AI-Augmented Practice
I added UX guidance to Delta’s enterprise AI kit in two days
Delta’s enterprise AI kit launched with no UX guidance. Teams could generate interfaces, but nothing told the AI how Delta designs them. Nobody asked me to fix that. I also used AI in my own prototyping and research synthesis.
Results at a glance
Three places I used AI
Two were steps where manual work slowed delivery. The third was a gap in a kit offered to every product team.
Prototyping
I turned notes from a discovery session into an interactive HTML prototype within hours.
Kiro enterprise AI kit
UX guidance for a kit that every product team could adopt. It had none at launch.
Try the UX auditor demo (opens in a new tab)
Research synthesis
AI summarized the transcripts from scripted sessions, and the facilitator checked the summaries.
Problem & stakes
Three steps in the process depended on manual work.
Stakeholder alignment, research synthesis, and prototyping all took a lot of manual work. After a discovery session, it took days before the team had something visual to react to, and the handoff between discovery and development was often unclear.
When Delta launched its Kiro enterprise AI starter kit for product teams, the kit had no UX guidance. Teams could use AI to generate interfaces, but nothing in the kit told the AI how Delta designs them.
Process
I started with my own work and then offered it to other teams.
Prototyping with Kiro
Starting in late March 2026, I used Kiro to turn notes from discovery sessions into interactive HTML prototypes within hours. I did this on three to four features. Teams had something concrete to react to right after discovery. Before, that took several days.
Contributing to the enterprise AI kit
Within days of the kit’s launch, I noticed that it had no UX guidance. Without being asked, I built three components over two days: a steering file that loads Delta’s personas, design principles, and artifact standards into every session; an on-demand UX auditor agent based on Nielsen Norman Group’s 10 usability heuristics; and a hook that runs the audit every time a UI file is saved. I sent them to the General Manager of Developer Experience at the end of April 2026.
AI-assisted research synthesis
Usability sessions followed a consistent script, so the transcripts could be compared. AI wrote a first draft of the findings, and the person who ran the sessions checked it against what they had observed before anything was shared.
Trade-offs
AI drafts, and a person makes the final decision.
AI made synthesis faster, but it can also misread a session. So the facilitator reviewed every summary. That used up part of the time saved, and I accepted that.
The prototypes were for discussion. They were not production code, and I did not spend time making them complete or polished.
I set strict rules for the auditor agent: it reports only issues with evidence in the code, leaves accessibility to a separate audit, never edits files, and says so when it finds no issues. These limits reduce what it covers, but they keep it from inventing problems.
The kit belonged to another team, and nobody had asked me for anything. The risk was that the work would not get used. I still do not know whether it was.
Outcome
Prototyping and synthesis got faster. I sent the kit’s owners the UX guidance, and I do not know if they adopted it.
Prototyping: the time from a discovery session to an interactive prototype went from days to hours on three to four features.
Research synthesis: analysis time went from 6 hours to 2, on average across several usability studies, with the facilitator reviewing every result.
Enterprise AI kit: in April 2026, the General Manager of Developer Experience told me that the Developer Experience team planned to use the files. The team had already asked the UX design department for similar guidance and had not yet received it. I do not know whether the files were adopted after that.
Public demo: the UX auditor runs as a public demo with the same agent rules, outside Delta’s environment.
“Alex has been a trailblazer when it comes to integrating AI into his daily tasks in a thoughtful way.”
— Kirsten Erich, Technical Product Engineer
Reflection
The prototyping workflow depended on me. I would document it with prompt templates, a standard format for capturing session notes, and output conventions, so that other designers on the team could use it.