Cheaper, faster production makes deciding what to build the harder job. What a design leadership panel said about ownership, shared measures, and proving value.
I joined a design leadership webinar yesterday expecting a conversation about speed. I left thinking about ownership.
The Design Executive Council hosted "Driving enterprise outcomes as a design executive: what it looks like in practice in the AI era." Founder Gordon Ching moderated four design leaders: Andy Vitale (Taxwell), Heather Winkle (Delta Air Lines), Matthew Menz (former AWS), and Catherine Winfield (Autodesk).
The panelists came from different industries and kept landing in the same place. When building gets cheaper, the hard part moves. It becomes deciding what deserves to be built, and showing that it worked.
Speed makes the wrong thing cheaper too
Andy, who leads product design and the AI Center of Excellence at Taxwell, described a seasonal tax business where timing is part of the value. An experience that misses filing season may not reach customers until the following year. So his team asks where customers are stuck and what can reach people while it still matters.
His designers now prototype in code. Getting an idea in front of customers used to take two to three weeks. Now it takes under three days. Because customers share financial details, he was clear that understanding people and earning trust have not changed.
Catherine Winfield, who leads product management and design for Digital Growth Experience at Autodesk, repeated a line she attributed to Claire Vaux: "It's never been easier to go faster in the wrong direction." Her fix is a hypothesis review in place of a design or product review. The question is whether the solution will solve a customer problem.
Design leadership becomes an ownership job
Catherine separated leading a strong design function from owning an organizational outcome. At the second level you think about how resources move across functions, not what your own team needs. She once recommended adding engineering capacity instead of design capacity because the designs could not otherwise reach customers.
Heather Winkle, VP of Digital Experience at Delta Air Lines, said her organization moved to shared KPIs instead of divisional ones. Andy talked about the harder personal shift: being willing to question work outside design, including technical decisions, because the organization's problem is the job.
Matthew Menz, formerly VP of CX at AWS, gave me a distinction I plan to borrow. Shared metrics, shared ownership, and accountability are three different things.
Judgment has to live in the system
Matthew argued that enterprise design has to move past reviews, redlines, and adoption campaigns toward mechanisms that scale. His version: make the easiest thing be the best thing.
He described design systems, instrumented components, analytics, and content working as one feedback loop, with a context layer that carries customer journeys across silos. He was clear this is a direction to build, not something enterprises have today.
Andy's team is working on the foundations: machine-readable design system foundations, brand and legal skills, and semantic layers with data partners. His reasoning is that cheaper building moves the bottleneck to deciding and judging, so every step needs evidence of value.
What this looks like in practice
Matthew's point is easy to nod along to and hard to measure, so here is what we see in customer evals at Knapsack.
A global pharmaceutical company ran 20 tasks with AI and no design system context. About 75% of the token values the AI used were invented or detached from the system, and roughly one in three components was made up or pulled from outside it. The work shipped fast. It also shipped wrong.
A consumer photo-products company ran 12 tasks, three trials each, with and without the system's context in the loop. Token fidelity went from 36 to 78, and cost per task dropped by about 60%.
The gap between those two results is the judgment Andy was asking about. Nobody had to review each screen. The system carried the decision.
What nobody solved
Andy asked the question I keep coming back to: "What value does that work actually create?" He wants speed metrics paired with decision metrics. Report efficiency alone and your function becomes a cost lever. Claim outcomes you do not control and you lose credibility.
The panel also left two questions open. What do you embed in enterprise tooling when the wrong practice scales as easily as the right one? And how do you build resilient teams when nobody knows the workforce model yet? Catherine's answer to the second was candor, psychological safety, and room to play.
The panel kept circling a point I see all the time. The tooling is rarely what stalls AI at scale. Who owns the outcome, what gets measured together, and how a team earns the right to scale something are org design and change management questions.
A safe place to test systems thinking, judgement, and the ownership mindset
That is why I am looking forward to UXDX San Francisco on November 5 and 6. Sam Alexander, Chris Strahl, and I are running "How to Keep AI-Generated Products Aligned with Your Design System" on Friday, November 6 at 1:15 pm PT at August Hall. It is a hands-on session on checking whether AI-generated interfaces follow a design system.
Workshops are included with your UXDX SF pass. If you are going, come find me. I want to hear what your organization is wrestling with, especially the change management and org design it takes to scale AI in a meaningful way.
