When Software Gets Easy to Make, Design Gets Harder to Fake

A product idea used to live in a notebook for a while. Turning it into something people could click, criticize, and revise meant finding an engineer, waiting for a sprint, and hoping the first prototype arrived before the thought went cold. Now a designer can hand a rough prompt to an agent and have a working first pass before the coffee does.

That should make design feel easier. It doesn’t, at least not for everyone. In Lenny’s survey of the tech workforce, designers and user researchers emerged as some of the most anxious, tired, and least likely to recommend their profession to newcomers. Ian Silber, OpenAI’s Head of Design, doesn’t dismiss that feeling. The rules are moving. Nobody can neatly say what a designer is expected to know now: Should they ship code every day? Rebuild their entire process? Compete with colleagues who happen to be quicker with agents?

The unsettling part isn’t that AI has already replaced design. It’s that making something has become startlingly cheap. The ability to turn a fuzzy idea into a screen, a flow, or an interaction used to be a meaningful barrier. That barrier is falling. A product manager with a coding agent, a founder with a little front-end knowledge, and a designer with a good prompt can all produce something that looks like software.

So the question gets tougher: when nearly anyone can generate an interface, who decides what deserves to exist?

Prototypes got faster. Judgment didn’t.

AI’s leverage in engineering can be obvious. A piece of code runs or it doesn’t. A bug is fixed or it isn’t. Plenty of tasks have a reasonably clear finish line. Design rarely works that way. You think a direction is strong; you make it and it feels wrong. You revise it; users still don’t understand it. You show it to the team and discover that nobody agreed on the problem in the first place. Many good-looking ideas need to be thrown away.

AI can increase the speed of those attempts, but it has not cancelled the work. It can produce five layouts, connect a flow, and make a clickable demo in minutes. It does not inherently know where a user is stuck, which visual hierarchy is clear, when an interaction should happen, or whether a feature should exist at all. Someone still has to own those calls.

That is why Silber can make the apparently counterintuitive case that this may be the best time in history to be a designer. Not because designers no longer have to learn, but because their expressive range is larger. A product hunch at midnight does not have to wait until next week’s review to take shape. By morning, it can be a thing the team can argue with, reject, or improve.

That matters because it shifts more of a designer’s time toward the expensive work: seeing the real problem, making trade-offs, organizing feedback, and building systems that other people can actually use.

The simpler the door, the harder the design behind it

AI products make this problem unusually visible. ChatGPT serves a range of people that barely fits in one product: someone looking for a dinner idea, someone drafting an email, someone exploring a new hairstyle, someone running a complicated workflow with Codex, desktop software, or work tools. Put every option in front of everyone and the product becomes noisy. Hide everything and experienced users feel trapped.

Silber describes a direction he calls a “universal input.” A person should be able to state what they need; the system should work out whether the moment calls for a quick answer, a deeper conversation, or an action taken on the user’s behalf. That is not a matter of cramming more buttons into a chat box. It is the work of arranging models, tools, context, inputs, and outputs into a path that feels natural.

Chat will remain useful, but it needn’t be the permanent shape of everything. Voice can become more natural. Results can become more visual and more actionable. Reusable, editable, collaborative workflows can replace the ritual of beginning every task from a blank conversation. Turning “help me write an email” into a block that can be edited, copied, and carried forward is a small but meaningful change: the product is no longer merely answering; it is helping with what comes next.

The challenge is not inventing one dazzling feature. It is making the default experience quiet enough for a newcomer while leaving a real path for someone doing serious work. The designer’s job is to keep one door open without forcing every visitor down the same corridor.

Roles won’t vanish. Accountability will move.

Will product managers, designers, and engineers become the same job? It is the standard question whenever AI changes how teams work. A more realistic answer is that the borders will blur, but responsibility will not evaporate.

Designers will prototype more and understand technical constraints more directly. Engineers will participate earlier in product judgment. Product managers may make the first version of an experience themselves. Small teams may change their mix of talent, rewarding people who can connect concept, prototype, and exploration. Yet in a complicated organization, someone still has to be accountable for direction, experience quality, and system reliability. “Everyone does everything” often just means the difficult decisions arrive late and unowned.

That is why the goal for a designer is not to cosplay as a full-stack engineer as quickly as possible. A better move is to bring AI into the work they already do: exploring product directions, making rough ideas visible, preparing context for research or review, and surfacing the work a team needs to follow up on. The point is not performing an impressive prompt. The point is whether the tool makes a judgment happen earlier and with more evidence.

Don’t pretend the new rules are settled

The most exhausting part of this transition may be the performance around it. New tools are often discussed as though they have already produced a stable professional playbook. Silber’s reminder is more useful: we are still early. An agent that failed yesterday may suddenly work next month. A workflow that looks advanced today may feel clumsy in a year.

That is not a pep talk. It means nobody has an answer key that will stay valid. The danger for designers is not missing one tool. It is becoming so afraid of looking behind that they stop getting their hands dirty: they stop trying bad ideas, stop showing unfinished work to users, and stop admitting that a new interaction pattern has not made sense to them yet.

When Silber talks about the turn from IGTV to Reels, he does not dress failure up as a legend. Product work is usually a sequence of wrong assumptions, ugly first versions, user pushback, and another iteration. AI can make that loop run faster. It cannot make it graceful.

Software will get easier to make. The hard part is still sitting at the desk: deleting what does not belong, preserving what matters, and making something a person wants to open a second time.

Source

• Lenny’s Podcast: OpenAI’s Head of Design: This is the best time in history to be a designer | Ian Silber