AI has killed the pure software moat—hard problems are your only defense

YC partners Tom Blomfield and Matild Sandor advocate that the cost and speed of AI have made pure software startups undefendable. The talk argues that the primary moat is no longer code but choosing a genuinely hard problem—like hardware, regulation, or deep science—coupled with exceptional execution. They observe that building software is now commoditized, so

durable advantage must come from other dimensions: distribution, network effects, or regulatory barriers. A critical counterpoint emerges: this shift redefines founder evaluation and team composition. The partners reveal they are testing whether deep technical skill still differentiates, given that AI enables 'vibe coders' to ship credible products. Yet they note that statistically, solo founders still

underperform, and the need for a co-founder’s emotional support remains amplified even as skill complementarity diminishes. For engineering leaders, the implication is organizational: teams must pivot their hiring and capability-building strategy. If writing code is no longer the bottleneck, then technical leadership must invest in deep domain expertise, sales engineering, and direct customer interaction. The

partners warn that delegating customer conversations to AI or junior staff is a mistake—context and feedback loops stay critical. The hidden tension: the same forces that make startups easier to launch also make them harder to scale. The talk suggests that capital is still required for truly hard problems, and that VC funding will flow