Aaron Levie
Episodes
Notes
a16z · The New Startup Playbook | Box CEO Aaron Levie
-
Why Will Better AI Models Help Companies More Than Consumers?
Aaron Levie says the next gains from better models land in companies, not in most people's personal lives. A point from Dario at Anthropic makes the split: moving a chemistry model from undergrad level to grad-student level will not change what a consumer asks in Claude. Almost no personal chemistry question needs that jump. The same jump is a large gain for Pfizer and other life sciences companies.
Personal use is already full. His mother's tasks are solved, and she would be fine if GPT-5 were the last model, because she is not inventing a drug. Songs, drawings, and stories for a six-year-old do not need to get better. A law firm is the other case. A contract run through a model can still
-
What Happens When AI Labor Gets Much Cheaper?
Aaron Levie says cheaper AI labor can unlock demand that the old price kept off the market. If a kind of work could be bought at a tenth of the old cost, there could be about 50 times the demand for it. High-end legal review at $2,000 an hour, or $1,000 an hour, can be done by an agent for $5 or $10 an hour. People did not hire as many lawyers as the work required, because the price shut them out. At the lower price, more review happens, including more patent filings.
The same shift shows up inside a company budget. Hiring a person can cost $100,000, $200,000, or $300,000, depending on the job. An agent might cost $1,000. Work that always lost the budget fight gets funded, because there is
-
Why Do Strategy and Taste Matter More Than Coding Speed?
Aaron Levie says the premium in a startup has moved off coding speed. Hacking something together used to be the scarce skill. That work is fairly commoditized now, because agents can do it from a clear spec.
The new leverage is knowing what to build. Startups that get ahead write very clear prompts and in-depth specifications, the practice called spec-driven development. Being good at the spec, and being clear about the product, is extreme leverage. The questions that matter are whether there is a real market, whether the idea is good, whether it looks good, and whether it has taste. Companies that are religious about getting strategy and taste right are the ones that pull ahead.
-
Can a 10-person Startup Beat a Company With Hundreds of Employees?
Aaron Levie says a three-, five-, or ten-person team can now produce the output of a 50-, 100-, or 200-person team, and that changes how a startup is run. Founders and early employees are becoming managers of work that agents do. They are no longer limited by how fast they can type code, marketing assets, or blog posts. One founder reviews thousands of lines a day that an agent wrote. Reviewing and orchestrating is about ten times the leverage of creating the work, because many agents can run at once and the person checks them when they finish.
A startup used to be two or three people against an incumbent with about a hundred times more people. The small team could move faster, and the big
-
Does AI Change Startup Distribution?
Aaron Levie says the take might look wrong in five years, and distribution does not change that much with AI. New channels will show up, including a place in the OpenAI app store, but AI is not a fundamental transformation or disruption of the tech distribution model, and it is not a go-to-market disruption.
A consumer company still has to figure out how to go viral. An enterprise company still has to build a go-to-market machine. When the buyer is technically savvy and can self-serve, the path is PLG plus a go-to-market engine. There will be no new invention of go-to-market because of AI.
An amazing agent does not bring customers on its own. The startup gets drowned out by 10,000 other st
-
How Much AI Coding Market Cap Stays With Startups?
Aaron Levie is comfortable underwriting a couple hundred billion dollars of market cap in the AI coding space that do not go to incumbents. Coding agents are the first knowledge-work tools to hit critical mass, and the same opening exists in every other field.
Nearly infinite startup opportunity follows a complete shift in software. Industries and parts of business that previously had no software will have software for the first time as AI agents, and most domains of knowledge work will have an agent for that work. Incumbents will take some categories, while startups can still build knowledge-work agents in every category. Replit, Cursor, Windsurf, Lovable, and Cognition are the clear early
-
When Did Aaron Levie Drop Out of College?
Aaron Levie dropped out of college in the middle of junior year in 2005, when being 20 and doing a startup was still fairly rare. A few dozen people in Silicon Valley were doing it. He thinks it was prompted probably mostly by Facebook having made dropping out a thing, and it was still a relatively rare event. The number of 20 year olds dropping out is now a standard playbook. How many young people are getting started and jumping straight into building startups is super exciting, and he still feels pretty old.
The playbook is incredibly different. The way startups are being run right now is the most different of any period of technological change he has ever seen. Running Box in 2005 compar