AI Platform — Infrastructure & Admin Controls
Turning years of ambiguous GPU cost and utilization requests into a shared vision, one anchor page that shipped, and a roadmap the team is now building against.
01 · Point of view
Knowing what AI is good at, and where judgement still matters
With all of the disruption of AI in UX, it's important to understand what it is good at, where human judgement is important, and when we can let it do the work so we can solve other problems. This project was where I put that to the test on a genuinely messy problem.
02 · The problem
Expensive GPUs, no way to track the cost
One of the biggest problems with deploying AI in an enterprise setting is keeping track of costs and making sure usage of expensive and in-demand GPU resources is optimized. This is a new, large feature area with many new requests from customers and sales, often ambiguous problem statements, and many disparate feature needs from PM, some dating back years.
03 · Making sense of the backlog
Using AI to build a vision the whole team could share
To make sense of all the requests and give the team direction, I started by working in Cursor to create a vision that could be shared across PM, front-end engineering, back-end engineering, and field sales. I fed in numerous strategy and idea documents from PM, tech specs and implementation explorations from engineering, and sales and field requests. With all of that ingested, I created a plan for what to include in a vision and the steps I needed to take to get there.
04 · Aligning on the workflow
One map of every user and every touchpoint
The first step was to create a workflow tracking all the different users and where they would want to interact with the new feature. That gave me something concrete to share with stakeholders to get common agreement on what we needed to build, and on the pages we needed to design to build it.
05 · The anchor page
An Infrastructure overview that earned the rest of the vision
With agreement on that workflow, I dove deep into what was obviously the most important page, the one that would anchor the rest of the experience. I designed an Infrastructure overview page that lets admin users track GPU usage and strain across the product. It kept a customer happy at a critical juncture and provided fuel to make sure the rest of the vision continued to be implemented.
06 · Extending the vision
Connecting pages, and making scheduling legible
As that page was being implemented, I refocused on the larger vision. Again using Cursor, I mapped out high-level work connecting pages together and began defining future pages and interactions. These let users see more detailed information, understand the technical and complicated nature of the workload scheduling hierarchy, and eventually control all of it through the UI.
07 · Mentorship
Coaching an associate designer through the detail work
While creating the high-level vision I also worked with an associate designer, mentoring and coaching them toward creating the more detailed designs we hand to engineering. That let us make progress on both the larger vision workstream and the specific higher-fidelity pages that tie the vision into what is actually being built into the product.
08 · Where it stands
Vision work becoming product
We're now at an exciting point where the vision work is coming to life in the product. The next steps are managing the time pressure and the often competing priorities to deliver a quality product to customers on their timetable. The work we spent on the vision provides a great scaffolding to make that a reality.