01 / Heard
A recurring user problem
I listened closely enough to recognize that scattered requests, market nuance, and review complexity were symptoms of a larger workflow gap.
Product design + strategy / User-informed prototype
Content Operations Agent
I turned a problem raised by the UX design manager leading myPepsiCo into a working proof-of-value prototype that created immediate pilot demand and an active funding case.
View working prototypeRole
Discovered, researched, designed, built
Timeline
3-week value prototype
Stack
React, TypeScript, Node
Outcome
9 users engaged → Canada pilot proposed
My design position
AI content generation is not enough. A trustworthy product must connect creation, standards, human review, ownership, and governance.
From Heard Problem to Pilot Demand
02 / Learned
User-informed direction
I worked with UX partners and the people closest to the work to understand user needs, trust requirements, and enterprise constraints.
03 / Built + tested
A prototype to prove value
I designed and built the working experience, tested it with users, and iterated on the moments that shaped clarity, control, and confidence.
04 / Demand
Evidence created demand
The tested build made the opportunity tangible, created urgency from users, and led to Canada being proposed as the first real pilot market.
Working prototype used to test the agent workflow with users and make the product value tangible to stakeholders.
Agent workflow demo
Open live build ↗
What I drove
I created the opportunity—not just the interface.
The work moved from an observed problem to a tested product direction because I took initiative across discovery, UX strategy, prototyping, and stakeholder alignment.
Recognized and framed the opportunity without a formal brief.
Designed, built, and tested the end-to-end human–AI workflow.
Turned user evidence into pilot demand and an active funding case.
The problem
Knowledge creation was fragmented by design.
Creating knowledge articles was difficult, inconsistent, and treated as extra work—yet those articles help employees understand nearly every part of PepsiCo.
Effort
Experts became accidental content operators.
Authors relied on content teams for support, pulling both groups away from their primary responsibilities.
Quality
Standards changed from article to article.
Structure, tone, accessibility, and publishing requirements were applied inconsistently.
Governance
No one could see content health end to end.
Ownership, status, review dates, duplication, and stale content lacked a central view.
Why it matters
When internal knowledge is inconsistent or outdated, employees do not know what to trust—and teams spend more time creating, correcting, and moderating it.
Research synthesis
Three roles. One connected lifecycle.
User testing revealed distinct patterns in who creates, reviews, and governs knowledge—and where each role needs confidence.
Create
Content Writer
Turn subject-matter expertise into an employee-ready article.
- Guided writing support
- Clear requirements
- Readiness feedback
- Simple approval path
Review
People Manager
Create content and manage the work of a team.
- Review and approval queues
- Ownership and status
- Audience and access
- Quality signals
Govern
Content Owner
Keep organizational knowledge trustworthy at scale.
- Organization-wide visibility
- Roles and permissions
- Content health reporting
- Areas requiring action
Role-to-workflow map
Role
Create
Improve
Review
Govern
Employee author
Start with what they haveConversation, structured form, or document upload
Receive guidanceAI applies standards, structure, tone, and accessibility
Submit confidentlySee readiness and respond to requested changes
Track progressKnow ownership, status, and next action
People manager
Support team contentSee drafts across direct reports
Strengthen qualitySpot gaps before formal review
Manage the queueApprove, return, or route work
Maintain accountabilitySee owners, audiences, and access
Content owner
Set the guardrailsDefine templates, criteria, and source context
Standardize qualityManage reusable organization-wide standards
Resolve exceptionsHandle permissions, escalations, and policy risk
See system healthMonitor stale, duplicate, and at-risk content
User feedback that informed the build
Nine users shaped what the prototype became.
The nine participants all create or support content alongside their primary jobs. Across three interviews and four prototype feedback sessions, I studied the current workflow and used each demo to validate, extend, or pivot the product direction.
Insight 1
We made conversation the primary path without removing the form.
Users were already turning to ChatGPT and Copilot to draft articles. The redesign meets that learned behavior immediately, while preserving a structured alternative for people who prefer not to use AI.
Insight 2
We expanded the admin view from managing articles to understanding content health.
Content owners told us a list was not enough. The latest direction introduces system-level health, discovery, maintenance, and follow-up insights so administrators can see where attention is needed—not just what articles exist.
Insight 3
“Let me upload what I have and have the agents write it for me.”
We added a more passive creation path that accepts existing source material and asks the agents to turn it into a standards-ready draft. Users can review the result in My Articles rather than building every article field by field.
Additional findings
Standards
Important criteria were difficult to apply manually.
Writers valued PepsiCo standards but rarely had time to check and rewrite against every requirement. We embedded structure, tone, accessibility, and publishing guidance into AI-assisted creation.
Learnability
Infrequent use made simplicity essential.
Because this would not be a daily tool, users could not relearn a specialized workflow each time. We used familiar AI patterns, simplified onboarding, and kept guidance visible.
What users said
“We can see our feedback influencing what was implemented since the last time we talked. This is so awesome.”
“How can we get this next week? This is going to provide so much value, and we need to get it out ASAP.”
How UX worked
Build to learn. Test to earn belief.
I kept research, design, prototyping, and validation in one continuous loop—moving from a heard problem to demonstrated user demand.
01
Hear
Recognized a systemic problem without waiting for a formal brief.
02
Define
Learned from UX partners and future users, then framed what the product needed to prove.
03
Build
Designed and coded a realistic end-to-end prototype with AI-assisted development.
04
Test
Observed real users, listened for trust gaps, and iterated on their feedback.
05
Build support
Used evidence from the working product to create urgency, propose a first pilot, and support active funding proposals.
Product model
One system for the full knowledge lifecycle.
The prototype connects creation, quality, governance, and scale instead of treating each as a separate tool.
01
Create & improve
- Guided conversation or form
- Existing-content upload
- AI-assisted writing
02
Standardize quality
- Content and design standards
- Consistent tone and structure
- Accessibility guidance
03
Review & govern
- Approval workflows
- Roles and permissions
- Review dates and duplicates
04
Manage & scale
- Central article library
- Status and health visibility
- Canonical source of truth
Product experience
What would you like employees to understand or do?
Structure and readability✓ Meets standard
Accessibility1 recommendation to review
ArticleOwnerStatus
Benefits guidePeople teamReady
Travel policyFinanceChanges
HealthyOn track
ReviewDue
StaleAttention
Content health by area
Value chain
Self-service + standards + governance
→
Less effort + higher quality
→
Greater trust + sustainability
Section · 03
The Multi-Agent Architecture
I architected the interaction model around a visible orchestrator, giving human authors transparent control over modular AI sub-processes instead of hiding automation inside a black box.
CoreOrchestrator
Part · 01 of 07
Intake
Checks that the brief has the minimum context needed to begin.
HandoffPasses a complete brief to Clarifier
AI as an accelerator
The prototype reduced the cost of learning.
AI-assisted development helped me turn product strategy into a realistic experience quickly. UX judgment still determined what to build, what to test, and what the evidence meant.
Lower risk
Validate before full investment
Test the problem, demand, and experience before committing to full development.
Faster feedback
Put real interactions in users’ hands
Learn from behavior and reactions instead of relying only on static screens.
Stronger evidence
Make the opportunity tangible
Give users and leaders something concrete to evaluate, discuss, and support.
Key Product Decisions
01 / Permission model
Separate admin governance from author workflows
Problem
A single view mixed enterprise governance with everyday article management.
Move
Split the experience by permission level: admins get system-wide controls; writers and managers get focused ownership tools.
Result
Each role sees the right scope without extra operational noise.
Admin view
Monitor article health, spot stale or low-performing content, and alert owners to review, update, consolidate, or remove.
Non-admin view
Manage owned and direct-report articles, track stale reviews, filter by owner/status, transfer ownership, and make updates.
02 / IA model
Sector before market, not market alone
Problem
Design conversations made it clear that market-only organization ignored how content ownership actually works.
Move
Made sector the primary organizing tier across profiles, filters, and articles.
Result
Authors choose "PFNA → United States," so the prototype feels specific to PepsiCo's operating model.
Technical Alignment
Ensured data indexing paths match PepsiCo's operational hierarchy, mapping API calls directly to sector boundaries.
03 / Workflow surface
Merge review queue into content library
Problem
Users needed to understand what was in progress, what needed attention, and what was already live.
Move
Collapsed review and library into one All Articles surface with three tabs.
Result
Reviewers and authors can move between work-in-progress and live content without losing context.
Technical Alignment
Optimized state management in React to prevent rendering lag when toggling between dense, real-time localized dataset tables.
04 / Grounding model
Sources as profile data, not per-article attachments
Problem
Repeated source setup would make the prototype feel slower than the workflow it was meant to improve.
Move
Put sources on sector, market, and audience profiles instead.
Result
Grounding follows ownership: PFNA sources guide PFNA content; US sources guide US procedures.
Technical Alignment
Caching system data globally on sector profiles drastically reduced payload sizes and context-window token overhead for subsequent Claude API inference cycles.
From Prototype to Pilot Demand
AI-assisted creation
Users can create standards-aligned articles with less manual checking and rewriting.
Canada test readiness
The team is maturing the conceptual prototype so it can test speed, independence, and reduced support effort.
Connected operations
End-to-end governance, administration, messaging, notifications, and deeper PepsiCo workspace integration.
Current outcome
Users asked for the product immediately, and Canada was proposed as the first pilot market.
Pilot test
Measure faster article creation, greater employee independence, and less hands-on support from content teams.
Status
The prototype is still being developed toward pilot readiness; funding proposals are active, with no decision yet.
What this shows about how I work: I recognize opportunities, stay close to users, and use working prototypes and evidence to help an organization decide what should become real.