A friend put me onto the company and recommended I buy the stock, too. He loved the stores, those cavernous athletic-shoe big box superstores you could wander around in for an hour. Peter Lynch’s mantra about “investing in what you know” to spot great stock ideas from your experiences and daily routines was a concept which he subscribed to. It made sense to me, so I created an account at a new online brokerage called E*Trade and bought it.

Shortly afterwards, as it expanded too fast, the company went into bankruptcy and the stock went to zero. I lost a thousand dollars, which at that time for me was real money!

A $1,000 tuition bill for the difference between a view and an edge.

I’d had no real insight. I had walked the aisles of a store in North Carolina, and watching the company open new superstores as fast as it did, it seemed obvious it would just keep on winning. That was a reasonable read. It was also one available to anyone who had walked into the same store and noticed the same thing. That is what a view without an edge feels like.

I have not bought a single-name stock since, instead largely sticking to indexing, which has cost me real upside but protected me from larger losses. This experience taught me the difference between having a view and having an edge. In the public equities market, I only had a view.

In the private venture market, I do believe I have an edge, one built over twenty years of doing this job professionally. However, it must be reimagined as we enter our industry’s AI age.

The clearest map I know here comes from the one corner of finance that already handed its decisions to machines, quantitative hedge funds. Systematic shops like Renaissance Technologies and Two Sigma have spent decades automating the judgment that used to belong to a trader, and it is all too tempting to import that story into venture as “the machine will pick the winners.” A number of thoughtful people have already explored exactly that. Katelyn Donnelly wrote a prediction of quant coming to venture a little while back, AngelList has published what its own data shows about power-law returns, and firms like SignalFire and Hone Capital built models to score seed companies before a partner ever weighed in. Moreover, Correlation Ventures has been practicing an early version of this quantitative approach for well over a decade.

But hedge funds and venture funds are notably different in structure and approach. Data scientist Joe Hovde has laid out why “quant” struggles in venture specifically: venture returns follow a power law, the sample size is tiny, the feedback arrives a decade late, the act of investing bends the outcome, and the biggest winners often looked the most like mistakes. You cannot backtest your way to the next category-defining founder. Simply stated, the machine does not pick in venture the way it trades in equities.

Yet the real transferable lesson from hedge funds is quieter. In a mature quant shop, humans have a more important role than picking or executing trades. Instead, they invent the hypotheses to act on the data that only they own.

Any general signal you can simply buy, the same feed your competitor can buy, decays to zero edge the moment enough people hold it. Chamath Palihapitiya recently warned that most companies with AI “bleed their edge into a model while believing they are building one.” The only alpha that does not decay is proprietary data multiplied by a proprietary hypothesis.

The same dynamic has always operated inside venture. Years ago I founded a quarterly event, the Web Innovators Group, where hundreds of people gathered to watch startups demo. For a long stretch it was a real edge, wiring me into the local ecosystem and putting me in front of companies before anyone else. Then it faded, the way every conference eventually does, as the startup world filled with places to gather and a copyable format hit its half-life. The event was distribution, buyable and copyable both, that fed me deal flow for a while. What endured as a long-term edge was the tacit read underneath it, the pattern sense from sitting through all those rooms, meeting every company’s founder and hearing the story.

That is how you build a decision engine when the models themselves are a commodity, rented by everyone from the same few providers: Anthropic, OpenAI, and the other frontier labs. Your competitor firm is calling the identical model, so the edge has to live where the model cannot reach, in the conjectures about what will matter that no one else is testing, and in the record of every call you have made and scored. A better-funded rival showing up in 2028 with superior models and an empty archive does not have that. You rent the model. You own the judgment and the record it leaves behind.

When AI makes everything downstream either rented or copied, the intelligence from the model providers and the distribution that used to come from “proprietary” deal-flow networks, insight is the only thing a firm keeps that never commoditizes. Rent the same intelligence everyone else rents, and a partner’s processing power stops setting her apart. Her judgment is the whole edge, the call about which conjecture is worth testing and which founder is worth backing. That is the pick, and in an AI-automated firm it matters more, not less, just as the rest of the job falls away around it.

A VC partner’s job isn’t going away, but it certainly changes drastically. Automate the whole venture capital engine with artificial intelligence, and the partner’s judgment moves upstream to the only ground where alpha was always going to live: in the human insight.