Adobe MAX 2026 shows Firefly turning into an AI studio that works on its own

- Reading
- 6 min
- Tables
- 1
Linha fina atual mantida.
At Adobe MAX 2026, the company launched a campaign about human authorship at the same event where it turned Firefly into an agent capable of planning, choosing models and executing tasks on its own.
At Adobe MAX 2026, the company did two things that seem contradictory but aren't. On one hand, it launched a brand platform called Made to Create, with a film produced in partnership with agency 72andSunny bringing together more than 30 artists from 18 countries to make the case that taste and decision-making remain human. On the other, it turned Firefly into an AI studio with agents that plan, choose models and keep working after the person steps away from the screen.
The tension is deliberate. Adobe knows people who make a living from creative work fear becoming machine supervisors, so it uses the campaign to draw a line: execution can be delegated, authorship cannot. But the technical announcement that came alongside it is what really matters to people who design products, because that's where the company decides, in practice, where that line sits.
Made to Create: the yardstick Adobe set for itself
The film moves through illustration, fashion, animation and film, featuring names such as Ukrainian director Tanu Muiño, illustrator Space Dawg, digital artists YoveLourself, Malik Afegbua and Edita Žirgulienė, and Colombian director Santiago Carrasquilla, whose studio Art Camp also co-produced it. One of the pieces reinterprets the word "Create" in dozens of different styles and techniques, a visual attempt to show what the company considers hard to automate.
AI puts creation within reach of more people; for others, it makes this moment uncertain.
Lara Balazs, CMO at Adobe
The campaign will continue over the coming months, with an extension planned for the NFL's Thanksgiving broadcast in the United States. From a brand standpoint, it's a well-calculated defensive move: Adobe needs to convince its own creative audience that selling an AI tool isn't the same as replacing the people who use it, precisely at the event where it announced the most autonomous tool in its history.
Firefly got a project manager that never sleeps
The new Firefly AI Assistant, still in beta, changes what it means to "ask" the tool for something. Instead of generating an image from a prompt and stopping there, the user describes a goal in natural language and the assistant builds a multi-step workflow to get there, deciding on its own which Firefly tools to use at each step.
According to a report by B9, a Brazilian tech outlet, about the event, there are two operating modes, and the choice between them is the most interesting design decision in the announcement:
- Plan Mode: the agent shows the plan before executing, and the person reviews, edits or approves each step.
- Automatic Mode: the agent simply executes the task from start to finish, with no checkpoints.
This duality is, in practice, an interface pattern for building trust in an autonomous system: you can audit before running it, or you can trust it and let it run. Anyone designing a product with generative AI inside will run into this same choice, and Plan Mode is a concrete example of how to turn a "black box that decides on its own" into something the user still feels they control.
According to B9, Adobe is also testing always-on agents, which take on recurring tasks such as organizing incoming material, checking the quality of assets or preparing content for different channels, continuing to work after the person closes the program. This is the point where a "creative tool" starts having its own initiative over how to reach the result, not just speed of execution.
From in-house model factory to third-party model hub
The most relevant strategic shift for anyone thinking about product, not just assets, is a different one: Firefly stopped being synonymous with Adobe's own model. According to B9's report, the platform now brings together, depending on the task, models from Google, OpenAI, ByteDance, Kling, ElevenLabs, Runway and Topaz Labs, automatically selected by the Creative Agent or manually chosen by the user.
This reveals where Adobe decided to compete. In a market where a new image, video or audio model shows up almost every week, trying to be the best generator in every category is a losing race. The bet is to be the place where those models meet files, version history and an already-established professional workflow, instead of forcing creators to migrate to a new app with every release.
For anyone deciding on a product stack, this is the exportable lesson: competitive advantage can lie in orchestration, not generation. Becoming a trusted hub for third-party models, instead of owning the best model, is a product strategy just as valid as the other, and much cheaper to sustain in the long run.
From storyboard to final cut, in a suite that became five products
The expansion also moves Firefly closer to being a complete creative suite, with modules that tackle different stages of the workflow. The descriptions below follow what Adobe presented at the event, as reported by B9:
| Module | What it solves |
|---|---|
| Firefly Video | Builds a story, organizes scenes on a timeline, generates transitions, captions and supporting images |
| Firefly Design (beta) | Brings together illustration, imagery, animation, 3D and print materials in a single workspace |
| Firefly Image | Edits objects, text, lighting and targeted retouching without redoing the whole image |
| Firefly Graph | Creates repeatable workflows, useful for applying brand guidelines across dozens of campaign assets |
| Firefly Builder (private beta) | Generates prototypes of websites, apps and presentations from natural language, then editable visually |
Firefly Builder is the one that matters most to people working on digital products, not just campaign assets: it blurs the line between a design tool and a functional prototyping tool, territory currently contested by Figma, Framer and a wave of AI app builders. If Adobe delivers this in a stable way, product teams get one more path from idea to clickable prototype without opening a separate tool.
What remains open
None of this solves the most annoying problem of autonomy: who answers for it when the agent gets it wrong. Delegating model choice and background execution means accepting fewer human checkpoints, and Adobe's materials don't detail how licensing, image rights and brand liability work when the output comes from a third-party model automatically chosen by the Creative Agent.
For agencies and product teams, this means something practical: before turning on automatic mode for a client project, it's worth testing first in Plan Mode, reviewing each step, until you understand what kind of asset the agent delivers well and where it still slips into generic territory. The autonomy Adobe sold at MAX 2026 is real, but the work of supervising hasn't disappeared: it has only changed shape.
Translated from the Brazilian Portuguese original · Read the original
