AI cheapened one corner of the triangle — which is why it stopped being the advantage
Building a company with AI: skill × market × leverage, validating by hand, and the moats that survive everyone having the same tools
The question: If the organisation is you, what should you build — and which advantages survive a technology that makes building cheap for everybody?
Everything in this module assumed an organisation with a board, a budget and a portfolio. This lesson runs the same three disciplines — strategy, unit economics, a ninety-day decision — for a company of one. It is also the only lesson in the course where most of the available evidence is marketing, so it doubles as an exercise in reading a genre honestly.
What the lesson covers
The organising idea is the **founder's triangle**: skill × market × leverage. Something you are unusually good at, people with a problem and a budget, and a way to produce output without trading hours for it. The three are multiplied rather than added, which is the whole point — any one of them at zero and the result is zero however strong the other two. Skill with no market is a hobby. Market with no skill is a wish. Both with no leverage is a job, and there is nothing wrong with a job.
What AI changed is precisely one corner. **Leverage** used to require hiring, funding or an audience; it is now available to one person with a laptop and a subscription, which is why the current wave of one-person businesses is real rather than rhetorical. But notice what follows: a corner that becomes available to everybody stops being a differentiator. The bottleneck moved to the two corners that did not get cheaper, and neither skill nor a market can be downloaded. Most people, offered this diagnosis, keep working on the corner they are already strongest in.
Before building anything, **deliver the result by hand**. Not a survey, not a landing page measuring interest — produce the actual outcome, manually and expensively, for people who pay. Ten of those is the cheapest evidence in existence and almost nobody buys it, because building feels like progress and selling does not. Two things reliably happen: some of the ten do not renew, which is the finding; and the product you then build is not the product you would have built in month one, because doing the work by hand is the only way to learn which part customers actually value. Measure **retention, not applause** — signups measure your launch, and "that's a great idea" costs nothing to say.
The shape of a one-person product follows from one constraint: every option you add is a support ticket you will answer yourself. So: one job done completely rather than a platform; no roles, permissions or admin panel, since those serve organisations buying for teams; the simplest version whenever you are uncertain, because you can add from evidence but cannot remove without apologising. Price the outcome rather than the seat — ten customers at €2,000 and four hundred at €50 are the same revenue and roughly a fortieth of the support. And know your **cost per completed task** with your own minutes in it: a product that saves a customer an hour and costs you ninety minutes to run is a hobby with a payment link.
Finally, the honest part. Five advantages survive everybody having the same tools: **distribution** (an audience that already listens), **proprietary data** (outcomes nobody else recorded — which appreciates, per the model-collapse lesson), **trust and permission** (regulated access, a name somebody will put in front of their board), **embedded workflow** (you are in the process, not beside it), and **taste** (knowing which of forty generated options is right). Being early to a tool is not on that list; it expires the week the tutorial appears. And every account of this subject you will read, including the sources behind this lesson, is told by somebody it worked for — so take the mechanisms, which are cheap to test, and discount the outcomes, which are not evidence about you.
Key points
- Skill × market × leverage, **multiplied**. AI cheapened leverage, which is exactly why leverage stopped being the advantage — the bottleneck moved to the two corners that cannot be downloaded.
- **Deliver the result by hand** for ten paying customers before building. Measure retention, not signups; encouragement is not evidence and a card number is.
- One job, no settings, simplest version. Every option is a support ticket you will answer yourself, and price the outcome rather than the seat.
- Know your **cost per completed task with your own minutes in it** — the largest line in a business of one is never on an invoice.
- Five moats survive: distribution, proprietary data, trust, embedded workflow, taste. Being early to a tool is not one, and it expires in weeks.
Framework — Skill × Market × Leverage
Score each corner out of five, honestly, before building anything. Skill: what are you unusually good at, that has not just become free? Market: who has this problem, and who holds the budget? Leverage: how does output happen without your hours? Multiply, do not add — a 5 × 5 × 0 is zero. Then work on the lowest number, which is almost never the one you have been working on.
The lab
Score your own triangle, find the one result somebody would pay you to produce by hand, and name a moat you could start this quarter.
Open this lesson, its lab and its quiz
Sources and further reading
- The founder's triangle, the DREAM framework, and moats AI cannot erase — Sandeep Swadia
- Build a real business, not a cool demo — validating before building — Young Zhao (OpusClip)
- Six founders on what they would build today — Replit · ElevenLabs · Benchmark and others
- The rise of anti-startups — small and profitable as a destination — Enrico Tartarotti
- Survivorship bias — the missing base rate in every founder story — Wikipedia
- Full AI prompting course — the leverage corner, taught properly — Andrew Ng · DeepLearning.AI