Korva is a scroll of the internet’s best ideas.
Bet bigger when value is exponential
Flipping the portfolio
A traditional growth team puts 60 to 70 percent of its time into small and medium bets and 20 to 30 percent into larger swings. Anthropic's team flips that, closer to 70/30 or 50/50 toward the big ones. A grocery delivery app might add 30 to 50 percent more value for its users in two years, and small, high-conviction optimizations can capture a decent share of that. An AI product could deliver 100 to 1,000 times what it does today.
Agentic coding
Agentic coding shows the pattern. It didn't exist a year and a half ago, and its value is already bigger than the earlier market for AI coding. The team still runs the small optimizations, because the compounding value is real and nobody else will do them.
The Chrome extension
One big swing came from the growth team itself: the Chrome extension that now underpins a number of use cases in Cowork and Claude Code. An engineer was bullish on it, no one else was building it, and it became a research-heavy AI product. The approach fits companies whose core value rests on AI, such as Lovable or Cursor, and fits less well when AI is a side feature.
Why do most growth teams favor small bets?
A mature product sits near a local peak, so small changes to it reliably produce small, measurable gains. A/B tests suit those changes because they can detect small effects with enough traffic in a few weeks. Large changes are harder to isolate in a controlled experiment, and their results are slower to arrive. Kohavi, Tang, and Xu describe this large-scale approach to incremental testing at Microsoft, Google, and Amazon in Trustworthy Online Controlled Experiments.
Why do a few winners dominate returns?
In venture capital, a small number of investments typically produce most of a fund's total gain, a pattern called a power law. Funds accept that many bets will fail because one outsized winner can return more than all the losers cost. The same shape shows up in product work when a feature opens a whole new use case instead of lifting an existing one a few percent.
Why does a browser extension help AI agents?
A Chrome extension runs inside the user's own browser, so it can act in sessions where the person is already logged in. Its content scripts can read what a page displays and interact with its elements, such as clicking buttons and filling fields. That gives an AI agent a way to work with sites that have no programming interface. Chrome's extension platform controls this access through the permissions declared in the extension's manifest.
More from Lenny's Podcast: Product | Growth | Career
