Commitment is the willingness to hold a durable point of view about what you’re building and why. It’s the thing that gives individual decisions coherence over time. Without it, every choice is evaluated fresh, against whatever signals are loudest in the moment. With it, choices compound into a product that feels like someone is steering.
Commitment combines two capabilities. Sensibility: a deep, internalized understanding of what the product should feel like and how it should behave, the kind of understanding that lets a team member feel when something has gone off-track without needing a review. And conviction: the organizational willingness to maintain a direction despite uncertainty, competitive pressure, and the constant reasonable arguments to abandon it.
Teams with strong sensibility have specific words for what they care about. Not “clean” or “simple”; those words mean nothing because they could describe any product. They have language that functions as decision filters, specific enough that two people can apply them to a novel situation and arrive at the same conclusion.
A team with weak sensibility using AI will produce a lot of competent, forgettable work. A team with strong sensibility will produce work that feels like it could only have come from them. The difference isn’t in the tools. It’s in whether the team has a shared understanding of the product’s character that goes deep enough to govern output they didn’t individually review.
Sensibility is what makes design systems more than component libraries. The system encodes decisions about what the product is, and those encoded decisions hold the line even when no single person is checking every screen. As AI generates more of the surface area, that encoding becomes load-bearing in a way it never was before.
Commitment isn’t tested in principle. It’s tested in these exact moments. Nobody walks into a room and announces they’re abandoning the product philosophy. The erosion is gradual. It comes from arguments that sound like strategy, made by smart people with good data and legitimate concerns, framed as pragmatism.
Three arguments show up reliably. Each one appeals to something real. Each one gets more potent as execution gets cheaper, because the cost of acting on them drops while the frequency of encountering them goes up.
“Competitor X just shipped this. We need a response.”
When execution is cheap, competitors ship more, faster. A feature that would have taken a rival team a quarter now ships in weeks. The cadence of “they just launched something” accelerates, and the window where you feel pressure to respond shrinks. The cost of responding drops too, which makes each individual reaction feel low-stakes.
That’s the trap. One reactive feature is easy to justify. But reactive features compound. Three quarters of them and the product starts to feel like a composite of other products’ priorities, a patchwork of external responses wearing your brand. Users feel this even when they can’t name it. They experience it as a product that seems uncertain about what it’s trying to be.
Most of the time, this argument is a category error. It assumes that because a competitor made a choice, the choice was right, or that it’s right for your product. It replaces “what should we build?” with “what are they building?” and treats the answer to the second as if it settles the first.
The response that preserves commitment: acknowledge the signal, then run it through your own filter. Does this matter for what we’re building? Would we have built this if they hadn’t? If the answer is no, the information is useful but the action is wrong. That filter has to be fast now, because the signals come faster. The thesis has to be sharp enough to apply under pressure, not something you retrieve from a strategy doc once a quarter.
“The metrics say this variant wins.”
You can now run more experiments faster, across more variants, measuring more things. The risk has shifted from not having enough data to having so much that you can find a number to justify almost anything. The data-backed case against a principled design choice is easier to construct than it used to be, because somewhere in the results there’s a variant that outperforms on something measurable.
AI-powered optimization tools compound this. They can run continuous experiments without anyone explicitly choosing to test. The experimentation becomes ambient, the metrics accumulate, and the gravitational pull toward whatever converts best in the short term gets stronger without anyone making a conscious decision to follow it.
The trap is in the metric, not the data. Every test measures something specific, usually short-term: click rate, conversion, time on page. It tells you which variant performs better on that measurement over that window. It tells you nothing about long-term trust, coherence across the product, the cumulative effect on how people feel about your product after six months, or whether the winning variant is consistent with what you’re trying to build.
The version of this argument that preserves commitment: “The data shows a short-term preference for the flashier option, but our thesis is that long-term trust depends on users understanding the product, and the quieter design serves that better.” That’s data informing judgment. The version that erodes it: “The numbers say B, so we ship B.” The metric becomes a shield against having to hold a position. When the volume of data goes up, the temptation to outsource the decision to the numbers goes up with it.
“It’s good enough already.”
There’s an older version of this that most people recognize: “we’ll fix it later.” The next cycle rarely comes, the compromise becomes permanent, the bar drifts. That pattern still exists. But cheaper execution has introduced a more dangerous form.
When the generated output is 80% of the way there, clean and functional and within the general bounds of the design system, the question shifts. You’re no longer deferring quality to a future cycle. You’re accepting adequacy in the current one. The output passes review. It works. It’s close enough. “Why spend time refining something that already functions?” is harder to argue against than “we’ll come back to this,” because it doesn’t acknowledge a gap. There’s no deferred promise to hold anyone accountable to. The standard lowers and nobody experiences it as a decision because nobody made one. The generated output set the bar, and the team accepted it.
If you haven’t defined what “good” means with enough specificity to distinguish it from “adequate,” the AI’s output becomes your de facto standard. The system’s ceiling becomes your ceiling. And since the system optimizes for statistical likelihood, that ceiling is the average of everything it was trained on. Regression to the mean, delivered at scale, accepted by default.
The thing that separates real triage from quiet acceptance: triage has a specific plan, a specific owner, and a mechanism to reopen the conversation. Acceptance of adequacy doesn’t even have the language. It passes in silence.
Each argument appeals to a real value: market awareness, empirical rigor, execution quality. Each one, unchecked, substitutes an external signal for an internal conviction. The competitive argument substitutes their priorities for yours. The data argument substitutes a metric for a philosophy. The adequacy argument substitutes the system’s ceiling for your own.
Cheaper execution makes all three more potent simultaneously. The competitive signals come faster. The data volume is higher. The baseline quality of generated output is better than it used to be, which makes the gap between adequate and good harder to perceive and harder to justify closing. The moments where someone needs to say “no, this isn’t what we’re building” arrive more frequently and with less obvious justification each time.
Commitment means having a theory of your product clear enough to evaluate these arguments against at that pace. When someone raises the competitive signal, you can respond from your thesis. When someone cites the data, you can articulate what the measurement isn’t capturing. When the generated output is adequate but not right, you can name the gap and explain why it matters in terms the team can act on.
I’ve been talking about what commitment prevents. The drift. The flattening. Products that work and say nothing. That’s the defensive argument, and it matters for obvious reasons. But there’s something else happening that I keep thinking about.
For as long as I’ve been making things, the hardest part was never knowing what the product should be. It was having the time to get there. You’d have a clear picture, how the interaction should feel, how the system should behave, what the experience should be on the hundredth use, and then you’d ship a fraction of it. The calendar ate the rest. Details that would have made the thing feel yours got pushed to a quarter that never showed up. The best version lived in someone’s head and died in a sprint planning meeting.
That constraint is loosening, and it changes what’s possible in ways I don’t think most people have internalized yet. A small team can now reach surface area that used to take an entire org. Someone who carried a product idea for years but couldn’t write the code can build a working version this month. The distance between what you can imagine and what you can put in front of people has gotten smaller than at any point in my career.
The products I’ve respected most had a specific quality: everything in them felt decided. Not optimized. Decided. Someone had an opinion about the weight of a transition. Someone wrote the empty-state message as if an actual person would read it on a bad day. Someone made an architectural choice about how information was structured that you’d never consciously notice, but you’d feel its absence if they hadn’t made it. The accumulated weight of a few hundred choices like that is what gives a product its character.
The people who make those choices have always been around. What they lacked was runway. Every cycle was a negotiation between the vision and the deadline, and most of the time the deadline won, not because anyone wanted it to, but because the cost of execution forced a constant triage of ambition.
That triage is easing up, though unevenly and with a catch worth naming. The same tools that give time back to people with a clear point of view also give infinite output to people without one. The acceleration doesn’t care what’s behind it. It amplifies whatever’s already there, which is the whole argument for doing the depth work first.
But for the teams that did, the ones who figured out what they believe, built systems to hold it, developed the language and the standards and the shared sense of what their product should feel like, something genuinely new is available. The limiting factor on what they can make is, for the first time, the quality of their ideas rather than the cost of expressing them.
I want to be clear about this because the hopeful version of this argument can sound naive. I’m not describing a world where tools automatically produce better work. I’m describing a specific condition: people who invested in recognition and commitment before the tools arrived are now able to operate at a level of ambition that wasn’t reachable before. The ceiling will be raised by them.
Everyone else gets more of the same at a higher volume.
That gap, between the teams that did the depth work and the teams that didn’t, is going to become the most visible dividing line in product quality over the next few years. You’ll see it in the difference between products that feel coherent across every surface and products that feel like a collection of generated parts. Between brands that hold together under scrutiny and brands that dissolve the moment you look closely. Between experiences that reward sustained attention and experiences that are forgotten the moment you close the tab.
The tools made the bet on depth worth more, not less. That’s the part of this moment worth building toward.