You can build an app in two hours; a decent product takes two months. AI has crushed the price of "writing code" down to nothing, but the two things that actually decide life and death — whether anyone wants it (demand) and whether you can reach them (distribution) — haven't gotten one cent cheaper. This piece takes the classic theories along the "distribution vs. demand" and "incumbent vs. startup" axes and adversarially verifies them one by one: which hold up, which are just widely-repeated bumper stickers, and which remain unsettled in the AI era.
"Build it and they will come" is Silicon Valley's most expensive comfort. Its error isn't that "the product doesn't matter" — it's treating distribution as downstream busywork you do after the product is built. But distribution is itself an independent, power-law-governed life-or-death variable.
In chapter 11 of Zero to One, Peter Thiel offers the observation people so often overlook: distribution follows its own power law. In his words, "a single channel that works is enough to build a great business" and "one channel usually dominates all the others." This yields a counterintuitive tactical conclusion:
Most companies can't get even one channel to work — poor distribution, not a poor product, is why they die. What's truly scarce isn't "can you build it," it's "can you reach them." Zero to One · Ch.11 (independently corroborated by the Bullseye framework in Weinberg & Mares, Traction)
The Bullseye framework in Traction arrives at the same answer from the other end: list all 19 categories of distribution channel, test them lightly in parallel, then pour all your resources into the single channel that performs best. Two sources converge — the right posture for distribution is focus, not breadth.
The widely circulated line "poor sales, not a poor product, is the number-one cause of startup death" — that one did not pass in this round of verification (see §09). It's Thiel's rhetoric, not a reliable statistical fact. The core claim ("concentrate and nail a single channel") is well-supported, but don't cite that cause-of-death assertion as data.
Many founders think that once they've reached product-market fit they're safe ashore. Brian Balfour (former VP of Growth at HubSpot, founder of Reforge) offers the counterexample: companies with every PMF signal that still can't grow are everywhere.
The reason is that growth requires four meshing gears to be aligned at once, none of them optional:
| Fit | Meaning | What missing it looks like |
|---|---|---|
| Market ↔ Product | The product satisfies a market that really exists | Classic PMF; having it still isn't enough |
| Product ↔ Channel | The product's shape fits the channel you want to use | You built a product that needs high-touch sales, yet you're counting on organic growth |
| Channel ↔ Business model | The channel's economics can support your pricing | A $10/month subscription can't fund an enterprise direct-sales team |
| Business model ↔ Market | Your pricing matches the market's ability to pay / its size | Forcing a high-volume low-price model onto a niche market |
The one that stings indie developers most is the assertion about ordering:
The product is built for the channel; the channel won't bend to the product. Brian Balfour · Product Channel Fit (TikTok and Pinterest both built their products around the channels they relied on to take off)
corollary Don't finish building and then wonder "how do I get this out there." Do it the other way — figure out first which channel you'll live on, then let the product grow into the shape that channel likes. Distribution isn't downstream of the product; it's an upstream constraint on product design.
The word "demand" gets used loosely; pulled apart, it's two entirely different ledgers.
| Demand capture | Demand creation | |
|---|---|---|
| Who you target | People searching right now, ready to buy | People who don't yet realize they want it |
| Typical channels | Search / ASO / price comparison / direct-intent traffic | Content / brand / social / word-of-mouth building |
| Economics | Cheap, high-converting, but capped at existing search volume | Expensive and slow, but can open up a new pie |
This maps to marketing's "95:5 rule" (Ehrenberg-Bass): at any given moment, only about 5% of people in a category are "in the market" and ready to buy, while the other 95% aren't buying for now. medium confidence
That "5%" comes from long B2B purchase cycles; it varies by category and is not a constant. For a high-frequency, low-decision-cost consumer app, the share of people "in the market right now" will be significantly higher. Don't copy 5% over as an iron law. On top of that, the claim "you must create demand before you can capture it" did not pass verification (§09) — the two aren't sequential; they're two ledgers you can run in parallel.
corollary If what you're making is something people are already actively searching for (a tool, a vertical service, a category with established awareness), your first battlefield is almost certainly demand capture — reaching them at the exact moment they've already asked, rather than burning money to educate people who don't want to be educated. Existing high-intent demand is the most underrated lever an indie developer has.
Even if you've nailed a channel, it won't work forever. This is an iron law that Andrew Chen (a16z GP, growth advisor to Dropbox / Uber) and Balfour hammer on repeatedly.
The Law of Shitty Clickthroughs high confidence: the first banner ad (HotWired, 1994) had a clickthrough rate above 70%; two decades later the banner average is around 0.05%. The mechanism is crowding from competitors copying you, plus user habituation. Balfour backs it with another set of numbers: email lightbox popup conversion fell from over 10% all the way to 5–6%; "the Facebook ad that worked 90 days ago doesn't work today." The conclusion is plain and brutal — there is no single permanent channel; growth is an adapt-or-die environment.
So what actually holds up? The answer is to build distribution into the product itself. Andrew Chen's read on Dropbox is representative:
Dropbox's real engine was "inherent virality" — you share a folder and pull a new user in along the way. That famous referral program merely captured word-of-mouth that was already happening, rather than manufacturing it. Compared with building a great viral feature, a referral program will always ride in the back seat. Andrew Chen (via the Intercom podcast)
Andreessen elevates this into a company's evolutionary path: successful tech companies shift from "product-centric" to "distribution-centric" — treating the channel itself as an asset and pushing one new product after another through it. high confidence
Ask yourself one question: is there an action in my product where "the user, purely to get more out of it for themselves, pulls someone else in along the way"? Collaboration, sharing, comparison, invitation — any feature that naturally requires a second person present is a candidate for built-in virality. It's worth more than any "invite for a reward" bolted on after the fact. If your product naturally has such a surface, don't treat it as a tab; design it as your growth engine.
On the same distribution-demand axis, incumbents and startups stand in completely asymmetric positions. Understanding that asymmetry is the prerequisite for a startup to find a way out.
The mechanism as Clayton Christensen defines it in The Innovator's Dilemma: high confidence disruptive innovation lets a new entrant come in at the low end of the market (or open an entirely new market), and eventually turn on the established leader. And the reason incumbents don't chase isn't stupidity — it's rationality:
So it actively hands the opening to startups. This is the structural crack through which a startup fights a giant.
The mechanistic definition of the innovator's dilemma is reliable and extremely widely cited, but its empirical predictive power has been questioned by scholars (Jill Lepore 2014, King & Baumann 2015, and others) — it's good at explaining after the fact, poor at predicting before it. Use it as an analytical tool, don't quote it as a guarantee.
The Holloway venture guide breaks startup success into three insights: market insight, product insight, distribution insight. medium confidence Of these, "distribution insight" is defined most precisely — what do you understand that others don't about how to reach this particular group of users? (The guide's example: if your team spent ten years in search, do you hold search-acquisition knowledge that others don't?)
corollary A startup lacks distribution and still has to prove demand; the only thing that can turn the game around is finding a distribution channel the giant "can't or won't use." "Can't" might be compliance / licensing / supply barriers; "won't" might be that the market is too small, the margin too thin, or the brand doesn't want to touch it. A counterintuitive result: the dirty, slow paths you've walked — the ones the big players look down on and won't touch (compliance in certain heavily-regulated arenas, the manual community-building of a cold start, some vertical scenario a giant can't be bothered with) — can themselves be a moat no one can copy. The wheel can be copied; your exclusive understanding of "how to reach this crowd" cannot.
Every discussion of distribution has one precondition: you first need a product a market wants. Get the order backwards, and distribution only speeds up how fast you burn cash.
Two laws proposed by Andy Rachleff (Benchmark partner) and popularized by Andreessen's 2007 The Only Thing That Matters: high confidence
Good team meets bad market, market wins; bad team meets good market, market wins. In a good market, the market pulls the product right out of the startup's hands… and a market that doesn't exist doesn't care how smart you are. Marc Andreessen · The Only Thing That Matters (paraphrasing Rachleff's law)
The direct corollary: talking about scaling before PMF = premature, and premature scaling is the number-one startup killer (in Startup Genome's empirical data, roughly seven in ten failed startups stumble at this step). What distribution scales must be a loop that has already been validated and whose unit economics work; scaling something not yet validated just makes the losses bigger.
The question this piece most wants to answer, yet most honestly can't give you a definite answer to: when AI drives the cost of building toward zero, does distribution become the only moat?
In this round of adversarial verification, not a single primary claim on this question survived. The existing frameworks — power law / four fits / channel decay — were all born before AI. The items below come from primary sources dated 2025–2026, but they're only search snapshots, not fully verified, and can be read as prevailing winds only, not conclusions:
Assemble the verified parts and you get a judgment that isn't radical but is solid: when building trends toward free, the value of "being able to build it" trends toward zero, and the center of gravity of competition necessarily gets squeezed toward the two ends of distribution and demand. But "distribution is the only moat" is too strong a claim — distribution solves "being known," retention / momentum solve "staying alive," and neither can be skipped. The moat hasn't moved to the "only" distribution; it has redistributed, from the now-vanished dimension of "can you build it," into the harder, less-copyable dimensions of distribution, demand, and retention.
This section is the piece's integrity collateral. The following claims circulate widely online and sound smooth, but in multiple rounds of independent adversarial verification they were rejected (a 2/3 dissent vote required). Don't write them into your pitch.
| Popular claim | Why it was rejected |
|---|---|
| "Poor sales, not a poor product, is the number-one cause of startup death" | Thiel's rhetoric, not a verifiable statistic; as an empirical assertion of a "cause-of-death ranking," the evidence is insufficient. The core claim (concentrate on a single channel) holds, but this attribution does not. |
| "Distribution must be built into the product itself" attributed to Thiel's Zero to One | "Built-in virality is strongest" holds (it comes from Andrew Chen), but treating it as Thiel's argument is misattribution; that specific attribution was rejected. |
| "Demand creation and capture are sequential and complementary: you must create before you can capture" | The relationship between the two is unsettled; they're more like two parallel ledgers than a mandatory sequence. |
| "The only lasting defense against channel decay is finding the next new channel before rivals saturate it" | Finding a new channel is one solution, not the "only" one; a built-in loop and deeper retention are equally defensive. |
| "Growth loops compound multiplicatively while funnels only add linearly, opening a 7.7× gap after 24 months" | 0/3 unanimous rejection That specific numerical model doesn't hold up under scrutiny. The qualitative distinction (loop vs. funnel) is valuable, but don't cite its specific multiple. |
The diagnostic itself (worth keeping): turn off paid acquisition — if growth stops completely, what you have is a funnel; if growth continues, what you have is a loop. An indie developer should prioritize building a loop that self-sustains even with ads turned off.
Compress the verified parts above into a few actionable judgments.
AI didn't make building products easy; it just made the least important part (writing code) free. The energy you save should go entirely into the two things it never solved for you: whether anyone wants it, and whether you can reach them.
Method: 6 search angles in parallel → fetch primary sources → extract falsifiable claims → independent adversarial verification of each with 3 votes (a 2/3 dissent required to reject) → merge, deduplicate, and synthesize. The confidence labels in §01 come from this process; rejected claims are in §09 and don't enter the body.