There is a story you have heard. It goes: pick a lane. Become the best in that lane. Compound for ten years. The wealth, the reputation, the audience, all of it follows from depth.
The story is correct, with a single asterisk that nobody quite wants to read out loud.
The asterisk is this: depth is most valuable when nobody else is willing to do the depth. When everybody has access to a depth-on-demand engine, which is roughly where we are now, depth becomes more like running water in a flat; assumed, undifferentiated, and only noticeable when it is missing.
Naval is right. He is also right about a different thing than the one he is taken to be.
Naval Ravikant's specific-knowledge thesis is the standard articulation: 'specific knowledge is found by pursuing your genuine curiosity'. It is correct. The curiosity has to be real. The depth has to be earned. You cannot fake your way into specific knowledge by reading a single book.
The bit that gets quietly dropped from the popular retelling is that Naval explicitly describes specific knowledge as something that cannot be taught. By implication, it cannot be replicated by a model that has read every book. By further implication, the species of specific knowledge most valuable in 2026 is the kind that lives at the intersection of two or three fields that no single training set has integrated.
Cross-domain combinatorics, not vertical depth, is the moat.
The four-practices stack
I am an advocate and solicitor of the High Court of Malaya. I run a law firm. I also run a software company that ships AI products to small businesses. I produce dance music for an underground label. I write books that have no audience yet.
By the orthodoxy this is professional malpractice. I should pick one. I should compound. The intersections are not legible to my LinkedIn audience and not legible to my booking page and not legible to my mother.
The intersections are exactly the thing that makes the work possible.
Court documents are a software product because I am a lawyer and a software engineer at once. AI in WhatsApp is a viable channel-partner business because I am both an operator who runs the bots and a lawyer who can write the indemnity contract. Personality assessment becomes a paid product instead of an academic curiosity because I can wire the Hogan-grade science to a Stripe checkout the same afternoon. The label survives because I treat the contracts and the masters as the same artefact instead of two separate professions trading invoices.
What the literature actually says about deep work
Cal Newport's Deep Work is the other side of the specialisation orthodoxy, and it is correctly read as a defence of long, uninterrupted blocks on a single problem. It is sometimes incorrectly read as a defence of long, uninterrupted blocks on a single profession. Newport's actual claim is about the structure of an hour, not the structure of a career.
You can practice deep work across four practices. Each session is single-focus. The switching cost between sessions is real and worth measuring. The combinatoric advantage of having four problem spaces feeding each other is also real, and is generally larger than the switching cost; unless you are doing the switching badly, which is a separate problem with a separate fix.
What the labour economics says, post-AI
Daron Acemoglu's 2024 NBER paper on the macroeconomics of AI makes a careful case that AI compresses the wage premium of mid-skilled cognitive work most aggressively, and that the workers least exposed are those whose value sits in the combination of multiple skills rather than the depth of any one.
Translation: the lawyer who only practices law gets paid less, slower, every year. The lawyer who can also ship the product the law firm needs gets paid more, faster, every year. The producer who can only produce gets compressed. The producer who understands the publishing contract under the master license gets a bigger cut and the legal letter is not a separate procurement.
Three objections, each correct, none fatal
The first objection is that this requires being above-average at all four practices, and being above-average at any one is already hard. True. The answer is to be properly good at the intersections, where the bar is lower because the field is newer, instead of trying to be the best in the field for each one.
The second objection is that this is a story told by people who happen to have made it work, and the field is littered with people who tried it and ended up as four mediocre practitioners. Also true. The honest version of the four-practices thesis is that the practices have to feed each other; if they don't, you are not running four practices, you are running four hobbies.
The third objection is that 'refuse to specialise' is a slogan rather than a strategy. Correct. The strategy is: pick two or three fields that genuinely talk to each other, become operationally competent in each, and let the intersection be your specific knowledge. The slogan is a side-effect of the strategy working.
What this looks like in practice
I do not recommend running four practices. I recommend running two, or three at the outside, that genuinely talk to each other. I would not recommend, for example, becoming a tax lawyer and also a competitive eater; the intersection is not a real business.
I would recommend becoming a lawyer and a builder. Or a doctor and a writer. Or a producer and a copyright specialist. The pattern is that two of the four practices should generate the artefacts that the other two get paid to handle. The legal practice writes the contracts that the software practice deploys. The music practice writes the songs that the label practice releases. The writing practice writes the books that the publishing practice publishes. Everything pays back to the next thing.
The orthodoxy will keep saying pick a lane. The orthodoxy is correct, with that one asterisk. Pick the lane that is wide enough to hold two professions.