Uploaded August 2026 | Updated September 2026, 19 hours ago
Soomee Kang and Andreas Eiken, along with the Tanzu UX and design team, have spent about six months rebuilding how design works at Tanzu around AI agents.
What came out of it: the handoff to engineering is no longer a Figma spec, it's an HTML file that deliberately doesn't look like the product. The design system stopped being just documentation for humans and became the context layer that stops an agent inventing its own product. And Soomee's read on the job itself - "we like to call it behavior engineer, no longer UX designer."
Also in here: the thing they were sure would work and didn't, why "too much tokens" changed their whole prototyping approach, a feedback librarian that turns weeks of customer research synthesis into under an hour, and Andreas on whether an agentic pipeline is quietly rebuilding the waterfall that XP spent twenty years dismantling.
Soomee runs UX for Tanzu Hub and Intelligence Services. Andreas is a product designer on her team who came up through Pivotal Labs.
This is the design half of a conversation that started with product management. The first one, with Chris McClanahan:
youtube.com/watch?v=kvpnindfsAk
And the write-up of that one:
blogs.vmware.com/tanzu/using-ai-for-product-management-the-tanzu-platform-product-management-experience
FIND THE GUESTS
Soomee Kang: linkedin.com/in/soomeekang
Andreas Eiken: linkedin.com/in/andreaseiken
Andreas's portfolio: https://www.andreaseiken.com?utm_source=cote&utm_medium=youtube&utm_content=tanzu-ai-ux
MENTIONED
BMAD Method, the agentic pipeline they adopted: bmadcodes.com/bmad-method
Clarity, the open source design system underneath theirs: https://clarity.design
CHAPTERS
0:00 Introductions
1:40 Andreas's background
3:28 How AI has changed the UX workflow
8:43 Perfection vs. speed
11:09 The BMAD pipeline and human in the loop
14:28 Evaluating the output: three pillars
17:15 Human judgment vs. AI
19:46 What's working: synthesizing data and the feedback librarian
21:47 Fixing things directly in the code
23:07 Fit and finish with the actual front-end is too many tokens
24:31 HTML prototypes instead of Figma specs
26:45 Can you just give the agent a goal?
27:59 A design process in small context windows
29:31 Design systems
31:25 AI as the automated governor
32:26 More time on the design system, or less?
34:01 Everyone can work on the design system now
35:18 More documentation discipline
37:00 All the undocumented stuff
37:51 Team structure and roles
39:39 Pods
40:58 Balanced teams and eXtreme programming
43:07 What's the next experiment?
44:28 Designing an interface an AI can summarize
45:17 Making the pipeline our own
46:47 Wrap-up and where to find us
Soomee Kang and Andreas Eiken, along with the Tanzu UX and design team, have spent about six months rebuilding how design works at Tanzu around AI agents.
What came out of it: the handoff to engineering is no longer a Figma spec, it's an HTML file that deliberately doesn't look like the product. The design system stopped being just documentation for humans and became the context layer that stops an agent inventing its own product. And Soomee's read on the job itself - "we like to call it behavior engineer, no longer UX designer."
Also in here: the thing they were sure would work and didn't, why "too much tokens" changed their whole prototyping approach, a feedback librarian that turns weeks of customer research synthesis into under an hour, and Andreas on whether an agentic pipeline is quietly rebuilding the waterfall that XP spent twenty years dismantling.
Soomee runs UX for Tanzu Hub and Intelligence Services. Andreas is a product designer on her team who came up through Pivotal Labs.
This is the design half of a conversation that started with product management. The first one, with Chris McClanahan:
youtube.com/watch?v=kvpnindfsAk
And the write-up of that one:
blogs.vmware.com/tanzu/using-ai-for-product-management-the-tanzu-platform-product-management-experience
FIND THE GUESTS
Soomee Kang: linkedin.com/in/soomeekang
Andreas Eiken: linkedin.com/in/andreaseiken
Andreas's portfolio: https://www.andreaseiken.com?utm_source=cote&utm_medium=youtube&utm_content=tanzu-ai-ux
MENTIONED
BMAD Method, the agentic pipeline they adopted: bmadcodes.com/bmad-method
Clarity, the open source design system underneath theirs: https://clarity.design
CHAPTERS
0:00 Introductions
1:40 Andreas's background
3:28 How AI has changed the UX workflow
8:43 Perfection vs. speed
11:09 The BMAD pipeline and human in the loop
14:28 Evaluating the output: three pillars
17:15 Human judgment vs. AI
19:46 What's working: synthesizing data and the feedback librarian
21:47 Fixing things directly in the code
23:07 Fit and finish with the actual front-end is too many tokens
24:31 HTML prototypes instead of Figma specs
26:45 Can you just give the agent a goal?
27:59 A design process in small context windows
29:31 Design systems
31:25 AI as the automated governor
32:26 More time on the design system, or less?
34:01 Everyone can work on the design system now
35:18 More documentation discipline
37:00 All the undocumented stuff
37:51 Team structure and roles
39:39 Pods
40:58 Balanced teams and eXtreme programming
43:07 What's the next experiment?
44:28 Designing an interface an AI can summarize
45:17 Making the pipeline our own
46:47 Wrap-up and where to find us
