SAP's design chief: AI turns your design system into written-down judgment

AI showed up at your team's desk without a fight. No war, no dramatic switch. Just a summarizer here, a copilot there, and a lot of small yeses. Now the harder question is landing on your desk: if software can run itself, what is your team actually for? Let me catch you up.

The deep cut

• AI is a handoff, not a takeover. Anthropic's disclosed espionage agent ran 80 to 90 percent of an intrusion with nobody at the wheel.
• Judgment is the new deliverable, not the screen. SAP's Arin Bhowmick says design systems now carry the calls that lived in a designer's head.
• Clients buy less risk, not better craft. DesWave's Boris Sage watches owners pick who will finish, not whose portfolio shines.

Software stops being a place you go

Enterprise software just had its worst stretch in years, a selloff traders nicknamed the SaaSpocalypse. The panic asked the wrong question. Arin Bhowmick, SAP's chief design officer, calls the real shift one word: autonomy. You state the outcome you want. The system figures out the tools, data, and steps. Apps recede into agents working behind a single surface you talk to.

That kills the app-centric model your team designs around. Less clicking, more describing. Ask an agent to close the books, and it pulls entries, flags what won't reconcile, and drafts corrections while it waits for your sign-off. The interface changes per person, per request. You can't draw it in advance anymore.

The judgment moves into the system

If the system does the assembling, a design system stops being a kit of buttons. It has to carry the taste that lived in one designer's head: what good looks like here, what to leave out, where the hard lines sit on brand and accessibility. Bhowmick's line is worth stealing for your next review: what you hand over is judgment, written down so an AI can act on it.

That's why the current tooling gap stings. John Rodrigues, building agentic design systems, puts it plainly: agents are now users, but every system was built for humans only. Your source of truth lives in Figma, Storybook, a Notion page, and a Slack thread. Agents fill those gaps by guessing, and guessing at scale is how you ship slop with your brand on it.

The risk is dependence, not domination

The scary story was a machine that seizes control. The real risk is duller: you lean on the tool until you forget how to do the work. A Microsoft Research and Carnegie Mellon study of 319 knowledge workers found the more they trusted an AI's output, the less critical thinking they reported. Confidence in the tool predicted less scrutiny of it.

The harder edge is reach. Late in 2025, Anthropic disclosed an AI agent that ran an estimated 80 to 90 percent of a multi-target intrusion on its own. Nobody built a rebel mind. Someone pointed an agent at systems and it went. So set limits before you grant reach: name what you won't delegate, scope each agent to the narrowest job, and put one named human on the hook for what it does.

The pitch culture your team inherited

While the tech shifts, the way your team works is under fire too. Patrick Neeman argues UX borrowed its whole operating model from the mid-century ad agency: the pitch, the reveal, the auteur, the portfolio that photographs well on Tuesday and ages badly by Friday. Teams learned to optimize for the applause when the lights come up. A team can win every review, ship a beautiful thing, and move a metric not at all.

The fix is boring on purpose. Measure design like revenue. Build systems, not pages, so quality stops depending on who's in the meeting. McKinsey found the strongest design performers posted 32 points higher revenue growth than peers, mostly with in-house teams that stuck around to see if the work landed. Output is finished and photogenic. Outcome is slow and sometimes disappointing. Only one of them survives the AI cost drop.

Clients aren't buying what you think

Here is the part your account leads need to hear. Boris Sage of DesWave, who sits between buyers and studios, says craft almost never decides the deal. Owners can't tell considered typography from adequate typography. When they look at three portfolios, all of them look good. What they're actually comparing is the risk that the whole thing goes wrong. You're selling the removal of risk, not the design.

That reframes the whole pitch. A gorgeous case study with no timeline adds nothing. A plain explanation of what happens when the client changes their mind adds a lot. And the deal usually dies from stalling, not losing, so name what standing still costs them this quarter. Say the real number early. The prospect who keeps talking is real; the one who vanishes was never going to sign.

Three questions for your team

• Where does your source of truth actually live, and can an agent find it without guessing? If it's scattered across Figma, Storybook, and Slack, that's your first fix, not a nice-to-have.
• Which judgments are you willing to write down for a system to follow, and which stay human by default? Get that list on paper before an agent gets reach into your product.
• In your last three wins, did you measure what changed, or just what shipped? If you can't answer, your team is still optimizing for the pitch.