The New Founder Skill Is Conviction, Not Execution

For most of startup history, the most valuable founder attribute was execution speed. The founder who could write code, talk to customers, and iterate simultaneously had a structural advantage that was difficult to match. Investors converged on "formidable executor" as the primary signal of likely success. Shipping quickly — getting to a testable version before competitors, iterating faster on feedback — was the primary lever of competitive differentiation. The gap between what a great executor and an average one could accomplish in a given period was measured in months, which in startup time was often decisive. Agentic AI is collapsing this gap. When AI can write production-quality code, synthesize customer feedback, draft copy, and run tests autonomously, execution velocity becomes table stakes rather than a moat. What remains scarce is the quality of the founder's judgment about which problems are worth solving in the first place. Paul Graham's insight — that good startup ideas come not from trying to think them up but from becoming the kind of person who naturally notices them — is essentially a description of cultivated conviction: domain-specific beliefs, held with enough confidence to act on, that are hard to replicate because they are rooted in lived experience and pattern recognition. This is precisely the kind of tacit knowledge that automation cannot replace. As Michael Polanyi observed, we can know more than we can tell — and what experts know but cannot fully articulate is exactly what AI systems cannot absorb from explicit instruction. AI can execute a task you define clearly; it cannot originate the conviction that the task is worth doing. The objection is that conviction without execution is worthless — and historically, that is correct. But the objection misses the direction in which scarcity is shifting. We are entering a period where execution is abundant and conviction is rare. Multiple teams can now test the same hypothesis at similar speed with similar tools. What separates them is the quality of the hypothesis — how specifically and verifiably the founder has thought about the problem before writing a line of code. Bad convictions will fail faster with AI than without it; good convictions will compound faster. The new founder advantage is not moving faster, but starting from a better-calibrated belief about what direction is worth moving in at all.