Scaling UX content review with AI
How I led an AI-powered content review system that became SentinelOne's most-used AI agent
Company: SentinelOne
Role: Senior Staff Content Designer, project lead
Timeline: 2025–2026
Focus: AI agents, content operations, UX strategy, content governance
Impact at a glance
1,100+ uses of the UX Content Review agent
48% team adoption for automated content checks
Most-used AI agent at SentinelOne
Integrated into design review, content office hours, and self-service workflows
The challenge
At SentinelOne, product designers worked across a complex cybersecurity platform with highly specialized terminology, numerous interface patterns, and evolving product requirements.
Maintaining consistent, accessible UX content required designers to navigate multiple resources: the enterprise style guide, product glossary, Gravity design system, and UX writing guidelines.
Content review often happened through ad hoc requests. As the organization scaled, this created a challenge: how could we maintain content quality without requiring a content designer to manually review every interface?
I saw an opportunity to make our standards easier to access and apply by integrating them into an AI-powered workflow.
My role
As Senior Staff Content Designer, I owned the content strategy, conversational behavior, governance requirements, and evolution of the UX Content Review agent.
I partnered with designers, AI collaborators, and UX leadership to develop the experience, evaluate its performance, and integrate it into everyday design workflows.
My work included:
Defining the agent's role, instructions, tone, and response structure
Connecting recommendations to approved content standards
Designing guidance for content hierarchy, clarity, accessibility, and terminology
Developing enhanced prompting and review workflows
Testing recommendations against real product designs
Gathering feedback and identifying improvements
Driving adoption through documentation, training, and process integration
Designing a scalable review experience
The agent was designed to help product designers review interface content before requesting formal content design support.
Rather than simply rewriting text, it evaluated content against SentinelOne's established standards.
Its knowledge sources included the enterprise style guide, product glossary, taxonomy, Gravity component guidance, and UX writing heuristics.
I structured the experience around a more useful sequence:
Understand the user's goal and the purpose of the interface.
Evaluate the information hierarchy and content structure.
Identify clarity, terminology, consistency, and accessibility issues.
Recommend improved content.
Explain the rationale and cite relevant standards.
This made the agent more than an automated copy editor. It became a way to bring content design reasoning into designers' existing workflows.
Improving the agent through evaluation
As usage grew, I identified opportunities to improve the accuracy and depth of its feedback.
The existing experience sometimes recommended content that contradicted our standards, including inconsistent capitalization, incorrect modal guidance, and unsupported formatting.
I developed an enhanced prompt specification and experimental V2 agent to address these issues.
My improvements included:
Stronger guardrails: Explicit requirements for sentence case, terminology, punctuation, and adherence to approved guidance.
More useful feedback: Recommendations addressing information hierarchy and user intent rather than focusing exclusively on individual strings.
Explainable recommendations: Source citations and rationale so designers could understand why a change was recommended.
Context-aware review: Proposed enhancements for different cybersecurity personas and interface patterns, including error messages, modals, and empty states.
I tested the updated agent against existing product designs, comparing its recommendations with the baseline agent and previous content design decisions.
The experiments demonstrated improvements in the quality of explanations and hierarchy recommendations while also revealing remaining inconsistencies that informed subsequent refinement.
Making AI part of the design process
Building the agent was only one part of the challenge. It also needed to become useful enough for designers to adopt.
I incorporated it into the broader content design operating model, including pre-review checklists, office hours, and self-service resources.
Designers were encouraged to use the agent before formal reviews to identify smaller issues and prepare more focused questions.
This helped establish a repeatable process for applying content standards without making every decision dependent on a specialist.
I also supported adoption through demonstrations, team communications, and feedback collection.
Results
By March 2026, the UX Content Review agent had reached more than 1,100 uses and become SentinelOne's most-used AI agent.
Forty-eight percent of the team was using it to automate content checks.
Usage had passed 500 by December 2025 and more than doubled over the following months.
Beyond adoption, the agent established a scalable way to apply content guidance across product teams. It was referenced in design process documentation, pre-review checklists, and engineering workflows.
The organization also identified the agent as a mission-critical AI workflow during migration planning.
What I learned
AI adoption depends on more than the quality of the model or prompt.
For an AI tool to become part of a design organization's workflow, it needs trustworthy knowledge sources, clear instructions, useful explanations, and a place in existing processes.
The most valuable part of this work was not automating copy suggestions. It was making content design expertise more accessible and reusable across the organization.
My takeaway: AI can scale specialized expertise when it is grounded in well-designed systems, continuously evaluated, and integrated into how people already work.