The question keeps coming up. At dinners, in DMs, on calls that were supposed to be about something else. Should I specialize more? Should I widen? Is it safer to go deep on one thing now that the work is drying up, or safer to pick up adjacent skills so at least one of them survives the next wave? The question arrives with a certain amount of panic folded inside it, though the person asking almost always pretends otherwise.
The reason the debate feels unresolvable is that it got settled two years ago and nobody announced it.
Specialization, the way it has been sold for the last twenty years, was an industrial idea. Pick a discipline. Get excellent at it. The market will pay a premium for depth because depth is expensive to acquire. The logic rested on the assumption that deep skill was scarce, which was true when a skill required a decade of quiet practice in a single room. Copywriters, illustrators, developers, researchers, editors, photographers: each profession priced itself around the time it took to become good enough that the output was obviously better than the commodity version.
That assumption broke in late 2023. It broke quietly at first, through GPT-4 writing copy that was almost as good as a mid-tier copywriter. By mid-2025 it had broken loudly. Cursor and Claude Code writing software fast enough that half the industry had to rethink its pricing. Midjourney, Runway, and Sora collapsing the middle tier of image and video production. Research summarization agents moving into the slide decks of firms that used to bill four hundred an hour for the same work. The middle of the skill distribution in almost every creative and knowledge profession stopped being worth what it used to be. The skill itself was still there. The cost of producing the output had collapsed, and once the cost of an output falls close to zero, nobody pays a premium for the depth that used to be required to reach it.
Specialization is now a bet against automation. The bet sometimes pays off. A surgeon will be fine. So will a senior architect, or a technically excellent glassblower. The thousands of professionals whose specialization was the middle layer of a knowledge field are watching their value compress in real time. Nobody is happy to say this out loud, but everyone paying attention already knows.
The generalist option has broken too. The idea worked when crossing disciplines required human time. Learning a new tool took months. Picking up an adjacent skill took years. A generalist accumulated range through effort, and the effort was itself defensible. AI flattened that. Today, an operator with good taste and good prompting can stumble into design work, code, research, and video editing in the same afternoon. The range is no longer scarce. It is the new baseline. A generalist with no organizing principle is now a person with a lot of tools and nothing to build.
So the specialist-versus-generalist debate is not being resolved. It is being retired. Both positions assumed a labor market that no longer exists. The debate keeps running because the people who built their careers inside the old frame do not want to accept that the frame is gone. We get it. We have spent most of our own working lives inside that frame too.
What comes next is not a new category of worker. It is a different way of being one.
The operators thriving right now are doing something we have been calling, in our own writing, running an architecture firm. We borrowed the metaphor from Dave Free and the pgLang team.
“A specialist is a contractor, excellent at building one kind of wall. A generalist is a contractor with many tools but no shape to the work. An architect is the person whose output is the building. Architects know how rooms relate, how materials meet, what the thing is for. They direct contractors. They do not compete with them. And crucially, they become more valuable as contractors get cheaper, because somebody still has to decide what is being built and whether it is any good.”
The specialist-versus-generalist debate was always a debate between contractors. What changed in the last two years is that contracting, in the AI sense, is now something anyone can do. Which means the architect layer, the only layer that was ever about judgment rather than labor, is the layer that still commands economic value. That layer runs on a skill the operator class has been undertrained in for decades. Taste.
Taste has been treated as decoration, a personal aesthetic choice, a thing you either have or do not. That framing is lazy and it is also expensive, because it hides the fact that taste is a working skill, trainable, accumulative, and in 2026 the rarest one in any room. Taste is what tells an operator which AI-generated draft is good enough and which is slightly wrong in a way that will cost the pitch. It decides which of the ten thousand things you could now make is actually worth making. And it is the compressed judgment of a thousand references you have internalized well enough to act on without needing to explain.
Taste is also the hardest thing to outsource to a tool, because tools do not have an inside. Taste comes from having read the book, watched the film, sat with the painting, worn the garment, argued with the thing long enough that you know why it works. It is the output of a life of absorbed attention. No model currently in existence absorbs attention the way a person does, and no near-term model will, because the thing that trains taste is not data. It is care. Care about one thing enough that the care trains discrimination. Care about thirty things that way and you have a worldview.
This is the part nobody wants to say in a career conversation because it sounds insufficient. "Develop your taste" is not a plan. It is not a degree program. It cannot be priced. It will not comfort anyone who wanted a one-sentence answer to the question of whether to go deep or wide. It is, however, the honest answer, and the operators who have understood it are already moving.
What we are watching in 2026 is the end of work as skill and the beginning of work as judgment. The distinction matters. Skill can be learned, automated, and eventually commoditized. Judgment is personal, accumulative, and cannot be bought. A career oriented around skill has a shelf life now. A career oriented around judgment can survive any tool.
The operators we pay attention to are all making the same move. They are deepening their point of view instead of adding more skills. Reading slowly instead of scanning. Revisiting their old work. Having specific conversations with specific people while building an audience. They treat taste as the job rather than the garnish. They act like architects and they let the contractors, human and synthetic, do the execution.
This kind of work does not show up as productivity in a spreadsheet. It looks like staring at a book for an hour, rewatching a film for the fifth time, sitting with a brand mark until the reason it works becomes articulate. The market is going to reward this kind of work more in the next five years than it has in the last fifty, because it is the only work left that is genuinely scarce.
There is a version of this argument that feels scolding. We do not mean it that way. The shift is hard. Many people spent careers becoming excellent at skills that are now being automated, and the honest response to that is grief. Grief is not a plan, though, and the plan that works is this.
Stop picking a lane. Build a point of view strong enough that the lane question stops making sense. Treat the taste you have spent your life quietly sharpening as the actual asset. Let the AI handle the walls.
You are here to draw the building.
If this made you think of something you're building, reply and tell us. We read everything.
~ native unit
ON OUR RADAR
Adobe launched Firefly
On April 15, Adobe launched Firefly AI Assistant, a conversational agent that orchestrates multi-step work across Photoshop, Premiere, Illustrator, Express, and the rest of the Creative Cloud stack. The language Adobe used in the announcement is the part worth sitting with: the creator provides the vision, judgment, and creative direction, and the assistant handles the orchestration and execution. The largest creative software company in the world just published, in plain English, the thesis of this week's article. Execution is no longer the paid layer. Judgment is.
Read the launch note: news.adobe.com/news/2026/04/adobe-new-creative-agent
A beer brand made "good taste" its entire positioning
On April 1, 1664 launched its "Unquestionably Good Taste" campaign globally, starring Robert Pattinson and directed by Brady Corbet. The film is playful and the campaign is a premium beer ad. The fact worth noticing is that a mass-market brand is now building its long-term positioning around the word taste. When the ad industry starts selling a concept back to operators, it usually means the concept has already moved in the culture. Taste has officially crossed the threshold from creative-class vocabulary into mainstream brand currency. Which is a tell.
THE REFERENCE
Taste for Makers (2002) by Paul Graham
Twenty-four years ago, Paul Graham wrote an essay arguing that taste is not a matter of personal preference but a set of trainable principles, and that makers who dismiss it as decoration lose the ability to produce work that lasts. He listed the principles: simplicity, timelessness, solving the right problem, suggestive power, a certain playful daring. In February of this year, Graham reposted the essay on X alongside a new prediction: "In the AI age, taste will become even more important. When anyone can make anything, the big differentiator is what you choose to make." The essay is more useful in 2026 than it was in 2002. Read it once a year.
Forward this to one person building something who should be reading it.
The newsletter is free. The conversation is on Instagram: @native.unit