I've been thinking a lot lately about what actually survives The AI Compression as I’m calling it: wherein design, product, and engineering all become compressed into a single-ish role…and design-to-execution becomes compressed even further. Specifically, what parts of design work are genuinely irreplaceable versus what we've just been assuming were?
As it turns out, the answer is clarifying. And…just a little uncomfortable.
This week's piece is about where design leadership is actually heading — not the anxious "will AI take my job" version of that question, but hopefully the more useful one: what does the job look like when the executional layer is mostly automated? I am trying to gain a clear point of view on this, and it's likely not the one most teams are hiring for right now.
— Justin Lokitz
Design Deep Dive
Design Leadership is Shifting to Where It Was Always Supposed to Be.
I have (obviously) been watching the AI/agentic design space very closely. As an optimist, what I am seeing with tools like Claude Design, Google Stitch, and Figma Design Generator is pretty amazing. And…as a longtime educator and design leader, the shifts the design industry (mostly in tech) is making is, how should I say, daunting, to say the least.
For instance, as part of a recent prototype I’m (slowly) building, I watched both Claude Design and Google Stitch design an entire onboarding flow in minutes. My initial reaction was: “holy shit!” But…upon reflection, at least in the case of Google Stitch, I realized the agent had no idea why the flow existed, and the output proved it. Technically clean, yet strategically incoherent.
Prototyping in Claude Design
That gap — between competent execution and encoded intent — is where design leadership now lives.
As you likely know by now, the executional layer is compressing fast. Pixel work, UI iteration, production handoffs: AI handles these well enough that optimizing for speed there is very likely the wrong bet. What remains is harder to automate, and frankly, more important is deciding what a product should feel like to a user who doesn't know what they want yet…and making sure every agent, every generated variant, every automated touchpoint honors that decision consistently. That is not a Figma skill, full stop. That, my friends and colleagues, is a judgment skill.
The Three Problems Worth Solving
The design leaders I notice staying relevant right now are focused on three things.
Agent orchestration. When AI generates twenty variations of a flow, someone has to decide which signals matter, where human review is non-negotiable, and how automated processes hand off without losing coherence. That's a design problem, not an engineering one.
Trust architecture. There are moments in any product where a user needs to feel certain way before they submit payment, before they share sensitive data, before they commit to something irreversible. AI doesn't intuit those moments. A designer has to locate them explicitly and build for them on purpose.
Behavioral consistency. Products that use AI-generated interfaces are already showing a specific failure mode: the experience stops feeling like a coherent thing. Tone shifts. Interactions contradict each other. Memory of who the user is degrades between sessions. Fixing this requires someone who can specify a product's identity precisely enough that agents can maintain it. That's a new kind of design brief.
On the Tools
What’s more, if AI agents are going to generate interfaces at scale, they need a source of truth for design intent, not just design components. That’s where some of the newest foundational design tools, like Open Design, Design Desk, and the DESIGN.md format, come in. If you know your design history, traditional design systems documented what things looked like…or are supposed to look like. The tools above go one (very human) step further: they want to document why decisions were made.
I think DESIGN.md is the most interesting to me because it treats design intent as a file that lives in the repository, versioned alongside code. The premise is that intent should be as legible to an agent as an API contract. Whether that specific format wins is less important than the principle behind it: design thinking has to become machine-readable, or it gets ignored entirely.
What none of these tools solve yet is evaluation. How do you know if an AI-generated output is honoring the intent you specified? That's still a human judgment call, and it probably will be for a while. With that in mind, the designers who matter most right now aren't the fastest generators; they're the best evaluators.
What to Actually Develop
The counterargument to all of this is that most product teams still need execution, and they need it now. That's as true as it’s ever been. It's also a reason to stop training designers primarily for execution, not a reason to keep doing it.
Simply put: the skills that compound over the next three years are product intuition, systems thinking, and the ability to articulate design decisions with enough precision that a non-human process can act on them. Can a designer on your team explain not just what they decided but why, in terms a constraint-checking agent could verify? Can they identify where trust breaks down in a user journey before users do? Can they override an AI output that's technically fine but wrong in ways that don't show up in a heuristic?
Those aren't new skills. They're the skills design has always claimed to value and rarely hired for, because execution was faster to evaluate in a portfolio review.
If you think about it, AI just made the trade-off visible.
The Actual Shift
Design leadership was always supposed to be about intent, judgment, and systems. For a long time, the discipline got distracted by tooling; by the craft of making things look good and the speed of making them look good faster. AI is compressing that layer whether the field is ready or not.
The designers who thrive won't be the ones who adapted to AI quickest. They'll be the ones who knew what they were doing well enough to specify it, and were honest enough to say when the machine got it wrong. That's not a new job description. It's the original one, finally stripped of everything that was covering it up.
Subscribe to Design Shift for more conversations that help creative professionals grow into strategic leaders.
What did you think of this week's issue?
