The AI market isn't the problem.
Poor validation is.
Before writing code or building an MVP, I run every AI idea through these stress tests.
Here are the exact prompts.
1️⃣ Expansion vs Replacement Test
Prompt:
Analyze this product idea: [insert idea].
Is AI expanding an existing workflow or replacing the core value of an existing SaaS category?
If replacing — which current products become vulnerable and why?
If the answer is "thin layer over LLM," risk is high.
2️⃣ Commodity Risk Test
Evaluate whether this idea can be easily replicated using a generic LLM + simple UI.
What defensible elements are missing?
If it can be rebuilt in two weeks with an API, there's no moat.
3️⃣ Real Cost Structure Test
Estimate the cost structure of this AI product at 10k, 100k, and 1M users.
Include LLM inference assumptions and margin pressure scenarios.
Most AI founders underestimate compute.
4️⃣ Unit Economics Alignment
Suggest possible pricing models (seat-based, usage-based, outcome-based).
Which model aligns best with the value delivered?
If value ≠ number of users, seat-based pricing will break.
5️⃣ Workflow Control Test
Does this idea control a critical workflow or function as an optional tool?
What happens if users stop using it for a week?
If nothing critical breaks, it's not a system-level product.
6️⃣ Data Moat Test
What proprietary data advantages would this product need to become defensible in 3–5 years?
Without a clear answer, long-term defensibility is weak.
7️⃣ GTM Reality Check
Identify the most realistic first customer segment.
Why would they urgently switch?
No pain, no business.
8️⃣ Incumbent Reaction Test
If a large incumbent adds similar AI capability, does this startup survive? Why or why not?
If the answer is no — it's a feature, not a company.
9️⃣ Automation Depth Test
What percentage of the target workflow can realistically be automated in the next 2–3 years?
AI often automates 30%, while founders promise 90%.
🔟 Downturn Survival Test
In a market correction scenario, which 30% of AI startups survive — and would this idea be among them?
Final thought
AI isn't the challenge.
The illusion of differentiation is.
If your idea doesn't pass at least 6 out of 10 stress tests,
it's not a startup yet — it's an experiment.
