Context

Why this matters

Lenny's own sentiment survey of the tech workforce found designers and user researchers report the worst scores of any role, most overwhelmed, least optimistic, least likely to recommend their own role. Ian Silber, Head of Product Design at OpenAI and previously eight years at Instagram, has a structural explanation for why design is lagging engineering's AI gains, and a specific answer for where humans stay valuable.

The Big Idea

AI hasn't sped up design the way it's sped up engineering, because the design process still runs on the same messy loop of trying things, throwing them out, and getting real feedback, and AI hasn't compressed that loop the way it's compressed a coding agent's binary "does it work" check.

Silber's counter to the anxiety this creates: because the field is genuinely this early, starting today still gives anyone a real head start.

Key Insights

AI's productivity gain skipped design

Coding agents turn a binary task, does this code work, into something AI can iterate on and verify itself. Design has no equivalent check: you still have to try an idea, get honest feedback, and often throw it out and start again, a loop that hasn't compressed the way code generation has.

AI is incredible, not yet best

AI already produces real design work and is accessible to anyone, but Silber says it isn't yet best at visual design, typography, information hierarchy, or interaction design specifically. That's a more useful claim than "AI can't design" or "AI replaced designers," and one a PM can actually act on when deciding where to route work.

New design has no training data

Multitouch on the iPhone and Snapchat's ephemeral messaging didn't exist before someone designed them, so there was no prior art for a model to learn from. AI is trained on what already exists; a genuinely new interaction has no precedent in that data by definition, which is why noticing an unmet need and inventing something new stays a human capability.

Systems thinking is now core

Silber describes designing composable building blocks, citing Notion's approach, rather than shipping one-off features, because AI-driven engineering teams now ship faster than an ad hoc design process can track. Lenny notes this is the fifth podcast in a row where systems thinking has come up independently, a signal both treat as real, not coincidence.

OpenAI builds in capability overhang

Most of ChatGPT's user base touches only a sliver of what it can do, an intentional choice: a simple default for most people, full capability available underneath for those who go looking. This contradicts a common PM instinct to expose everything to justify the investment behind it.

Mental Models & Frameworks

Capability overhang

The gap between what a product can do and what most users actually touch. Build a simple default for the median user and keep full depth available underneath, rather than surfacing everything to everyone. Silber's example: a ChatGPT user asking for a recipe sits alongside one automating an entire farm, and OpenAI serves both from the same simple default plus deeper power underneath.

Pick your durable battles

For the few surfaces you're confident will still matter in six months, Silber's example is the core chat composer, run the full research-prototype-test-iterate cycle. For everything else, the much larger surface still likely to change, move fast, try many things, and expect to throw most of it away without regret.

Trade-offs & Nuance

Roles blurring, not disappearing

Pure generalists work well at a startup. At larger scale, Silber argues you still need someone accountable for direction (PM), experience (design), and system integrity (engineering); doing all three well gets harder as an org grows, even as the underlying skills increasingly overlap. His resolution: track responsibility, not job title, since skills converge faster than responsibilities do.

Practical Application

Prototype in an agent first

Silber's own workflow: when an idea occurs to him, he puts it directly into an agent for a rough prototype before taking it further, a default first step, not an occasional shortcut.

Separate durable from exploratory work

Decide out loud, as a team, whether a feature belongs in the small set worth the full rigorous process or the larger set worth shipping fast, and revisit that classification as the technology changes.

Audit for accidental capability overhang

Check whether you're exposing everything to prove the value of what you built, when a simpler default with real depth underneath would serve most users better.

Hire specialists, not one generalist

Silber hasn't found one designer who covers visual craft, prototyping, and strategic thinking equally well. Hire a well-rounded mix of specialists rather than holding out for generalist perfection.

Questions to Consider

• Where does our product's process lack a clear yes-or-no success check, the way code either works or doesn't, that would let AI iterate on it directly instead of needing a human to judge the result?
• Are we exposing every product capability just to justify the engineering effort behind it, instead of defaulting most users to a simple experience with real depth available underneath for the people who go looking?
• Which of our features are stable enough that they deserve the full research-prototype-test-iterate design process, and which ones are we giving that same rigor to even though they're likely to change soon?
• If someone on our team feels behind on AI tools, is that a real skills gap, or a structural mismatch between how their function's work happens and how AI actually helps, the way Silber describes for design versus engineering?

Bottom Line

Nobody is actually ahead right now. The field is early enough that starting today still gives you a real head start, and the anxiety many designers feel is a rational response to genuine structural ambiguity, not a sign they're falling behind.

Focus on the outcome you're producing, not the tooling around it.

Case Studies Mentioned

Instagram's IGTV to Reels

IGTV, Instagram's long-form vertical video product, flopped, and Silber attributes it to the team holding onto wrong assumptions too long. The team reacted, changed course, and later shipped Reels successfully. The lesson was less the initial mistake and more the speed and honesty of the recovery, "how you react and fix forward" mattered more than getting it right the first time.

People to Follow

Kevin Weil

Former Chief Product Officer at OpenAI, quoted twice: for "the model we have today is the worst the model will ever be," and for the observation that natural conversation works as an interface because humans already communicate across a huge range of intelligence levels in everyday life.

Resources Mentioned

Notable Quotes

"We're all extremely early in this process. Nobody should feel behind right now. If you literally started today, you're going to have a leg up on pretty much most people." (Ian Silber)

"The model we have today is the worst the model will ever be." (Kevin Weil, former OpenAI CPO)

"We're all figuring this out as we go. Nobody has some perfect answer about how the design process should work." (Ian Silber)