Highspot: Scaling AI through a native text refinement pattern

The challenge

Highspot’s exploration into AI text generation had remained in an alpha state while the platform underwent a major transition to a new text editor. Once the migration was complete, we faced a market where Generative AI was commonplace, yet our internal authoring experience remained manual and fragmented.

We observed several critical friction points:

• The "blank page" hurdle: Creators struggled to initiate content, making authoring feel like a high-effort chore
• Workflow fragmentation: Users were "alt-tabbing" to external tools like ChatGPT to polish copy, losing the "deep context" of Highspot’s CRM data in the process
• Pattern gaps: Multiple crews were building isolated AI prompts, leading to a fragmented experience without a unified primitive for text refinement

Strategic framing

This project was a "Big Bet" for our product strategy. My role was to ensure the crew didn't just solve this for SmartPages, but built a common pattern that could be extended across the entire platform. 

We focused on:

• Unified patterns: Designing a refinement component other crews could adopt without reinventing the solution.
• Strategic sequencing: Prioritizing high-volume, one-click refinements (Shorten, Elaborate, Professional) to build habits before moving to full generation.
• Scalability: Building a foundation to eventually layer on Highspot-specific insights like CRM data.

Principles for project

Synthesizing complexity

We began by auditing how industry leaders like Notion, Canva, and Google handle the transition from text generation to refinement. 

I facilitated a whiteboard jam to align on several core hypotheses:

• Speed over robustness: For an early version, staying in context and building confidence may matter more than offering the most creative output
• Winning on refinement: Highspot may not compete with general-purpose tools on full content generation, but we win by helping users polish existing content directly where it lives

Refined internally

To ensure the first milestone of this project met our high standards for usability, we gathered critical feedback from the Crew to stress-test our initial patterns before moving into beta.

The key insights we prioritized included:

• Building trust through interaction: There was a strong preference for inline, "preview-before-insert" patterns to minimize distraction and create a more conversational, low-friction experience
• Architecting for platform scale: The crew wanted to ensure the refinement primitive was flexible enough to adapt across all surfaces, from the Email Composer to Item Properties
• Native editor alignment: To maintain the user’s flow, it was essential to focus on a seamless integration with the new text editing experience

This feedback acted as a vital "pulse check" for our first iteration, ensuring our foundational architecture was robust enough to support future agentic implementations.

Navigating tradeoffs

We approached the solution by investigating how to balance the power of GenAI with the speed required for a professional environment. We focused on building a native authoring primitive that reduced cognitive load and kept users in their flow.

Decisions we made:

• Inline over toolbar: While toolbar actions are familiar, they require a "travel tax" where users move focus away from their work. We chose an inline action that appears upon selection, meeting users exactly where they work
• Floating widget over modals and side panels: Modals provide space but obscure the page. Side panels also provide space but require more engineering effort. We landed on a floating widget to keep the experience in-context, allowing users to see changes applied in real-time
• Refinement over generation: We made a deliberate bet to prioritize one-click refinements. We recognized that Highspot wins not by competing with ChatGPT on broad creativity, but by helping users polish existing copy directly where it lives

Innovation through AI-assisted prototyping

As an AI-first organization, we wanted to embrace AI tools within our own design workflows. Once initial concepts were established, I used Cursor to handle the heavy lifting of building a functional prototype.

This allowed us to:

Reflection

This project was a powerful reminder that Staff-level design is about providing leverage, not just solutions. I needed to remind myself that to truly scale, I had to relinquish direct control of the pixels and lead through vision instead. By giving a more junior designer the space to own the execution while anchoring them in a clear experience model, I was able to scale my influence without becoming a bottleneck.

To keep the project moving, I focused on manufacturing momentum. I introduced short, frequent milestones, not to create pressure, but to establish a virtuous loop of feedback and iteration. This allowed us to return to Product and Engineering with validated ideas much faster than a traditional linear process would allow.

My takeaway on AI and Craft:

• Iterative prompting: Just like traditional design, prompting is a craft. I used Cursor and Figma mocks to build functional prototypes that prioritized "concept clarity" over "pixel perfection"
• AI as an extension: I learned that AI doesn't replace the design process; it accelerates the iteration cycle. Prompting is a new design medium—one that allows us to refine an outcome until it perfectly matches the strategic intent
• Organizational scaling: By delivering a "plug-and-play" primitive, we ensured that teams working on Email Pitches or Item Descriptions could adopt a high-quality pattern without starting from zero