When the middle compresses, what’s left?
The front edge: what actually needs to be built, what problem is being solved, and shaping the work clearly enough that an agent can act on it, and fail usefully when the direction turns out to be wrong.
The back edge: whether the output solved the problem, whether the thing that got built should ship, and whether to keep going or change course.
These edges require judgment. Agents don’t have it. That’s the practical consequence of The Disappearing Middle of Software Work.
What agents have is execution speed. They build exactly what you ask for. They don’t question whether it’s the right thing. They don’t notice that three feature requests are really one underlying user need. They don’t push back when the brief is solving for the wrong metric.
The shift underway is from producers to directors. Instead of asking how do I accomplish the goal?, the question becomes what goals do I want to accomplish, and how do I delegate them to AI?
The front edge is the work that now determines everything downstream.
Three positions that hold
Not every designer will reposition the same way. But three orientations seem defensible in this environment.
The first is direct integration. Use AI in production workflows to increase output and exploratory capacity, while keeping clear responsibility for the decisions that shape the work. The designer as a highly leveraged director of AI execution. This is the closest to how I work now, prototyping in Cursor and Claude Code against real operator workflows rather than handing off a file.
The second is a pivot toward interpretation, coordination, and evaluation over artifact generation. Less time producing screens. More time synthesizing research, pressure-testing directions, and evaluating whether outputs solve the right problem.
The third is anchoring practice in domains where the cost of error is high enough that automated output can’t be accepted without expert oversight: healthcare, financial services, critical infrastructure. Anywhere the failure mode is serious, judgment isn’t just valuable. It’s mandatory.
That third position is the one my career kept walking toward, from electronics and industrial automation into IT/OT cybersecurity. In that world a wrong workflow is more than a support ticket, and I don’t get to treat agent output as done because it compiled.
Each of these is the same move: shifting professional value away from production and toward judgment.
That sounds like an upgrade, until you remember the profession spent decades building its identity around production.
Also in April: The Disappearing Middle of Software Work · Designers Need a Sandbox Repo · Judgment Is the Territory
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.