Why companies are becoming a series of loops

Anish Acharya, a general partner at a16z who invests heavily in consumer and AI companies, discusses why AI fears like a 'permanent underclass' are overstated and how companies may become networks of AI 'loops' with humans supplying intuition. The conversation also covers model choice across functions, consumer AI opportunities framed as 'loop, make me happier,' and how durable moats and ambition shape startups.

01Key learnings

• Treat each business function as an agent loop: map its inputs, outputs, and the points where it gets blocked, then supply context, insight, or direction at those points.
• Expect loops to climb local maxima quickly, but plan for humans to supply out-of-distribution intuition that moves you to the next hill.
• When an agent gets stuck, have a human coach it through the problem; the captured traces let the agent handle that case next time.
• Match model tier to upside: frontier models suit high-upside work like research and engineering, while cheaper mid-tier or open-weight models suit bounded-upside functions like legal or finance.
• Build a personal chassis, such as a small app or project, and ship something regularly with each new model to develop real intuition about what models do well.
• Consider what consumers want beyond productivity: connection, love, progress, and fun, since people tend to want to spend time more than save time.
• Moats are often discovered by shipping rather than designed upfront; classic moats like network effects, scale, brand, and proprietary data still apply.
• Be more ambitious in company bets and product pricing; test whether a product could justify a premium price, which is a useful measure of product-market fit.

02Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.