Adobe has integrated AI agents into the entire Creative Cloud

On June 18, Adobe officially announced a substantial update to the Firefly platform — not just adding new features to Firefly itself, but embedding AI agents directly into Photoshop, Premiere Pro, Illustrator, InDesign, and even Frame.io. Starting today, all these apps enter public beta, and you’ll see a new dialog box in the sidebar when you open them.

The symbolic significance of this move is greater than the features themselves. Over the past three years, Adobe’s positioning of Firefly has constantly shifted — from its initial debut as a text-to-image tool aimed at competing with Midjourney, to being packaged as an “all-in-one creative AI studio” at Adobe MAX 2025 last year, to the launch of the unified Firefly AI Assistant this April. Today’s update effectively rolls out that “conversational creation” approach right onto every Creative Cloud user’s desktop.

Generate a Logo, Color Scheme, and Marketing Video in One Sentence

First, let’s talk about updates to Firefly itself. The core new feature this time is called Creative Skills. The logic is simple: users describe what they want in natural language, and Firefly’s backend coordinates model chains to get the job done.

It might sound like the same story every AI platform has been telling over the past year, but Adobe has added some noteworthy details:

• It will proactively ask follow-up questions. If your prompt is vague, Firefly won’t just force out an “okay-ish” result — instead, it will ask you questions: Who’s the target audience? Where will it be used? Is there a defined brand color palette? The experience feels more like talking to a junior designer than tweaking a generative model.
• Dedicated control panels instead of pure prompts. After generation, a task-specific refinement panel appears on the right-hand side of the interface — font, camera movement, brand colors — you can drag or select directly.
• Skills are orchestratable. Logo generation, brand colors, and marketing videos aren’t isolated features — they can be chained together. If you tell it “Create a complete visual identity for a coffee brand,” it will first propose logo options, then derive the color and font system, and finally produce social media asset templates.

This “conversation + panel” hybrid form is more approachable for non-technical creatives than prompt-only Midjourney or node-based ComfyUI. Adobe serves tens of millions of paying subscribers, many of whom are traditional designers unwilling to learn prompt engineering. For them, this hits the right accessibility level.

Photoshop, Premiere, and Others Get Their Own “Specialist Doctor”

What can truly change daily workflows is the AI Assistant inside each Creative Cloud app.

Adobe’s approach here is not to make one all-purpose helper, but to give each app a specially tuned agent. Photoshop’s AI Assistant understands layers, masks, and selections; Premiere’s understands timelines, transitions, and color grading; Illustrator’s is focused on vector paths and print specifications. Under the hood, it’s the same “conversational creative agent,” but with vertical fine-tuning for each app. Adobe officially describes it as “a specialist within each Creative Cloud app.”

This is actually practical. An agent that can work in Photoshop requires completely different tools and state-space understanding compared to one working in Premiere. Forcing a unified copilot risks making it average at everything, or relying on an overly long system prompt — which would slow responses and make them unreliable.

Tasks it can actually perform include:

• Asset organization. Drop in a pile of unsorted materials, and AI Assistant can auto-classify them by content, date, or color tone, and even rename files according to preset rules.
• Batch operations. For example, “Cut out all 50 product photos, replace the background with white, and export as 2000x2000 JPG” — a task that once required scripting is now done in one sentence.
• Design iteration. In Illustrator, you can say “Make this logo’s corners smaller, shift to a warmer color tone, and create three variants,” and the AI will alter the original path structure, not regenerate an uneditable bitmap lookalike.
• Rough video edits. In Premiere, you can have the Assistant detect dead air, verbal slips, or long pauses in interview footage, then auto-mark them or even produce a rough-cut timeline.

Redesigned Firefly Studio: Now It Remembers What You Did

Beyond in-app Assistants, Adobe is privately testing a “reimagined” Firefly AI Studio. Its core pitch centers on three terms: persistent context, reusable assets, organized workflows.

In plain language: it finally remembers what you’ve done before.

Previously, using Firefly meant each new session was like having amnesia — you had to re-feed the character design, brand colors, and typography settings from scratch. In the new Studio, you can name characters, objects, and backgrounds, then reuse them directly: “Use the mascot from last time, but change its pose,” and the model maintains consistency.

For creators developing IP or branded assets, this update is more practical than a new model. Consistency has been one of the toughest challenges in generative tools for the past two years; Adobe isn’t brute-forcing it through the model itself, but working around it with engineering — saving and referencing assets. If this approach works, it’s a big deal.

30+ Models on the Table — No More Holding Back

A major change behind this upgrade: Adobe has fully abandoned the obsession with “only using our own models.”

The Firefly platform now includes over 30 third-party models, such as:

• Google’s Nano Banana 2, Veo 3.1
• Black Forest Labs’ FLUX.2
• Runway Gen-4.5
• Kling 3.0 and Kling 3.0 Omni
• ElevenLabs Multilingual v2
• OpenAI’s GPT Image
• Luma Ray3, Topaz Labs, and more

Photoshop’s Generative Fill now lets you switch the underlying model — if you feel Firefly’s own photorealistic style falls short, want FLUX’s detail level, or Nano Banana 2’s text rendering ability, you can toggle with a click.

This is somewhat counterintuitive — Adobe has its own model teams, so why channel competitors? They’ve realized that the core moat in creative tools isn’t the model itself but the workflow. Aggregating the best models into their toolchain provides more value to users than trying to inch towards SOTA with an in-house model.

This thinking mirrors what many AI aggregation platforms do — developers use OpenAI Hub-like marketplaces to call GPT, Claude, Gemini, DeepSeek, choosing whichever is strongest. Adobe is essentially bringing this “model aggregation” approach into creative tools.

Enterprise: Firefly Foundry and Brand Consistency

On the enterprise side, Adobe has been active too. The GenStudio platform just launched two new offerings:

Firefly Foundry enables enterprises to train their own AI models on private data, ensuring output is copyright-safe and aligned to brand standards. This is critical for industries like FMCG, luxury, and automotive, where visual consistency is paramount — their reluctance to use general models has been more about legal risks than technical ones.

Firefly Custom Models lets enterprise users generate stylized outputs with consistent visual tone across asset sets. Together with Adobe Brand Intelligence’s brand rule verification, this can multiply an organization’s design productivity.

Some Not-So-Optimistic Points

Of course, the update isn’t without issues.

First is performance. AI Assistant in Photoshop relies on cloud inference, which demands a stable network. Anyone in China who’s used Firefly over the past year knows connectivity has been a persistent challenge. Adobe hasn’t offered a clear domestic access solution this time, so expect the same situation in the short term.

Second is subscription barriers. Firefly AI Assistant is currently only available to Creative Cloud Pro or paid Firefly plan users; free users get only basic image generation. Given the recent pricing trend for Creative Cloud, AI features are effectively a perk for heavy-paying subscribers.

Third is agent reliability. Whether the AI Assistant can truly take over “auto-organizing assets” depends on its performance in edge cases. The demo looks smooth, but faced with thousands of poorly named files, whether it can accurately identify and classify them remains to be seen. Adobe is calling this a beta, not GA — that distinction matters.

Final Thoughts

Beyond specific features, the most noteworthy aspect of this update is the shift in product philosophy.

Previously, Creative Cloud’s logic was “a complete toolbox, each tool doing its own job,” requiring users to jump between a dozen apps. Now, Adobe wants to condense all app capabilities into a single conversational interface — you describe the goal, and the AI orchestrates the toolchain. This isn’t just another new function; it’s a reconstruction of the creative interaction paradigm.

Whether this direction succeeds depends on two things: the quality of the underlying models, and the stability of the workflow orchestration. Adobe has solved the former by aggregating third-party models; the latter requires iterative refinement. But directionally, this is Adobe’s clearest AI statement in three years — they don’t intend to make the best model, but rather to own the indispensable workflow OS layer above the models.

For designers, it’s worth spending time getting familiar with this new interaction style. For developers, Adobe’s large-scale inclusion of third-party models indirectly validates the viability of “model aggregation” as a business model — whether in creative or developer tools, providing a single entry point to all mainstream capabilities is the more valuable product form.

References

• ITHome: Adobe Firefly AI platform upgraded, bringing agents into Creative Cloud suite — ITHome’s Chinese report on the latest Adobe Firefly update, covering Creative Skills, in-app AI assistants, and other core features.