The generalist product builder
The advantage is not breadth. It is owning the seams.
Thesis
Picture the person reading your résumé. It says product management, design and growth. They have thirty seconds and a pile of applications, and they think: no depth anywhere.
That reaction is not unreasonable, and it is also the wrong test, because it measures the candidate against a set of components rather than against the thing that actually fails in small companies.
Specialists own components. Almost nothing fails inside a component. Products fail at the joins: between what the product does and how it is described, between the onboarding flow and the pricing page, between the technical constraint and the design that ignored it, between the feature that shipped and the message that promised something adjacent.
Nobody is accountable for a seam. Everybody is accountable for a component. The generalist's actual product is coherence across the joins, and coherence does not appear on any skills list, which is why the value is systematically undersold by the people who have it and systematically misjudged by the people evaluating it.
Context
There is a real question underneath the posturing on both sides, and it is not about whether breadth or depth is better in general. It is about scale.
At a large company the seams are handled by process, and process works. There are handoff rituals, shared documents, design systems, review gates. Specialisation is correct there, because there is enough volume of each kind of work to keep a specialist fully occupied and improving, and coordination can be bought.
At small scale neither condition holds. There is not enough design work for a full-time designer in the first year, and coordination cannot be bought because there is nobody to coordinate. The question is not which is better. It is where the crossover point sits, and the answer is a function of volume rather than of taste.
Research
Conway, 1968. Organisations design systems that mirror their own communication structure. This is the single most useful thing to know about why products come out incoherent.
Split product, design, engineering and marketing into four groups with four managers, and you get a product with four visible personalities. The onboarding flow speaks in one voice, the settings pages in another, the pricing page in a third. Nobody chose that. The structure chose it. And overcoming it requires deliberate, continuous effort, which is what a design system and a content style guide actually are: mechanisms for fighting Conway's law.
One person building the whole thing gets coherence for free. That is not skill, it is structure, and it is the strongest argument in this article.
Brooks, 1975, on communication overhead. Communication paths in a team of n people grow as n(n-1)/2. Four people have six paths, eight have twenty-eight. Brooks used this to explain why adding people to a late project makes it later.
A single builder has zero communication paths, and this is worth more than it sounds. The cost of a coordination path is not just the meeting. It is the specification that has to be written for a decision that would otherwise have been made in three seconds, and the fidelity that is lost between the intent and the document.
Smith, 1776, on the limits of the division of labour. The chapter people skip: the division of labour is limited by the extent of the market. Specialisation requires enough volume of a task to occupy a specialist, and below that volume the gains are unavailable.
This is the economic core of the argument and it is not ideological. A product with four hundred users does not generate enough of any single kind of work to justify a specialist in it. The specialisation gain is real and it is out of reach until the volume exists.
Coase, 1937, on why firms exist. Activity gets internalised when the transaction cost of coordinating it externally exceeds the cost of doing it in-house. Applied to a small team, this predicts generalism directly: the cost of finding, briefing, reviewing and integrating a specialist's work on a two-week project frequently exceeds the quality difference. Not always. But the calculation is a calculation, not a philosophy, and it changes as the work gets larger.
Tetlock, 2005, Expert Political Judgment. Twenty years of forecasting tournaments, and the headline finding is that experts were poor forecasters. The more useful finding is the split within them. Tetlock's "foxes", who held several frameworks loosely and drew on many small pieces of knowledge, consistently outperformed "hedgehogs", who applied one deep framework confidently.
This is the best empirical support available for generalism as a judgement advantage, and the mechanism matters: it is not that foxes knew more, it is that having several frameworks made them less confident and better calibrated.
Hogarth, Lejarraga and Soyer, 2015, on kind and wicked learning environments. A distinction that resolves most of the deliberate-practice argument. In kind environments, feedback is immediate, accurate and unambiguous, and deliberate practice produces expertise reliably: chess, golf, sight-reading. In wicked environments, feedback is delayed, noisy, incomplete or actively misleading, and experience can produce confidence without accuracy.
Product decisions are wicked. The feedback arrives months later, confounded by five other changes, on a sample too small to separate. That is the environment where accumulated depth in one framework is least protective and where breadth of frameworks helps most, which is exactly Tetlock's result appearing again from another direction.
Argument
The seams are where the value leaks, and the list is longer than people expect.
Between the technical constraint and the design: a designer without technical intuition specifies things that are expensive for no user benefit, and an engineer without design intuition implements the letter of a spec and loses its point.
Between the product and the message: features get built and then described in language that promises something adjacent, which produces users with the wrong expectations, which reads as a quality problem and is a translation problem.
Between acquisition and the product: the channel determines what the visitor expects, and if the landing page and the first screen were built by different people they will disagree, and the disagreement shows up as a conversion number nobody can explain.
Between design and analytics: an event taxonomy designed after the interface measures whatever was easy to instrument rather than what the product is for.
Between pricing and architecture: a usage-based price on a system that does not meter usage per customer is a pricing decision that cannot be implemented, and it gets discovered late.
Each of these is a place where a coherent whole is worth more than the sum of good components. And each one is nobody's job in a divided team.
The correct shape is not evenly broad. It is one deep thing plus enough range to own the joins.
Complete evenness is the failure mode the critics describe accurately. Someone equally shallow in five disciplines has no standards anywhere, and no standards means no ability to recognise good work, including their own.
The version that works: real depth in one area, enough to have taste and to know what excellent looks like, plus working competence across the adjacent ones. The depth is what gives you a floor. The breadth is what lets you carry a decision across a boundary without losing its intent.
Know your ceiling per discipline, explicitly, and buy above it.
This is the discipline that separates a functioning generalist from a liability, and it is the part nobody writes down.
There are things a generalist reliably gets wrong, and pretending otherwise is how the label earns its bad reputation. Security architecture beyond the basics. Accessibility beyond the checkable rules, where actual screen-reader behaviour and cognitive accessibility live. Data modelling that survives scale. Brand and typographic craft at a level anyone would notice. Statistical rigour in experiments, where the failure is a confident wrong conclusion rather than no conclusion.
Naming your own ceiling in each area is more valuable than raising it, because it tells you which decisions to buy in and which to make. The generalists who fail are not the ones with shallow skills. They are the ones who do not know where their shallowness begins.
Generalism is a function of scale, so it has an expiry date.
Below roughly a handful of people, coordination cost dominates and the generalist wins on the Coase calculation. Above that, volume of each work type becomes enough to occupy a specialist, and depth starts compounding in a way breadth cannot.
The failure is not being a generalist. It is being a generalist past the crossover, which produces a company where one person is the bottleneck on every decision and quality is capped at their ceiling in five different disciplines simultaneously. That is a worse outcome than any specialist arrangement.
Examples
A four-person startup building one product. Generalists win clearly. Coordination cost is the dominant term, Conway's law is working in your favour, and there is not enough volume of any single discipline to occupy a specialist. Hire people who can carry a decision from problem to shipped.
A fifty-person company with a design system. Specialists win clearly. There is enough work, process handles the seams, and depth compounds. A generalist at this scale is mostly useful in a role that is explicitly about joins: product leadership, or the person who owns a whole surface end to end.
A migration or a redesign. Generalists win regardless of company size, because the work is almost entirely seams. Understanding what the old system meant, what depends on it, what the new one should promise, and how to tell users. Nobody's component contains that.
A security review. Buy it. This is the clearest case of a ceiling worth respecting, because the failure mode is silent and the cost is unbounded, and a generalist's intuition here is the specific kind of confidence Hogarth's wicked environments produce.
Counterargument
"Generalist is what people call themselves when they have not got good at anything."
The strong version is not snobbery, it is a claim about how expertise forms. Deliberate practice requires sustained focused attention on a narrow skill with tight feedback. Attention spread across five disciplines never accumulates that anywhere, and the result is someone who can produce a passable version of everything and an excellent version of nothing. In a market where customers can tell the difference, passable loses.
There is a second objection that is more practical and, I think, more damaging: the labour market does not buy generalists. It buys legible specialists. Hiring processes screen on discipline keywords, roles are defined by function, and interview loops test depth. Whatever the operational merits, "generalist" is a positioning liability, and positioning is what determines whether anyone ever evaluates the operational merits.
Where this is right. The market objection is largely correct and I would not argue with it. Generalism is harder to sell than it is to practise, and someone choosing this path should understand they are taking on a communication problem for the rest of their career.
The deliberate practice objection is right about kind environments and I would concede it fully there. If the work has clear, fast feedback, depth wins and the generalist should not pretend otherwise. The argument for breadth is specifically an argument about wicked environments, and it does not generalise past them.
Where I think it is wrong. The objection measures against a component and ignores what is actually scarce. Someone who can take a problem from a customer conversation to a shipped, described, distributed thing is rare, and their rarity has nothing to do with their ranking in any single discipline. That is a different capability, not a weaker version of a specialist's.
And the positioning problem has a solution that does not require becoming a specialist: lead with outcomes rather than with breadth. "I take products from problem to shipped, and here is one" is a specific claim. "Product, design and growth" is a list, and a list invites exactly the thirty-second judgement it receives.
Practical implications
Position on outcomes, never on breadth. The list is what triggers the depth objection. The outcome is what answers it.
Have one thing you are genuinely deep in, and be able to demonstrate a standard in it. Without a floor somewhere you cannot recognise good work anywhere.
Write down your ceiling in each adjacent discipline, and the trigger for buying above it. Security, accessibility depth, statistical inference and brand craft are the usual four.
Take the seams deliberately. In any team, offer to own the joins nobody has claimed: the gap between the feature and its description, the gap between the channel and the first screen.
Watch for the crossover. If you are the bottleneck on decisions in three disciplines, generalism has stopped being an advantage and become a cap.
In wicked environments, hold several frameworks loosely. Tetlock's foxes were better calibrated, not better informed, and calibration is the transferable part.
Do not claim depth you do not have. The label survives on honesty about its limits, and one overreach discredits the whole position.
My perspective
Opinion, and it includes a live failure of mine.
My role line is "Product Manager · AI-Native Builder · Growth Strategist", and I use it as written or not at all. The reason I am strict about that is not brand discipline. It is that my own site currently positions me three different ways across three surfaces, and one of those surfaces is the file that AI search engines read, where I am described as a marketer rather than a builder. A visitor and a language model can currently form two incompatible impressions of what I do.
That is the real cost of generalism and it is not the one people warn you about. Breadth does not make you shallow. It makes you hard to describe, and if you do not solve the description problem yourself, every surface you own will solve it differently and none of them will agree. I have a verification item open on exactly this, which is an uncomfortable thing to publish and a more useful illustration than any argument I could make.
On the capability side, the honest accounting: what I can do is take something from a problem to a shipped, deployed artifact without handing off, and the record of that is public and dated. What I cannot currently demonstrate is measurement rigour. There is no instrumentation, no evaluation harness and no analytics export anywhere in my work, which means the discipline where my ceiling is lowest is precisely the one that would have made everything else provable. Breadth without measurement produces a portfolio of things that exist and cannot be assessed, which is roughly what I have.
If I were advising someone taking this path: get deep enough in measurement to be dangerous, before anything else. It is the discipline that makes every other claim checkable, and it is the one generalists skip because it produces nothing to look at.
Conclusion
The generalist advantage is not that you can do many jobs. It is that products fail at the seams between disciplines, that no specialist owns a seam, and that Conway's law hands coherence to whoever holds the whole thing.
It is also bounded. Specialisation requires volume, so generalism is correct below a scale and wrong above it. It helps in wicked decision environments and loses in kind ones. And it comes with a real, permanent cost that is about description rather than capability: you will be harder to explain than a specialist, and if you do not solve that yourself, the market will solve it against you.
Depth in one thing, range across the joins, and a written list of where your ceiling is. That is the whole position.
Sources
- M. E. Conway, "How Do Committees Invent?", Datamation 14(4), 1968, 28–31.
- F. P. Brooks Jr., The Mythical Man-Month, Addison-Wesley, 1975.
- A. Smith, An Inquiry into the Nature and Causes of the Wealth of Nations, 1776, Book I, Chapter III, "That the Division of Labour is Limited by the Extent of the Market".
- R. H. Coase, "The Nature of the Firm", Economica 4(16), 1937, 386–405.
- P. E. Tetlock, Expert Political Judgment: How Good Is It? How Can We Know?, Princeton University Press, 2005.
- R. M. Hogarth, T. Lejarraga and E. Soyer, "The Two Settings of Kind and Wicked Learning Environments", Current Directions in Psychological Science 24(5), 2015, 379–385.
- K. A. Ericsson, R. T. Krampe and C. Tesch-Römer, "The Role of Deliberate Practice in the Acquisition of Expert Performance", Psychological Review 100(3), 1993: cited for the counterargument.
- Positioning conflict across devchopra.life, buildwithdev.xyz and
/llms.txt. Recorded as C9 inSOURCE_INVENTORY.mdand as V10 inVERIFICATION_QUEUE.md.
Related reading in this archive
- Many small products, and whether that is a strategy
- What seventeen products in five months actually taught me
- What AI-native product management actually means
- The Build Archive: the public, dated record referred to above