Google Cloud AI's Gemini Enterprise team set out to make multi-agent workflows feel simple and trustworthy. Sheta Chatterjee, Senior Director and Head of UX and Design at Google Cloud AI, describes the guiding principle behind the product: the user's focus should stay on their goal rather than on managing the AI. At the same time, the interface must make clear that the user can step in — people are handling sensitive information and making decisions with real business outcomes. Simplicity and transparency were prioritized across the chat interface, visual dashboard, custom agent builder, and shared project spaces.
A shared brand, differentiated by feature
Gemini Enterprise builds on the consumer Gemini app, and the team kept one visual language across both: the sparkle icon, guiding gradients, rounded shapes, and intentional motion. The differentiation happens at the feature level. The prompt box stays consistent with the consumer app, but surfaces connectors more prominently so the assistant can draw on the tools a business already uses — Google Workspace, Jira, and Notion among them.
The team also looked past the prompt box entirely. They see work shifting from single commands toward orchestration: multiple tools, multiple data sources, and tasks delegated to multiple agents. That led to the AI Inbox, a living dashboard that shows what agents are working on, what is complete, and what needs intervention — for example, a market analysis due the next morning that is ready for review. The visual workflow reads more like a team check-in than a chat thread.

From chat threads to a shared workspace
Because this is not a solo productivity tool, the design moved from individual chat threads to a persistent, shared project space in which AI acts as another team member — performing tasks, summarizing discussions, and pulling up project files. Everything happens where the whole team can see and reference it, and each request is attributed to a team member. That attribution supports accountability and helps others understand the context behind an agent's actions.
The assistant's constant presence also closes knowledge gaps: if an engineer uploads a technical spec, a designer can ask for details without hunting the spec down. Chatterjee frames silos as the biggest problem with work today, with AI serving as a single source of truth — an integration that turns the assistant from a productivity tool into a team intelligence amplifier.

Proactive help without interruption
An agentic system also needs to anticipate needs. The team is experimenting with non-intrusive nudges, such as offering to draft a status update when a project deadline approaches, or having the assistant raise its hand during a chat when it has something relevant to add. Such suggestions should feel magical rather than like interruptions, and seamless tool integration is treated as a foundational pillar so the AI understands the user's world.
Transparency as the basis for trust
The consumer app narrates what it is doing in real time; the enterprise version extends that thinking state at a finer level of detail. Asked to track the health of a recent launch, an agent states its intended plan — understanding the project, breaking down customer feedback in SurveyMonkey, analyzing support tickets in Jira, and drafting a memo of key takeaways. Chatterjee calls this a deliberate pause and a moment of designed friction that keeps the user the ultimate authority. Source citations are always included in responses.
For custom agents, the Agent Designer lets users define a harness: a governance layer specifying which data an agent can access, which tools it can use, and when it must stop and ask for permission. Security and logic sit directly in users' hands.
Keeping many surfaces consistent
Figma serves as the team's single source of truth from early concepts through shipping. FigJam is used up front to map user goals, understand journeys, and vote on ideas before moving to higher-fidelity work. Any change has to be evaluated for how it cascades and scales across everything the team plans to add. Building, documenting, and maintaining a design system in Figma is essential at this scale — it allows a level of detail and management of complex states that Chatterjee says is not possible with vibe coding. The single source of truth also lets the team cross-reference with the consumer team so a successful experiment on one side is reflected on the other.
What the moment demands of designers
Chatterjee compares the current period to the era before graphical interfaces, when computers ran on command lines. The principles that always mattered — clarity, information architecture, trust, visual polish — remain. Many teams are building fast, but the details decide which AI products people actually use. The terminology of agents and orchestration can be intimidating, yet end users simply want to get their work done, and making things feel simple, human, and delightful is what designers do well. In an age of rapid prototyping, when it is easy to build things that are not necessarily right, craft and taste matter more than ever.



