Last week, Carly's recap of UXDX NYC landed on the line "AI is an amplifier of whatever discipline you brought with you."
Here's mine from the engineering seat — three days, eight tracks, and a room that kept arguing about the same thing in different costumes: tension and what to do with it.
As mentioned in our last post, UXDX puts designers, engineers, and product leaders in the same room for three days. The May 2026 edition in New York put one question on every Main Stage backdrop: "AI can build anything. Can you decide what matters?"
A few themes stood out for me:
The bottleneck moved and so should we
Rory Madden, who founded UXDX, opened by naming what's underneath all of this. Software teams have been organized for decades around one assumption — that shipping is expensive, error-prone, and hard to repeat. That's why we sliced product work into functions: PMs to pick what to build, designers to shape it, engineers to deliver it, QA to catch the mistakes, project managers to keep the wheels on. Specialization made sense when execution was the bottleneck.
AI changed the cost. Execution is no longer the constraint. The bottleneck moved up the stack — from building the product right to building the right product.
According to Madden, the work AI is worst at — context, taste, framing the problem — is the work that just got more valuable, not less. The org chart we built around the old bottleneck is going to wobble.
In Goldratt's theory of constraints, the playbook for addressing a bottleneck is to identify it, exploit it and then subordinate the system to it – we're seeing this play out across the market in real time.
Agent Experience (AX) has joined the party with UX and DX
Dana Lawson, Netlify's CTO, gave the talk I came home thinking about. Her thesis up front: when she redesigned the Netlify platform for AI agents, "it did not just help the machines. I was approaching it like a true DevOps engineer, and it forced me to make everything better for humans." She calls the discipline Agent Experience (AX) — a peer discipline alongside UX and DX.
When an agent is the thing calling your API, deploying your code, or operating your dashboard, it is constantly asking which tool do I use? Is this context worth carrying forward? What should I do next? Most APIs are mute on those questions. Most admin panels are worse. Strong agentic tooling — Claude Code, Cursor, v0, Bolt, Lovable were Dana's examples — makes the answers explicit, and humans benefit from the same scaffolding.
The architectural shift she pointed at is real and uncomfortable. Traditional REST endpoints — post, URL, put this, get that — make agents reverse-engineer the sequence and the order. Agent-native systems expose intent-level operations instead: create a site, deploy repository, provision edge. And agents shouldn't have to poll for status; the system should emit events. Deploy started. Build failed. PR created. An agent watching a build pipeline shouldn't have to keep asking are we there yet?
Dana's punchline, which she earned: "Agent experience is just the human experience. When I thought I was doing this for agents, I was really doing it for you."
A related shift snuck into nearly every design-systems conversation after her talk: the component library is becoming rules, constraints, and infrastructure — not a shelf of pre-baked screens. If agents are generating the interface, the design system's job is to encode the guardrails the generation runs inside.
Workflow over features
Fahad Osmani runs design for Capital One's enterprise platforms, and the frame from his talk was all about the difference between building for a product and building for a platform.
"In enterprise platform design, we're much more concerned in a way about the floor rather than the ceiling."
As Fahad puts it — workflow beats features. Platforms organized around features rather than the user workflows leave seams that users trip over. This rhymes with the concept of stream-aligned teams in Skelton and Pais's Team Topologies — organize humans around the value stream, not around components.
His Jira story: "Everyone loves Jira ... but it turns out we were just speaking to the administrators." Then they watched the end users. "They hated it." The user of an enterprise platform is usually not the person paying for it. Design for the workflow of the person doing the work.
Two more things Osmani reinforced. Extensibility is a set of one-way doors — API shapes, schemas, terminology — and downstream teams live inside those doors for years - construct them carefully. And one of my favorites: the best innovations on enterprise platforms rarely come from the platform team itself; they come from business units solving an urgent problem, which the platform team then has to gather up and standardize.
A common theme under every debate: Tension and what to do with it
The many debates were the most interesting part of the conference, because they kept surfacing unresolved forces at play in many orgs:
- Donnie D'Amato (Design Systems House) vs. Alex Wilson (T. Rowe Price) on the future of design systems. Donnie: a design system is "an agreement," not a police force — "I'm letting you do whatever you want and I want to show you what's going on." Let tokens tell the truth about a system's inefficiencies; let the metrics carry the story. Alex, who works at a financial-services firm and self-identifies as "a control person," came back to trust: when a button drifts at the checkout, users hesitate at the credit-card form. The system should hold the line so that consistency does the trust work for you.
- Ellen Linardi (Synctera's Chief Product & Technology Officer) on whether product and engineering should report to one leader or two. Her case for unifying: combine the functions, kill the friction, sharpen focus. Michael Winnick (CEO, dscout) did an excellent job straw manning the opposing side, and the room was left with a clear sense of the tradeoffs.
- Dana Lawson (Netlify) vs. Christina Goldschmidt (VP, Product Design at Warner Music Group) on what AI is doing to roles. Dana: jobs are being lost, full stop, because cost-cutting agendas want them to be. The end result? the collapse of distinct roles into swiss-army knife job descriptions. Christina: roles are changing, not disappearing — and she brought April's hiring data to prove it. Product design and PM jobs were down a few points; design engineering, UX, and design ops jobs were each up by double digits. The shape is shifting, not shrinking.
My take on the theme — is tension a defect to be removed, or a feature to be stewarded?
Positive tension as a pattern
My take? Tension is a source of creativity and innovation. It's a signal that says: "there's an important decision to be made here." And the way in which you make that decision is your entire POV, strategy, and culture in practice. Like on the TV show Survivor, it’s all game theory until it comes time to vote someone off – when you have to decide, you see where you stand.
Christopher Alexander was a Berkeley architect who, in 1977, published A Pattern Language — a world-changing book about how to design buildings, towns, and rooms that people actually want to be in. His core claim was that a pattern is a resolution of forces in conflict — and the resolution honors the forces.
Honors them - it does not abate them. It does not appease them. It does not ignore or refute them.
Alexander's lifelong critique of modernist architecture was that it smoothed the competing forces away and produced dead places — wide flat plazas, identical buildings, places nobody wanted to be in.
Sound familiar? An LLM picks the most probable path — it collapses into the most statistically available continuation. Most likely. Least surprising. Predictably obvious. This is a fantastic feature of LLMs, but for design work, the hardest tensions have the least obvious resolutions.
Humans balance the tensions, AI executes within the balance.
And that's the shifting role that the product craftsperson – whether designer or developer – plays in the new world of UX/DX and AX.
