Aaron Levie says agentic coding could let engineering teams do three, five or ten times more work, and do it faster.
“you might be able to do it three or five or ten times faster.”
Person
Aaron Levie
5 episodes · 98 quoted claims
- 15 Jun 2026Aaron Levie, Box CEO: Advice for CIOs on AI Agents53m
- 14 Jan 2026How SaaS can survive the AI revolution, with Aaron Levie43m
- 24 May 2025Rapid Response: The enterprise AI revolution is underway, with Box’s Aaron Levie36m
- 20 May 2025The enterprise AI revolution is underway, with Box’s Aaron Levie40m
- 29 Jan 2025Aaron Levie, CEO of Box, on Box AI, Enterprise Enthusiasm, and the Evolution of SaaS50m
Aaron Levie estimates that 60 to 80 percent of engineering time goes into necessary work that does not itself differentiate a project.
“maybe sixty, seventy, 80% of of your engineering time is going into the work that doesn't really actually make the the project differentiated per se.”
Aaron Levie says engineering suits agents because the work is mostly text-based and models are trained on large amounts of code.
“The work is mostly in a text-based medium, so, you know, you're mostly just writing code.”
Aaron Levie says code is verifiable through regression testing, unlike contracts or financial analyses.
“if I build a bunch of code, uh, and I can go and do a regression test on it, I can see, you know, did I break anything or does it still work?”
Aaron Levie says engineers already have access to the code base for their project, so the agents they deploy do too, unlike in the wider enterprise.
“if you're an engineer working on a project, you already have access to the entire code base relevant to that project.”
Aaron Levie warns that organizations relying on security through obscurity can have agents leak data to the wrong people.
“you have too much access to information. Um, and we have sort of, s you know, a security through obscurity, uh, in your organizations, and now all of a sudden the the agent is leaking data, uh, to the wrong people.”
Aaron Levie argues CEOs are the furthest from real work in a company, which makes them prone to overestimating automation.
“Is like a CEO, by definition, is the furthest away from the real work that's happening in the company.”
Aaron Levie says Box tested Fable against Opus 4.8 on a project and that Fable did something Opus 4.8 could not.
“We we had a project just yesterday where we tested Fable versus Opus 4.8. And you'd you'd think, like, okay, these are should be on the same exact trajectory of of progress.”
Aaron Levie describes Fable and GPT 5.5 as a step-function improvement in what coding agents can do.
“If you're in software projects, something like Fable or GPT 5.5 actually is another step function improvement on what these agents were capable of.”
Aaron Levie predicts enterprises will have internal forward-deployed engineers, typically in IT but embedded in lines of business.
“I actually think there's going to be a role for effectively an internal FDE.”
Aaron Levie says forward-deployed engineers could come from vendors such as Box, Salesforce or Palantir, from systems integrators, or from new consulting firms.
“that Ford deployed engineer could come from a vendor like a Box or a Salesforce, or, um, or a Palantir, or it could come from a systems integrator or maybe a new consulting firm.”
Aaron Levie predicts that within one to three years the cost of top frontier models and the next-best models will diverge, making model routing important.
“you're going to see this sort of stratification between the cost of frontier intelligence and the cost of sort of the second best frontier intelligence.”
Aaron Levie says a period of venture-subsidized tokens for coding agents lasted about a year and appears to be ending.
“It was a beautiful moment in history. It lasted about a year, um, and we all had a great time.”
Aaron Levie says IT budgets typically run about 2 to 5 percent of revenue, which he sees as a cap on AI spending unless business units own budgets.
“IT budgets kind of run at depending on the industry, maybe three, five, you know, you know, two to two to 5% of of sort of total revenue”
Aaron Levie says Box does not run an internal token leaderboard or incentivize maximum token use.
“we don't have a leaderboard internally, we don't try and incentivize the most number of tokens.”
Aaron Levie says he uses Salesforce about five times more than before by connecting its MCP server to Claude Cowork.
“I use Salesforce more today than at any point in history, maybe like five times more.”
Aaron Levie predicts there could be about a hundred times more agents than people using software.
“we're going to move to a world where there's going to be, you know, maybe a hundred times more agents than people using software.”
Aaron Levie argues companies should not vibe-code core systems of record such as CRM or ERP.
“we don't think that you should be vibe coding like a core system of record for your documents and enterprise data.”
Aaron Levie warns that wrong prompts and limits could produce a $50,000 token bill overnight.
“you could probably, like, go and spend 50,000 and wake up the next day and, like, have that bill.”
Aaron Levie says that in 2026 Silicon Valley insiders have perhaps an hour's head start on AI information over everyone else.
“Today, uh, in 2026, there's like, you know, maybe I have an hour, you know, heads up of information from from anybody else.”
Aaron Levie says giving an agent too much context is a security risk and also leads to wrong answers because the agent has too much to attend to.
“You can't actually just shove as much context as possible to the agent.”
Aaron Levie says agents may be the most technical solution ever deployed to non-technical people, raising the importance of the CIO role.
“agents are are maybe the most technical solution that has ever been deployed to non-technical people.”
Aaron Levie was an intern at Paramount in 2004, the year he began the college project that became Box.
“got some internships in the entertainment industry, and then I was an intern at Paramount in 2004.”
Box launched its online storage product in early 2005.
“we launched it in early 2005 and then just sort of started growing from there”
Levie says earlier online-storage companies had died off in the late 1990s before Box launched.
“there had been companies that would let you store your data on the internet, but they all kind of died off in the late nineties”
Mark Cuban became involved with Box between Levie's sophomore and junior years of college, when Box was about four months old.
“between our sophomore and junior year of college. Box had been around for about four months or so.”
Levie says Mark Cuban advised Box not to hedge its bets between competing business and technical approaches.
“he basically said, you know, as a startup, just don't hedge your bets.”
Cuban's investment led Levie and his co-founder to drop out of college and work on Box full time.
“That led to actually us dropping out of college and betting on the company full time and pursuing it full time.”
Box pivoted into the enterprise market a couple of years after it started.
“once we pivoted into the enterprise market, which we did relatively early on, a couple years into into the business”
Levie says many banks, pharmaceutical firms and governments initially wanted on-premises software rather than a cloud version.
“there were a lot of enterprises, especially in banks, pharma companies, governments, that didn't want to have a cloud version of technology.”
Levie says cloud computing took about two decades to spread across businesses and the migration is still ongoing.
“It took two decades for that to actually play out across all all businesses, and still, we're still going through the migration phase.”
Levie says Box concluded within about ten days of ChatGPT's release that it should let users ask questions of their enterprise data.
“we realized it basically probably within ten days of ChatGPT”
Levie says Box tried a similar AI question-answering project about six years before ChatGPT but it could not work without very large models.
“like six years prior, we actually had a project to attempt that. It just was never gonna work because until you had, like, 200, 300 billion parameter model”
Box's generative AI effort began as a team of about ten to fifteen people.
“It started small. It was like a ten, f- fifteen person team”
Levie says Box offers agents that go through customers' contracts, invoices and other documents and extract structured data from them.
“give us your contracts, your invoices, your healthcare documents, your research files. And we have agents that go through all of that data and extract all of the structured data from that.”
Levie expects Box to charge a predictable per-user price of roughly $10 to $30 a month, with agent usage priced on top.
“There's gonna be seats for the user that will be a predictable price point, $10 a month, $20 a month, $30 a month, whatever the plan is that you're on.”
Levie says Box is comfortable with customers using agents in ChatGPT or Harvey to work on data stored in Box.
“If you never logged into Box, but you used an agent to interact with that data inside of ChatGPT or Harvey or somewhere else, we're totally comfortable.”
Levie argues that AI agents could increase the number of users on systems of record about a thousandfold.
“my general view is is that what what you basically have is like a thousand x increase in the number of users on these platforms.”
Levie uses a hypothetical of an agent given all company files without permissions to illustrate how it would reveal salaries or acquisition plans.
“you took all of your company's files and you put them on a server and you gave that agent the server.”
Levie estimates AI expands the software market by at least three to five times, possibly ten times.
“it's at least a kind of a three to five x increase in the size of of software, maybe a ten x increase in the size of software.”