This isn’t about designers writing code.
It’s about designers directing AI agents in an environment that produces production-adjacent output.
The distinction matters. A designer using Cursor in a sandbox isn’t replacing an engineer. They’re collapsing the distance between intent and implementation, doing the translation work that used to happen across handoffs and revision cycles, before it becomes someone else’s problem. The environment for that is the sandbox repo. The reason it matters is the disappearing middle.
It’s also not a fixed workflow. The sandbox can integrate Figma MCP, third-party design tokens, or whatever tooling the team already uses. The infrastructure is the point, not the specific tools. My own loop runs through Cursor, Claude Code, and the MCP connections that put design context and issue trackers inside the same session. The tools are interchangeable. The contract with production isn’t.
What becomes more valuable
When the middle disappears, two things become more valuable, not less.
Forming the right intent. What should we build, and why? Who is this for? What does success look like? That judgment doesn’t get automated. Agents need it as input.
Evaluating the outcome. Does this solve the problem? Should we ship this? Is this actually better? That judgment closes the loop.
The work between those two moments compresses. The judgment at both ends expands.
Judgment doesn’t live inside prompts. It was built from years of navigating bad briefs, difficult clients, and contradictory constraints, and earned by shipping things and watching them fail in ways no one predicted. A model doesn’t train its way into that.
I earned a specific version of it by watching operators abandon a workflow in five days because we had executed a precise, wrong idea. That failure is more useful to me than any portfolio piece from the years when craft could cover for a weak call.
AI doesn’t eliminate the need for design expertise. It relocates it, somewhere less visible, more contested, and more important than it has been.
The irony is that it took automation to make that visible. For most of design’s history, judgment was bundled inside production. You couldn’t see it separately because it was always attached to something you could point to. Strip away the artifact, and what remains is the thing that was doing the actual work all along.
The sandbox is infrastructure for that judgment. It gives design teams a place to test intent against reality, fast, without waiting for an engineering sprint to find out whether the idea held up. That’s where the leverage is now.
Whether that counts as good news depends almost entirely on where you were planning to stand.
Also in April: The Disappearing Middle of Software Work · The Two Edges of Judgment · Designers Need a Sandbox Repo
Continues in May with About Craft, Taste, and Speed.
Daniel Arevalo is Product Experience Lead at Omny Security. He builds AI-native tools and writes about design, product leadership, and what it means to direct systems that execute at a speed humans never had before.