AI-native Admin Hub
Designing a proactive control layer for enterprise admins
I led the 0→1 product direction for an AI-native admin experience that helped enterprise admins move from fragmented, reactive workflows to proactive oversight. The vision introduced Watchtower: a new admin home that surfaces prioritised recommendations, monitors critical signals, and supports guided investigation through AI-assisted canvas workflows.
Role
Senior product designer
Type
0→1 AI-native product vision
Team
With Lead PM, Senior Eng & Design Manager
Outcome
Aligned the team around Watchtower as the path for AI-native admin, with early LLM-powered proof-of-concept work approved
The problem
Current state: admins had to piece the system together manually
Before Watchtower, admins had to move across multiple product surfaces to answer basic questions:
What needs my attention?
Is this a risk, an opportunity, or normal activity?
What evidence should I trust?
What action should I take next?
The challenge was not simply to make dashboards better. It was to design a new operating model for admin work.
Why it mattered
A business strategy problem
The opportunity was not just a better admin experience. A more proactive admin model could support three business outcomes:
My design challenge
How might we help enterprise admins notice what matters, understand why it matters, and act with confidence — without forcing them to manually search across fragmented admin tools?
This gave the team a clearer product direction:
move from reactive to proactive
move from scattered dashboards to prioritised signals
move from raw data to explainable recommendations
move from chat-only AI to guided workflows
move from isolated admin tasks to a connected operating model
The vision
Watchtower was designed as a complementary admin “home,” not a replacement for the existing Admin Hub.
It brought together three parts:
Recommendations — surface the most important tasks, risks, and opportunities
Monitor — track important signals, anomalies, and system health
Canvas — help admins investigate, plan, and complete complex work in context
Together, these created a new AI-native operating model for admins: one that was proactive, explainable, and action-oriented.
Onboarding: personalising the admin contex
The onboarding flow helped establish each admin’s responsibilities, priorities, and context. This gave Watchtower a stronger foundation for personalising recommendations and surfacing relevant signals.
Instead of treating every admin the same, the system could understand what the admin cared about — such as security, compliance, user management, adoption, or cost optimisation.
Recommendation: helping admins notice what matters
Recommendations were designed to reduce the burden of manual discovery. Instead of expecting admins to search through dashboards, Watchtower could surface high-priority work items with context, rationale, and a suggested next step.
Each recommendation needed to answer:
What is happening?
Why does it matter?
What evidence supports this?
What should I do next?
Monitor: helping admins track signals over time
Monitor gave admins persistent visibility into the areas they cared about, such as usage, adoption, cost, risk, or compliance. While recommendations surfaced what required attention, Monitor helped admins understand whether the system was healthy over time.
This pattern was important because admins did not only need one-off alerts. They needed confidence that they could observe change, detect anomalies, and understand whether previous actions had improved the system.
Canvas: moving beyond chat into guided investigation
Canvas explored how AI could support deeper admin work beyond simple Q&A. It gave admins a workspace to investigate an issue, review evidence, compare options, and move toward action while keeping the reasoning visible.
This was important because enterprise admins need more than fast answers. They need confidence, traceability, and control.
Canvas helped translate AI from a conversational assistant into a decision-support workflow.
The system: one connected Watchtower framework
The final framework connected the three patterns into one operating model:
Onboarding builds context. Recommendations highlight what matters. Monitor tracks what changes. Canvas supports investigation and action.
This turned Admin Hub into a more proactive, personalised, and action-oriented experience.
AI interaction patterns - beyond chat
A chatbot could help admins ask questions, but it would not solve the deeper workflow problem. Admins needed a system that could proactively surface what mattered, explain why it mattered, and guide them toward the right action.
Process
5 layers to turn a broad AI opportunity into a focused product direction
1. Aligned with product and engineering on AI potential
I worked closely with product and engineering to define where AI could meaningfully improve admin work — especially across recommendations, signal detection, and cross-system visibility. We focused on where LLMs could unlock better recommendations and connect signals across systems, rather than using AI for novelty.
2. Studied the market to understand where admin tooling is heading
I looked at how major SaaS platforms and emerging AI-native products were evolving admin workflows. This helped ground the vision in real market movement and clarify where the opportunity was to lead, not just follow.
3. Explored broadly, then narrowed into a focused concept
The work started with a broader concept, Modes, which explored intent-based AI-driven admin journeys. I helped narrow that into Watchtower — a more focused hero experience centered on the most critical frictions: fragmented workflows, too many touchpoints, and limited visibility into what matters most. This made the direction clearer, more actionable, and easier to validate.
4.Collaborate with product & Eng to maximise ROI
The vision balanced ambition with realism. It proposed starting with in-context recommendations while building toward the fuller Watchtower experience, making space for incremental value without losing sight of the broader operating model.
5. Validated what felt useful, credible, and adoptable
I used early validation to understand how admins responded to AI-assisted ways of working. The key insight was that Monitor had stronger pull than Recommendations because it felt closer to how admins already work while still moving them toward a more proactive model. That helped shape the sequencing and rollout strategy.
Outcome
Alignment that the Watchtower path is worth pursuing
The AI-native admin vision creates two immediate outcomes: alignment that the Watchtower path is worth pursuing, and focused investment to prove it out quickly.
That investment allows us to rapidly build and iterate on lightweight LLM-powered proof of concepts, so we can start delivering value through early recommendations, learn what core system components are needed, and gain enough confidence to shape the next phase of investment.
Aligned on watchtower
Dedicated spike team to explore the core system components
Resource approved
Lightweight LLM-powered proof of concepts
Reflection
AI native, from the design process to the output
This project pushed me to adopt a more AI-native design process — using vibe coding, building a Replit prototype, and prototyping with Cursor and VS Code to explore ideas faster. It also deepened my understanding of the different characteristics of LLMs, LAMs, and eval-based evaluation, helping me shape both the product direction and the way I design in a more AI-native way.