete list of the endpoints it plans to make available.
Cloud MCP support will follow local access, with Firefly among the services planned for that later stage. Subramaniam did not give a timetable for that stage of the rollout.
Firefly workflows
"Are you creating by hands-on driving the application? Are you creating through a conversational interface, enabled by an agent? Are you creating by using your agent of choice driving our applications, or a hybrid of all of those approaches?" said Subramaniam.
Within Creative Cloud, Firefly supports content generation, concept development and workflows connected to Photoshop, Premiere, After Effects and Illustrator. Professionals use Firefly Boards to develop ideas before taking them into their usual production application, according to Subramaniam.
One demonstration involved a Photoshop user adding motion to artwork. The user described the movement in Firefly, generated a video animation and overlaid it on the work, instead of building the animation in After Effects.
Graph allows users to create and share workflows, including packaging them as capsules for colleagues to reuse. Firefly also offers simpler editing and prompt-based creation tools for people with less experience of Adobe's established professional applications.
Video is a focus across the range. Adobe's presentations included storyboarding and a social media video created in Firefly, alongside Premiere's timeline-based editing. Subramaniam described demand from professional editors, content creators and people exploring video as a medium.
Specialists also face pressure to broaden their skills. Graphic designers are being asked to add motion, while video editors face demands involving colour grading and sound. Adobe is using AI assistance to help users work in unfamiliar applications and formats.
Production work
"Here's the cool thing, though: we're testing them out against real projects, real deliverables. You can have a great product demo that kind of sells a feature, but we actually put that feature through the test. Real deadlines, real deliverables," said Resh Sidhu, vice president of creative, Adobe.
Sidhu leads Adobe's in-house creative team, which receives early versions of tools and features from the product team. Its feedback informs development alongside responses from the wider user community and public beta programmes.
Subramaniam cited Generative Expand as an example of a workflow emerging from public testing. People using Generative Fill in Photoshop's beta were extending photographs and other images, prompting Adobe to develop a dedicated expansion workflow for a subsequent beta.
Feedback also guides smaller enhancements, fixes and refinements. Adobe releases features when they are ready, incorporating user responses through beta and generally available releases.
Adobe's creative team is building repeatable processes and agents for production tasks. The production demands include creating different social media aspect ratios and producing material across languages. The team still has to meet production deadlines and work within finite budgets.
AI tools are also changing who can contribute early in a project. Sidhu described writers being able to visualise ideas and producers developing prototypes and concepts earlier in the process. Teams can examine proposals before committing to production, identify weaker ideas and develop alternatives.
Adobe's team is incorporating AI fluency into its understanding of creative fluency. Sidhu acknowledged that some artists and other creative professionals may choose not to use AI. Within agencies and creative departments that adopt it, teams are deciding which parts of a workflow to automate and where to apply specialist human skills.
Creative judgement
"I think craft, judgement, taste is fundamentally the thing that is going to make your brand and your team stand out. It's absolutely the question and the conversation, and we see it within our own team," said Sidhu.
"You can create 1,000 options of something; you still need someone to decide that's the thing that's worth making, and that's creative strategy," added Sidhu.