AI Speed Is Undermining Venture Capital's Rule Against Backing Rivals
Benchmark partners acknowledged at TechCrunch Disrupt that venture capital's pact against backing rival startups is eroding under pressure from AI.
The comments came from Eric Vishria, Chetan Puttagunta and Peter Fenton, speaking together on stage at the conference in the week before October 15, according to TechCrunch. Fenton described backing rivals as kryptonite for the firm and its toughest problem. Benchmark has built its business around concentrated, early-stage bets and close board work.
Vishria tied the strain to velocity. Companies are evolving and pivoting quickly in the current market, he said. A seed bet that looked clean at the term sheet can become overlapping twelve months later, without either founder changing thesis in bad faith.
Puttagunta gave the AI-specific mechanism. Near-zero costs for writing software, he said, make founders ambitious to claim broader surface area, meaning the range of tasks their product tries to cover. When code generation is cheap and iteration cycles compress, a team that started in one workflow can move into adjacent workflows, agents that act on their own, and platforms. The result is not two startups choosing to compete. It is two portfolio companies expanding into each other.
Both partners drew a line between that drift and a deliberate conflict. Backing a direct competitor at the outset is an easy no, Puttagunta said. A portfolio company later moving into another portfolio company's turf is different, and harder to police in advance.
Vishria said the firm stays committed to each founder and keeps what it learns confidential. He also downplayed how much a VC could transfer between rivals, saying VCs are not product or technology experts. Fenton placed Benchmark nearer the co-founder end of the investor-involvement spectrum and said transparency is a golden rule, with no surprise after the fact. If a conflict emerges, he said, a founder should never feel shortchanged. Vishria added that exclusivity remains really important at the very early stage Benchmark favors, but things are evolving.
The panel took place days after a public dispute involving Factory and Cognition. Factory, which builds an AI agent to write software, said it was firing board observer and adviser Chris Degnan for unethical conduct involving rival Cognition, maker of the Devin coding agent. Degnan said he had already resigned from Factory and denied sharing confidential information, then announced he was joining Cognition as chief revenue officer. Khosla Ventures backs both Factory and Cognition.
One firm holding positions in two direct competitors in AI coding agents would once have been treated as an exception. The partners noted that at the top of the foundation-model stack, the large base models other apps build on, it is now routine. OpenAI and Anthropic share dozens of the same investors.
The broader context here is that conflict rules were written for a slower map. In enterprise SaaS, or subscription business software, categories stayed legible long enough for a lead investor to check adjacency risk and for founders to accept that a firm would sit out a rival. In AI-native software, distribution, model access and agent scaffolding, the tools that connect models to real tasks, change quarterly. A cap table decision, meaning who is allowed to invest, made at seed can look different by Series A.
In my view, founders should read the Disrupt remarks less as a policy change and more as a disclosure about operating reality. That puts more weight on practical hygiene around board materials, observer rights, product roadmaps shared in updates, and what gets said in troubleshooting calls. Confidentiality commitments still matter, but they work best when paired with narrow sharing.
From a founder's side, the practical point here is that early exclusivity still has contracting value. If wider surface area is the norm, a founder can negotiate clarity about what happens if the firm later backs an adjacent team, who recuses from what, and how support staffing is separated. Transparency as a golden rule only helps if it is specified before the overlap appears.
The long arc here remains constructive for builders. Shared cap tables across OpenAI and Anthropic have not stopped either lab from shipping, and overlapping bets in coding agents have not slowed investment in developer tooling. The pact is narrowing to where it can still be kept, at inception, while later-stage overlap gets managed rather than avoided.
