Startups - cheap founding, expensive moats#
Contents
The startup system is the economy's formation layer: the set of institutions, capital pools, and cultural scripts that convert new ideas into new firms. Under cheap cognition it does not merely get faster - it changes shape. Founding a software company becomes something a small team (or a person) can do in weeks. Defending one becomes something almost no pure-software company can do at all.
That is the whole argument. Everything else is mechanism, category map, and falsifier.
The argument in four claims#
1. Formation cost collapses; selection pressure rises. Idea-to-ship falls by an order of magnitude for anything whose product is symbols. The result is not more durable companies but more attempts, most of which fail faster and cheaper. → Formation
2. Value accrues only where Game 3's complements still bind. Pure software without data, distribution, license, liability capacity, or physical lock-in is Red Queen pure - adopt-or-die for customers, zero rent for the vendor. The startup categories that work are the ones that own an inelastic complement. → Selection
3. Venture capital is a market for power-law outcomes; cheap founding floods the left tail. When thousands of teams can ship an MVP for the cost of a seed check, deal flow explodes and signal quality collapses. The fund that survives is the one that underwrites moat quality, not product demos. → Venture
4. Exit paths concentrate. Acquirers buy distribution, data, and people more often than code; IPOs require durable margin, which pure SaaS increasingly cannot show. Death becomes the modal outcome faster; acqui-hires and quiet shutdowns replace the mid-tier "successful enough" exit. → Exits
5. The constructive edge is a request list. The same selection filter implies a short list of company shapes worth founding - verification, atoms, and institutions - not another generation wrapper. → Ideas
Why this sits under cognitive domains#
Most of the formation cost that is collapsing is cognitive: writing code, drafting legal scaffolding, generating marketing, running diligence, building the first sales deck. The physical and contested worlds still require capital, permits, wet labs, and supply chains - those startups do not get free founding, and that difference is itself a selection force. Filing startups under cognitive is a claim about where the shock hits first, not a claim that every new firm is a SaaS company. The selection page is where the physical and licensed categories re-enter as the places that still work.
This is also the page that answers a question the rest of the corpus mostly leaves open: if incumbents must adopt and mostly do not profit (Game 3), who does form the AI-native firm? Entrants, not transformers - but only entrants that hold something the model layer cannot manufacture. The failure mode for the whole Part III framing, stated in the domains hub, is exactly an entrant that rebuilds the workflow rather than selling tools into it. This section is that failure mode treated as the object of study.
The single sentence#
Cheap intelligence makes starting a company easy and makes being a company hard, unless the company owns something intelligence cannot make.
That sentence is Game 3 applied to firm birth rather than firm survival. The rest of the section is the applied map.
Where the four claims sit on the clock#
| Layer | Moves when | Observable |
|---|---|---|
| Formation | Already; accelerating through 2028 | Median team size at seed; months to first revenue |
| Selection | Continuous; visible in cohort outcomes by ~2029 | Category mix of funded rounds vs durable ARR |
| Venture | Fund vintage 2024–28 marks by 2030–32 | Seed multiples; power-law concentration of returns |
| Exits | Lagged 5–10 years after founding | Acquisition vs IPO mix; median exit multiple |
| Ideas | Continuous (founder action now) | Whether capital and talent actually flow to complements |
Failure mode for the hub argument#
All four claims assume that customers can and will switch when a cheaper or better-looking alternative appears, so pure software cannot hold price. The reading that breaks them is permanent distribution lock-in at a few platforms: if OS defaults, app stores, enterprise procurement, or model-provider marketplaces intermediate almost all demand, then the rent sits at the platform and the "startup" becomes a feature farm feeding whoever holds the default - still Red Queen for the farm, but a different beneficiary than consumer surplus. Watch platform take rates and default placement more carefully than demo quality. A second break: if learned verification retires the master asymmetry in categories currently "hard," the selection map reshuffles without killing the formation claim.
What to watch#
| Signal | Reading |
|---|---|
| Solo / two-person seed share among software startups | Formation cost collapse is real |
| Vertical SaaS churn and internal-tool build rates | Long-tail pure software dying as predicted → software |
| Seed-to-Series-A conversion and time-to-kill | Selection pressure up, or capital still over-funding the left tail |
| Share of venture dollars into licensed / physical / data-native categories | Selection map working |
| Acquisition target composition (talent vs product vs data) | Exit mechanism resolving |
| Outcome-priced revenue at AI-native vendors | Reliability belief; B5 |
Sections#
| Formation | Team size, capital intensity, idea-to-ship, and the apprenticeship of founders |
| Selection | Category map: what survives Game 3 and what is demo economics |
| Venture | Fund structure, signal collapse, and who underwrites moats |
| Exits | Acquisition, IPO, acqui-hire, and the death rate |
| Ideas | Requests for startups - what to build, filtered by this forecast |
Related: Game 3 - Firms · Software · Inference economics · Capital · Assets · Insurance · Compressed