Max Schoening, the product & design lead at Notion, dropped this line in an interview, and it's a short sentence that lands uncomfortably. In an era flooded with AI coding tools and stories about anyone spinning up an app from a few lines of prompting, he argues that the quantity of output and the quality of output have started to come apart. The moment a tool takes over execution, the question becomes: what's left for the human?"Vibe coding gave us more software. It didn't give us better software."
This question isn't just for developers. If you run a business alone or lead projects with a small team, the core skill is no longer how well you operate AI tools — it's what you're able to ask when you're standing in front of what those tools produce. The problem isn't how many tools you have. It's the posture of the person holding them.
Notion's mantra: "Drive it like it's stolen"
In an interview for Lenny Rachitsky's newsletter, Schoening described Notion's internal product culture with a striking phrase: "drive it like it's stolen." The expression floats around the startup world from time to time, but Notion uses it in a slightly different sense. It doesn't mean go fast. It means take the resources and authority you've been given and use them to the fullest — make your own calls and move. In other words, rather than waiting for instructions to come down from above, you define the problem yourself and chart the path to execution. That posture is what he calls agency.
The core phenomenon he points to goes like this. As AI tools become ubiquitous, the skill gap is closing fast. With a little prompt engineering, a junior developer, a junior designer, or a junior copywriter can produce far better work than before. But at exactly that moment, something strange happens: the volume of output goes up while the ceiling on quality tends to drop at the same time. Vibe coding is the prime example. The AI takes a request and writes the code; the human checks whether it runs. But "it runs" and "it's good code" are entirely different judgments. An AI can verify whether something works. Whether it's maintainable, whether it scales, what it does to the user experience — those remain squarely in the human domain.
This is where Schoening expands the definition of agency. Agency isn't simply having a proactive personality. It's "defining the right problem and taking responsibility for it." The more AI handles execution, the more sharply humans need to judge two things: what should be built, and whether this output is good enough.
When skills get commoditized, attitude is what's left
If you break competence down into skill, knowledge, and attitude, the areas AI replaces first — and fastest — are skill and knowledge. You can write code without memorizing a language's syntax, and you can receive distilled insights without researching a field yourself. Most people can already feel how rapidly those two are being commoditized.
What about attitude? Attitude is a far more stubborn territory. AI won't teach you to ask, "Should this be better?" Faced with an unfinished result, whether you shrug and say "good enough" or hold onto the nagging sense that something's off and dig in — that decision is still yours. The agency Schoening describes is built through the repetition of exactly that judgment.
This is where the vibe coding phenomenon becomes meaningful again. When people gain the ability to make more, watch what they actually do and they split into two camps. One camp ships whatever the AI produced and moves on to the next task. The other treats the AI's output as a starting point — fixing what falls short and layering in intent. The first camp may produce more. But over time, the second camp's work is what survives.
This isn't a difference between lazy and diligent. It's the difference between someone who habitually defines problems for themselves and someone who only moves within the frame the tool presents. And the gap widens as AI tools improve — because the better the tool gets, the more valuable the person who knows what to ask of it becomes.
The message Schoening reportedly emphasizes to Notion's team — "cultivate agency" — comes from the same place. Building the posture of judgment matters more than acquiring skills. The person who updates their skill list every time a new AI tool ships is always slightly behind. But the person with the habit of asking "does this output actually solve the problem?" stays useful no matter how the tools change.
What this actually means if you work alone
For people who work solo or in very small teams, agency isn't optional — it's a survival condition. With no team, there's no one to delegate judgment to. Yet paradoxically, as AI tools multiply, more and more people are losing their grip on it. They follow the direction the tool suggests, ship whatever comes out, and try the next tool. They've gotten faster, but they've never once chosen the direction themselves.
It helps to apply three questions to whatever you're working on right now.
When the AI produces something, check whether you're asking "is this what I want?" That's a different question from "is this what I asked for?" Requests and intentions often don't match. Refining your prompts matters, but the habit of looking at the output afterward and asking whether it solves the real problem — that's where agency begins.
Next, when choosing a tool, distinguish between "does this tool assist my judgment, or replace it?" A tool that assists your judgment helps you reach conclusions faster and more accurately. A tool that replaces your judgment lets you skip reaching a conclusion at all. The more comfortable the latter feels, the more likely your judgment muscles are quietly atrophying.
Finally, spell out "which part of this work am I accountable for?" This means separating what you delegated to the AI from what you decided yourself. For a solo founder, a brand is an accumulation of judgments. What you chose to make, what you turned down, which direction you revised toward — those pile up into how customers perceive you. If an AI made those judgments for you, your brand gradually converges toward an average that belongs to no one.
The agency Schoening urges people to develop is not the same as learning a particular tool or adding a particular skill set. It's the repeated practice of feeling "this problem is mine," judging "is this output good enough," and taking the stance that "the choice here is mine to make." And that practice holds up when you're not using AI tools at all — and it will still hold up after AI tools have changed beyond recognition.
The more AI tools take over execution, the more deciding what to execute becomes the one job left to humans. A year from now, the quality of your work will be determined less by how skillfully you wield the tools than by the posture you take when you stand in front of them.
