basis ia restructure & ai agentic experience

The strategic context

This work emerged directly from my transition off the Reporting project into platform-level work. The VP of Product Experience identified two connected problems that needed dedicated staff-level design attention: the platform's underlying IA needed rethinking, and AI was arriving faster than the organization had frameworks to handle it well.

Kicking off the persona work

As part of the discovery work for an updated information architecture for the platform, I helped initiate an expanded persona effort in collaboration with the UX Research team. The goal was to move beyond traditional persona definitions and layer in roles and responsibilities, capturing what different users are accountable for, what decisions they make, and how their needs shift across the campaign lifecycle. This gives product teams and designers a more actionable foundation for scoping features and resolving competing priorities.

Once the framework was established, I handed the ongoing work to UX Research to lead, allowing me to focus on the IA and AI initiatives. The expanded personas have since been shared with the broader product organization with much enthusiasm from Product and Engineering.

Proposed IA & Navigation

The IA work was intended to be presented to the broader product organization. Before that could happen, the VP redirected my focus toward the AI framework — recognizing that teams were already moving and that getting ahead of fragmentation was the more urgent priority. The IA work isn't shelved; it's being held as a foundation to inform and connect future cross-functional conversations as the platform evolves.

Defining the AI interaction framework

The core challenge for the AI framework wasn't technical — it was conceptual. Most teams default to the same pattern when adding AI: a chat panel, a modal, or a separate page. These approaches treat AI as a destination users navigate to, rather than something woven into the flow of work. That model is familiar but limited, and applying it inconsistently across a complex platform would create exactly the kind of fragmentation the initiative was designed to prevent.

My exploration started with a foundational question: where should an AI agent live in relation to the content a user is already working with? That led to defining four distinct modes that describe how an agent can be present across different contexts and levels of user intent:

Driver — the agent takes center stage. The primary focus of the screen is interaction with the agent, with content surfaced below or alongside. Suited for task completion flows where the user is actively directing the agent to make changes, set parameters, or complete actions on their behalf.

Immersive — similar to Driver in layout, but the agent offers suggestions and guidance rather than making decisions. The agent sits above the content with contextual recommendations, moving below the content when data density requires it. The user remains in control; the agent informs without directing.

Minimized — the agent recedes but stays available. Suggestions appear below the content or in a collapsible panel; the agent can be opened when needed but doesn't compete for attention when the user is focused on the work itself.

Sleep — the agent is present but inactive. Represented by an avatar with a notification bubble the user can choose to open, this mode ensures the agent doesn't disappear entirely but creates no interruption when the user doesn't need it.

Across all four modes, the agent also serves as the home for help and notifications — consolidating two functions that are often handled separately into a single persistent presence that users can learn to rely on.

The framework also addressed interaction feel: AI text revealed through a typing animation to make responses feel seamless and conversational rather than transactional; an agent avatar that users can reposition or pin; navigation that can be pinned or hidden depending on user preference; and personalization settings including agent personality and tone, dark/light mode, and UI color preferences.

Organizational impact

The VP of Product Experience recognized the AI modes framework as directly aligned with a decision quadrant she had developed to help the product organization evaluate which AI/UX patterns were appropriate for different feature contexts. That alignment accelerated the framework's credibility — it wasn't just a design artifact, it was a tool product teams could use to make better decisions about how they introduced AI into their own areas of the platform. That's the outcome this kind of work is designed to produce: not a single feature shipped, but a shared language that prevents a category of future problems.