I write the same things because they keep working. Treat building a startup like a disciplined probability game: reduce variance, expose problems early, and increase your odds. This is camel thinking: resilient, capital-efficient, customer-first, not unicorn theatre.
1. Profitability first. If you're not profitable, you don't control your destiny. Example: prove a €1k/mo customer segment before hiring sales.
2. Retention is the only proof of value. Growth without retention is vanity. Example: ship one onboarding tweak that cuts 30-day churn.
3. Do less. Do the right things. Focus until it hurts. Example: cut three features; double down on the one that moves retention.
4. Don't scale before PMF. Order matters: PSF → PMF → Scaling Fit. Example: stop hiring SDRs until your funnel converts predictably.
5. Capital efficiency is a superpower. Burning cash hides problems; frugality exposes them early. Example: run 6-month runway experiments to force learning.
6. Advisors must reduce founder burden. If they add complexity, they're noise. Example: ask each advisor for one deliverable that saves you time.
7. Early teams must match the stage. Hands-on, customer-close, able to ship. Example: hire a product person who can ship, not a VP title.
8. Most startup deaths are self-inflicted. Choices kill: building too much, hiring too early, scaling before PMF, chasing investors over customers, ignoring unit economics. Example: replace “growth” KPIs with weekly unit-economics checks.
9. Build repeatability before growth. Sales, delivery, and value creation must be repeatable. Example: document a 3-step sales playbook any rep can follow.
10. Lead with predictive metrics, not history-based ones. MRR tells you what already happened. Pipeline conversion, activation rate, and 30-day retention tell you what's coming. Example: swap one lagging KPI in your weekly review for a leading indicator you can actually move this month.
AI and product development: what the data actually says
AI is not a magic shortcut. The most-cited 2025 randomized trial on developer productivity (METR) is a useful reality check: experienced developers using AI tools believed they were 20% faster, and were measured 19% slower. Perceived speed and measured speed are not the same thing. At the same time, AI tooling and inference costs are climbing fast, and many engineering orgs are quietly absorbing the bill.
What this means for founders: treat AI like any other lever. Measure cost per experiment. Track whether faster delivery actually improves retention or unit economics. Test small, measure the metric that matters, then scale. If you can't tell me whether AI made your last release more profitable or just more expensive, you don't have an AI strategy. You have an AI bill.
Pick one rule. Fix one metric this week. Which one is your team breaking right now?