Tiny teams, expensive distribution
AI lets a three-person team build like twelve, but being found, believed, and bought costs what it always did, and it eats the freed margin.
Bob Bass
Something real happened to the cost of building software. A three-person team with modern tooling and a capable assistant can now ship what took a dozen people and a couple of years not long ago: account flows, billing integration, an admin panel, dashboards, transactional email, documentation, and support automation that each used to need a specialist for a quarter. Teams of two and three are posting real revenue doing it, and the pattern is common enough that it is no longer a story about exceptions. The part that never got its discount is everything that happens after shipping: being found, being believed, and being bought. Distribution is still expensive, and it eats exactly the margin the engineering savings freed.
What actually got cheap
Run an inventory of the boring layer that used to sit between an idea and a sellable product. Account flows with verification and recovery. A payment integration. An internal admin tool that only employees see. A dashboard worth showing a customer. Transactional email, docs, a changelog, an onboarding sequence, and the long tail of things that make a product look like a company. A decade ago that list was months of work spread across three or four hires. Today much of it is weeks, and pieces of it (the first draft of the docs, the glue between systems, the v1 of the admin panel) genuinely are an afternoon with good tools.
The deeper change is that the boring infrastructure is no longer the reason small teams needed funding. What held tiny companies back was rarely the idea and almost always the quarter of engineering per unglamorous subsystem. When every subsystem got a compression step, the minimum company that could carry real customers got smaller, and the contract between headcount and ambition loosened. That is the part worth admiring: not smarter products, but a smaller viable company, arriving at the point where the engineering constraint that dominated software for two decades stopped being the binding one.
Distribution never got the same discount
Distribution did not get cheaper for a structural reason: it is the market side, and the number of attentive buyers did not multiply along with the tools that create supply. Cheap building multiplied products, launches, and competent content, all competing for the same fixed attention. Every time the cost of producing an essay or a demo falls, the bar for being noticed rises, and the channels reprice to match.
Paid reach repriced first. Auction platforms price a click up to what the highest bidder can rationally pay, and an incumbent with a mature funnel can rationally pay far more than a three-person company can. A small team outbidding that price is spending its freed margin on the most expensive free thing available. Content channels followed the same path: search results filled with machine-assisted pages of respectable quality, which quietly lowered the value of being merely good, and feeds reward posting volume in a way that a two-person team cannot sustain while also running the company.
Trust repriced too, and this is the part no build tool touches. Polished no longer proves an organization exists, because everyone can now ship a beautiful website in a weekend, so buyers learned to discount polish and check whether the substance behind it holds. Proof now has to come from things that do not generate on demand: a changelog that shows judgment over months, an answer in a community that was genuinely useful, a demo that survives someone pushing on its edges, a customer willing to take a reference call. That evidence accrues slowly, and it cannot be parallelized the way the build can.
Where the freed margin goes
The engineering windfall is real, which makes the tempting failure mode worth naming: the velocity trap. Shipping is cheap, so shipping more feels like the rational response to a good week, and the roadmap fills with improvements to a product nobody has heard of. A tiny team can now out-build its own distribution, which is a genuinely new condition for software.
The teams converting the savings into revenue spend the other half deliberately. They pick one channel and staff it with a person who owns it for a year, because the founder writing or presenting is the differentiator the channels cannot cheaply copy. They answer the support inbox personally for as long as possible, since the founders answering support is marketing, research, and trust built in the same hour. They hand-polish the demo instead of the onboarding doc, chase case studies even when it feels awkward to ask, and keep a boring weekly hour of being useful in a community because it compounds precisely because it is not automatable. For a growing number of tiny teams, the first non-engineering hire is a distribution hire.
None of that is glamorous, and almost all of it is slower than shipping. But the bottleneck moved rather than disappeared: the build got cheap enough that reach, reputation, and pipeline are now the expensive half of the business. The teams that treat that as a staffing and calendar problem, instead of hoping a good product sells itself, are the ones that get to stay tiny and keep the margin.
From the desk that wrote this
This blog is run by the team behind Layerbase, so every post comes with a conflict of interest attached rather than concealed behind it.