Design at the Edges
When agents absorb the middle of the workflow, what's left is what always mattered.
Last week I wrote that most teams have deployed AI but haven’t redesigned the factory. This week on the blog, I shared stories of a handful of teams that show what the redesign actually looks like. And they’re converging on the same answer about where design value lands.
Intercom’s design team published the numbers: 90% of their pull requests are now AI-authored. John Moriarty, product design director at Intercom, draws a clear line through what that means for designers. Design’s value concentrates at the edges: deciding what to build at the start, judging whether it’s good enough at the end. Agents own the middle, the build itself.
The same pattern showed up at Cisco, where one of Jason Cyr’s directors pointed Claude Code at their design system and got 44 detection detail panels in ten minutes, every decision tracing back to real customer research. The design system was the design review. Cyr’s sharper point is about what happens upstream: “When agents can generate ten options in an hour, the person who can look at all ten and say ‘none of these—here’s why’ becomes the most important person on the team.”
Inside Figma’s own team, designer Gui Seiz describes the cost of exploration collapsing. His team pulls production code into Figma via MCP, edits visually, pushes changes back to the codebase. Seiz spends more time now planning upstream and polishing craft downstream. The rushed middle phase—where designers used to race to get specs to engineering before priorities shifted—is the part that’s compressing.
Intercom, Cisco, and Figma run different stacks and build different products, but all of them are landing on the same conclusion: the work that matters is at the beginning and the end, and agents are taking over everything in between.
NN/G has a name for it. Sarah Gibbons and Huei-Hsin Wang call it process compression: what looks like “skipping the process” is really an experienced designer running an internalized version of it at speed. “The intuition designers trust was built by the very process they dismiss.” The double diamond didn’t die. Instead, it just got faster. But the compression only works if you’ve already done the reps.
And that’s where it breaks. The middle of the workflow is also where junior designers have always learned. The wireframes and component specs were never the deliverable. They were the mechanism through which designers built judgment. I wrote about this for Fast Company: construction figured out the pipeline problem a century ago with formal apprenticeships. The contractors I work with don’t debate whether to invest in training during a downturn. They know that if they stop training apprentices, they won’t have journeymen in four years. We’re hollowing out the middle of the workflow and the middle of the career ladder at the same time.
David Hoang proposes forward-deployed squads as one response: three people, embedded on the company’s hardest problems, building working prototypes instead of producing decks. The designer finds the problem and builds the first cut of the solution. That’s an edge job. It requires the kind of judgment that only comes from years of working in the middle.
At the far end of this spectrum sits StrongDM’s Software Factory, where humans write the roadmap while agents write, test, and ship the code. No human reviews the implementation. The expectation is that every engineer on the team spends $1,000 per day on AI tokens. The middle isn’t compressed there. It’s fully delegated.
Consider AI researcher Ethan Mollick’s observation: “We can see the shape of the Thing now, but we can still influence the Thing itself.” Design doesn’t have its version of this rulebook yet. For design teams on the frontier, the edges are becoming the whole job: direction at the start, judgment at the end, agents in between. The teams writing those rules now are setting precedent for everyone else. And the people best positioned to write them are the ones who spent years in the middle before it disappeared.
What I’m Consuming
The Curse of the Cursor. Alan Kay designed the original mouse pointer for the Xerox Alto by straightening one edge of a 16x16 pixel arrow to avoid jagged lines. That shape stuck through the Alto, Star, Lisa, Mac, and Windows. Marcin Wichary traces the full history, including Apple’s 2020 attempt to redesign the cursor for iPadOS and the quiet reversal five years later. The best observation comes from a commenter on Posy’s excellent companion video on cursor history: “I’ve never thought of the mouse cursor as an arrow. My mind was blown when I realized that it was just an arrow the whole time.” (Marcin Wichary / Unsung)
Consistency Is Primitive. Christopher Butler argues that when AI makes software creation nearly instant, the economic imperative for standardization disappears. We standardized software because building once and selling many times was the only model that worked, not because uniformity made for better experiences. When creation becomes individualized, software becomes bespoke because there’s no reason for it not to be. He takes the idea further than expected, connecting it to why sufficiently advanced technology might look wildly inconsistent rather than uniform. (Christopher Butler)
Strategic, Surgical, and Scrappy: A Career Mindset for Rough Terrain. Dee McCrorey revisits a Silicon Valley internship program she helped design 15 years ago and asks what it would look like today. Her answer: the skills needed to thrive alongside AI demand a total mindset shift, from orchestrating AI workflows to building scalable knowledge systems. She also takes a sharp look at Dario Amodei’s “AI jobs apocalypse” warning and asks why it took so long. (Dee McCrorey)
How Fast Will A.I. Agents Rip Through the Economy? Ezra Klein sits down with Anthropic co-founder and head of policy Jack Clark to talk about what happens now that AI agents can program autonomously. Clark argues we’ve crossed a threshold where the models that were always being promised are actually here, and the implications for labor markets, stock prices, and organizational structure are playing out in real time. Worth the full 90 minutes. (Ezra Klein / The New York Times)
There Are Only Two Paths Left for Software. I hate this but it’s a warning sign we should all pay attention to. David George at Andreessen Horowitz lays out a stark ultimatum for software CEOs: accelerate revenue growth by 10+ points through genuinely new AI-native products, or rebuild for 40%+ true operating margins including stock comp. No middle lane. His playbook for path one reads like a war plan: find the five people in your org who will deliver 100x value, put them on information-gathering sprints, then watch which VPs get on the bus. (David George / Andreessen Horowitz)