Illustration: AI Assistant in product

nextgen
ai assistant

From a single rail to an eight-modality landscape across every Adobe CXO product. Design, engineering, and PM, working horizontally.
Scroll to read
ai assistant

It started as a Summit demo.

The rail helped users learn the product, get answers in context, and surface insights about their objects and data.

First-gen AI Assistant · AEP · 2024
Adobe Experience Platform home screen with the first-generation AI Assistant rail open on the right.
aep-rail.png
First-generation AI Assistant rail — state 1 [ first-gen rail · screen 01 ]
First-generation AI Assistant rail — state 2 [ first-gen rail · screen 02 ]
First-generation AI Assistant rail — state 3 [ first-gen rail · screen 03 ]
First-generation AI Assistant — three rail states from the 2024 build.
First-gen AI Assistant · Fullscreen · 2024
First-generation AI Assistant in fullscreen mode — the modal experience competitors had built.
modalfirstgenaia.png
ai assistant

Stakeholders wanted fullscreen.

Leadership wanted a full AI experience with fullscreen chat, like competitors had built.

The reframe
The ask was "make it bigger." The real problem was context.
ai assistant

We ran workshops to figure out what right looked like.

We started with the Audiences case study and worked it through a four-dimensional space. AI intelligence and agency. User expertise. AI presence. Workflow complexity.

Workshop board with sticky notes for User, Agent, and Collab placed along four dimensional sliders — AI Intelligence & Agency, User Expertise, AI Presence, Workflow Complexity — with the Audiences case study as the workshop subject Early framework — four dimensional sliders for AI Intelligence & Agency, User Expertise, AI Presence, Workflow Complexity, used to characterize any AI surface Early modality landscape — seven AI Assistant modalities (no AIA, contextual, prompt bar, hub, rail, split screen, canvas, fullscreen) mapped along a complexity gradient The framework as a working Replit tool — dimensional sliders, dimension weighting, and a recommended modality output panel
ai assistant

Tools work in fixed flows. AI doesn't fit one.

Traditional interfaces follow fixed flows. With AI, interfaces must become fluid, dynamic, and context-aware.

The landscape
Contextual
Prompt bar
Hub
Rail
Split
Canvas
Fullscreen
Immersive

Contextual.

AI shows up inside the artifact you're working on.

Inline edits, annotations, tooltips. It recommends, you decide.

Prompt Bar.

AI as a quick query.

Type in natural language. Get an answer. Move on.

Hub.

AI as a command center.

One place to review activity, search across objects, start new workflows.

Rail.

AI in a persistent rail.

Always present, never demanding. Familiar by design.

Split-Screen.

AI side by side with your work.

One half conversation, one half content. Built for iteration.

Canvas.

AI on a hands-on surface.

Drag, drop, edit, undo. For non-linear work where you stay in control.

Fullscreen.

AI takes the full screen.

For deep focus, brainstorming, and strategy.

Immersive.

AI as a co-creator.

Shared environment, shared artifacts. AI-first by design.

ai assistant

The framework in action.

A sample of modalities as shipped or envisioned for product.

ai assistant

Five findings that shaped the landscape.

"My first reaction is just to go in and edit it."
On Canvas Users default to manipulation, not conversation, when stakes are high.
"I would expect an AI to be a companion. No other app provides this."
On Split-screen Co-creation, not chat, is the bar users were measuring against.

Modality maps to complexity. Prompt Bar for queries. Full-screen for ideation. Split-screen for structured workflows. Canvas for complex orchestration. Rail for explanations. Contextual overlays for just-in-time guidance. The progression was intuitive to users. The absence of it is what failed them in existing products.

Prompt-only workflows risk abandonment. Users described prompting as unnatural for visual work, feared loss of precision on small tweaks, and worried prompt-based systems would misinterpret intent. The chat-only path was the most likely abandonment point in the entire research.

Switching needs UI controls, not prompts. Modality transitions in existing products were mostly prompt-only, sometimes one-way. Users wanted bidirectional UI controls. The landscape only works if movement between modalities is as easy as the modalities themselves.

Synthesized from generative research with Audiences case study participants, 2025.

ai assistant

Eight shapes need three axes to compare against each other.

A landscape isn't useful without a coordinate system. We picked three dimensions that let any modality be placed against any other — and let teams reason about which shape fits which job, without re-litigating it every time.

01

AI Reasoning

The AI's ability to understand context and solve problems. How much intelligence does this workflow need to support?

02

Workflow Complexity

How demanding the task is to complete, both in steps and in judgment. How heavy is the lift?

03

AI Autonomy

How independently AI acts on behalf of the user. How much of the work does the human still own?

Try it

The framework, live.

Three sliders. Eight modalities. Pick a modality to see its three-dimension fingerprint, or move the sliders to find the modalities that match your situation.

What the framework unlocked
AI workflows don't follow fixed paths. Modalities gave us shapes for that. Eng, design, and PM had a shared language for where AI should appear and why. That opened up hybrid UI in product, meeting users in the work they were already doing.
ai assistant

Nextgen AI Assistant was born.

We sold the vision to leadership. They gave us buy-in to ship more than just a fullscreen chat experience. It reshaped the story.

Nextgen AI Assistant.
ai assistant

Three months to design and ship.

Three months to design components and patterns, and to work with agent teams on the end-to-end experience. We prioritized by what was most feasible for the deadline.

June
1
Month One — Foundation
July
2
Month Two — Partnership
August
3
Month Three — Ship
12
Agents Shipped
700+
Bug bashers
100+
Design bugs logged
3
Months to GA
30
Customers onboarded
1
Monumental release
Adobe press release: General Availability of AI Agents for Adobe Experience Cloud
announcementga.png
Adobe announces General Availability of AI Agents.