Where Does Excellence Live?

On Speed, Craft, and the Two Innovation Modes That Separate the Founders Worth Backing

Joerg with Claude, 3.3.2026

This piece started with a video interview. Jenny Wen, Head of Design at Anthropic, sat down in early 2026 to talk about how design works inside one of the world’s most capable AI companies. The headline observation was striking: the traditional design process — research, diverge, converge, mock, hand off — is effectively dead at the companies building fastest right now. Designers have gone from spending 60–70% of their time on mocking and prototyping to around 35%. The mock, Wen argued, has become a bottleneck rather than a blueprint.

I watched the interview, wrote a first version of this piece with Claude, and then stopped. Not because the observation was wrong — it isn’t. But because something felt missing. The essay celebrated speed and iteration in a way that made me uncomfortable, and I couldn’t immediately name why. So I sat with it, and in a conversation the next morning I found the question I’d been circling: where does excellence live in all of this?

What follows is the result of that dialogue — between my discomfort and the original framework, between my 17 years of watching founders build and the specific dynamics Wen describes from inside Anthropic. The piece that emerged is different from the one I first wrote. I think it’s sharper. And the process of writing it — a human investor sitting with an idea until the real question surfaced, then using AI to develop it further — is itself part of the argument.

The False Binary

The story Wen tells is seductive in its clarity. Engineering velocity has exploded. AI agents can deploy code faster than any design team can mock it. In that environment, the comprehensive Figma file that once represented professionalism now represents latency. Ship fast, iterate in production, build trust through responsiveness rather than polish. It’s a compelling frame — and for a certain kind of product, in a certain kind of market, it’s probably right.

But it contains a hidden assumption: that the path to excellence is always iterative. That if you ship enough and learn enough, you will eventually arrive somewhere worth being. And that’s not always true.

Some products don’t get iterated into existence. The iPhone wasn’t a response to user feedback requesting a touchscreen keyboard. It was a conviction — about what people would recognize as right once they held it in their hands, even though they’d never asked for it. Nobody riding horses in 1900 was requesting a car. The need didn’t exist in language yet. It had to be created — through design, through craft, through someone’s refusal to let what currently existed define what was possible.

The mistake isn’t celebrating iteration. It’s treating iteration as the universal mode. There are two fundamentally different innovation patterns, and they require different founder psychology, different team structures, and different evaluation criteria from us.

Two Modes of Innovation

The first mode I’d call discovery through iteration. The hypothesis is loose, the problem is real but not yet fully defined, and the fastest path to understanding is building, shipping, and watching what happens. In this mode, Wen’s framework is exactly right. The mock is the enemy. The 10-day sprint — build, ship, gather real feedback, decide — is the most efficient path to learning. Speed is a form of rigor.

The second mode is different. I’d call it vision through conviction. Here, the founder sees something the world doesn’t know it needs yet. The insight isn’t extractable from user feedback because users are answering from within the current paradigm, not the one being created. In this mode, iteration without a strong initial vision doesn’t accelerate toward excellence — it circles within mediocrity. You can iterate endlessly on a motive that isn’t strong enough at the root. The result is a better version of something that shouldn’t exist.

Apple in its best years operated in the second mode. So does the best Bauhaus-lineage furniture design — products so formally resolved that they look inevitable in retrospect, but required a designer to hold a conviction against market pressure long enough for the market to catch up. There are digital equivalents. Some of the most respected interfaces in the world — the ones that still feel right a decade later — weren’t iterated into existence. They were composed.

The important thing is that neither mode is inherently superior. The mistake is being locked into one when the other is required. The best founders can feel the difference — and switch.

In practice, the boundary between these modes is blurrier than any framework admits. The best founders don't neatly operate in one mode or the other — they switch mid-sprint, sometimes mid-conversation. A team might hold deep conviction about what the product should be while running rapid discovery on how to get there. Or they might start in pure iteration mode and find, three sprints in, that they've stumbled onto a conviction worth defending. The two modes aren't rooms you walk into and out of. They're more like frequencies you tune between — and the skill is knowing which signal to follow in any given moment. A clean binary makes for a better essay; messy fluency makes for a better company.

The Hard Evaluation Problem

This is where it gets practically difficult — and where I think the real value for us as investors lies.

The founder with genuine design conviction and the founder with a perfectionism problem can look identical from the outside. Both move slowly. Both push back on shipping before they’re ready. Both talk about quality in terms that sound visionary. The difference is internal: one has an earned intuition about what the product should be, developed through deep immersion in craft and the problem space. The other is optimizing for their own comfort, using the language of excellence to avoid the vulnerability of public failure.

There’s a mirror problem on the other side. The fast-iterating founder who ships constantly and talks fluently about learning loops can be genuinely excellent — or can be using velocity as camouflage for the absence of a real idea. Fast iteration without a strong original signal is just fast burning. It looks like progress until it doesn’t.

So the evaluation question isn’t “are they fast or slow?” It’s: do they understand which mode they’re in, and why? Can they explain what their design conviction is grounded in — not just as aesthetic preference, but as a specific theory about what users don’t know they need yet? And when they iterate, are they learning something specific, or just moving?

I find that the founders worth backing in the vision-conviction mode share a particular quality: they can walk you through their reasoning at the level of the craft decision. Not just “we believe in simplicity,” but “we made this interface element work this specific way because of what we observed about how people approach X, and we think that’s counterintuitive enough that users won’t request it, but right enough that they’ll recognize it.” That level of articulation is rare — and it’s the thing that distinguishes conviction from stubbornness.

Excellence in the Age of AI Collaboration

There’s a third dimension here that neither mode fully captures: how the best founders are learning to combine human judgment with AI capability — and how this changes what excellence looks like.

AI doesn’t choose between discovery and vision modes for you. It accelerates both. In discovery mode, it lets you build and test hypotheses faster than was previously possible — compressing the feedback loop dramatically. In vision mode, it lets you prototype the conviction faster, stress-test edge cases the founder might not have anticipated, and iterate on execution details without compromising the original design intent. Used well, AI preserves craft while accelerating exploration.

But this requires something AI can’t provide: the founder must know what they’re asking AI to build. They need enough design conviction to direct the tool, and enough humility to let reality talk back when they ship. This is the synthesis — not human versus machine, and not human replaced by machine, but human judgment operating at a higher bandwidth because AI handles the execution weight.

The founders who are doing this well right now have internalized something important: AI doesn’t lower the bar for excellence. It raises the stakes for judgment. When anyone can generate a functional product quickly, what differentiates is the quality of the vision directing that generation. Taste, conviction, and the ability to recognize excellence when you see it — these become more valuable, not less, as execution gets cheaper.

This piece is itself a small example. It started with Wen’s interview as a crystallization point, moved through a first draft, stalled on an uncomfortable feeling, surfaced the real question in a morning conversation, and then used AI to develop the argument further. The human brought the discomfort, the question, and the conviction about what mattered. The AI brought synthesis speed and structural pressure. Neither could have produced this alone. That’s not a footnote about process — it’s a preview of how excellent thinking gets done.

What This Means for How We Evaluate

Several of these observations translate into questions we can bring to founder conversations — and ways of listening to the answers.

Which mode are you in, and do you know why? This is the foundational question. A founder who can’t articulate why their approach to building matches the specific nature of their innovation is operating on instinct alone. That’s sometimes fine — but it’s fragile. The founders who can say “we’re in discovery mode because the problem space isn’t yet defined and users can’t tell us what they need” or “we’re in conviction mode because we have a specific theory about what exists that the market doesn’t know it wants yet” — they have the metacognitive clarity that survives hard moments.

What is your excellence grounded in? For vision-conviction founders, ask them to go one level deeper than aesthetics. Not “we care about design” but “here’s the specific thing we believe about how people experience this problem, and here’s why current solutions miss it.” The answer reveals whether the conviction is earned through immersion or borrowed from a category narrative.

What did you learn from the last thing you shipped? This question applies in both modes. In discovery mode, a strong answer shows that iteration is directional — that each ship taught them something specific that changed the next decision. In vision mode, a strong answer shows that conviction is being tested against reality, not protected from it. The founder who can’t answer this question is either not shipping or not learning.

How do you use AI in the work itself? Not as a productivity question — as a judgment question. The founders operating at the frontier aren’t using AI to replace design thinking. They’re using it to amplify it. What they direct AI to do, and what they insist on doing themselves, reveals the quality of their judgment about where human agency is irreplaceable.

The Shadow Side

A few honest caveats about how this framework can be misapplied.

The two-mode frame can be used to rationalize anything. A founder who isn’t shipping can claim to be in vision-conviction mode. A founder who’s spinning in circles can claim to be in discovery mode. The frame is a lens, not a verdict. What it does is give us a better set of questions — not a way to avoid the hard work of judgment.

Excellence in design is not the right bar for every company. Wen’s observations describe a specific context — a technically elite, well-funded AI company building consumer-facing products where design is part of the value proposition. For a B2B infrastructure founder, or a vertical SaaS team solving a workflow problem, a different kind of excellence applies. The question “where does excellence live?” has different answers depending on what the company is. Applying design-conviction criteria to every founding team is its own form of category error.

Survivorship bias runs through the conviction narrative. We remember the founders whose vision turned out to be right. We don’t remember the ones who held convictions just as strongly and were simply wrong. Conviction is not the same as correctness. The differentiator isn’t the confidence of the vision — it’s the quality of the reasoning behind it, and the founder’s willingness to update when reality contradicts the hypothesis.

Connection to HEARTFELT Thinking

Agency

The block-shaped builder Wen describes — someone performing at the 80th percentile across design, product, and engineering simultaneously — is an expression of agency. But agency here means something specific: the fluency to do whatever is needed, not the freedom to do whatever you want. The founder with genuine design conviction is exercising agency at the level of the vision. The fast iterator is exercising it at the level of execution. What we’re looking for is founders who have both — who are fluent across the full problem, not just their slice of it.

Timing

The compression of planning horizons that Wen describes — from 2–10 year visions to 3–6 month windows — is real, and it’s relevant. But timing doesn’t mean short-termism. It means being native to the leverage of this specific moment. In vision-conviction mode, timing means understanding that some groundbreaking ideas require the market to catch up, and that the founder’s job is to hold the vision long enough for that to happen — without becoming disconnected from whether it’s actually happening. In discovery mode, timing means knowing that the window to find product-market fit before the money runs out is short, and that learning speed is the only real currency.

Founder Quality

The deepest signal isn’t which mode a founder is in. It’s whether they have the judgment to know which mode is right for their problem, and the flexibility to switch when the moment changes. The iPhone required conviction. The first year of iteration after launch required speed. Jobs had both. The founders I’ve backed who went furthest shared that quality — not just vision, not just velocity, but the metacognitive awareness to know which one the situation demanded.

That’s what the question “where does excellence live?” is really asking. Not whether to optimize for craft or speed. But whether the founder has an honest theory of their own innovation — and whether that theory is sophisticated enough to hold both.

Source: Jenny Wen — The Evolution of Design in the Age of AI (Interview, 2026)

HEARTFELT Library — Selection Frameworks