Use humans to do AI's work as pre-launch validation
Before building AI/automation features, have humans manually perform those tasks for early customers. This validates that (1) customers will pay for the outcome, (2) the work can be done reliably, (3) you understand all edge cases and data requirements. Document what context, data, and judgment the humans use - this becomes your AI training ground. Better than building AI in vacuum and discovering it doesn't work. Creates hybrid model where humans bridge gaps while you automate incrementally.
When to use
When building AI-powered products or automation tools where you're replacing human labor. When you need to validate demand before investing in ML/AI development. When understanding edge cases is critical to product quality. Best for B2B where customers care more about outcome reliability than whether it's AI or human behind the scenes.
Don't do this
Building AI features before validating humans can do the work reliably - may discover task is harder than expected or customers don't value outcome. Also avoid if you can't afford human labor costs while building automation (need runway). Don't use if customers specifically require NO human in the loop (rare).
1 Founder Who Did This
Launched MVP as just an API and pitch deck with no functional UI. Rezaei personally acted as the AI system, manually analyzing customer data and writing personalized website copy until the dataset was large enough for algorithms.