Corporate America is taking a lesson from Silicon Valley: Fail fast

• Companies are adopting a "fail fast" mentality to stay competitive amid the AI boom.
• Execs from companies like Okta, Salesforce, Blackstone, and Snowflake share their take on the strategy.
• While AI enables some companies to move more quickly, it also carries risks.

For some companies building with AI, failure isn't a setback. It's the strategy.

Okta, Salesforce, Snowflake, Blackstone, and others are rapidly launching AI products, fully expecting that not all of them will pan out. Executives say they can't afford to meticulously test ideas in isolation if they want to outpace their competitors, and AI makes it easy to deploy new offerings at scale anyway.

"It's increased the pace of what's possible," said Eric Kelleher, president and chief operating officer of digital-identity company Okta, which has rolled out around 120 AI pilots and features over the past two years, keeping half and scrapping the rest. "You can spin up a lot more pilots at the same time."

This so-called fail-fast method isn't new, as tech startups have been doing it for decades. It's now gaining traction among larger, established businesses due to the AI boom.

Leaders say they're under pressure from their boards and shareholders to justify their AI spending, and that the fail-fast approach is helping them meet the moment.

"The stakes are higher, the risks are bigger, and to keep pace with the innovation churn, we need to fail faster than ever," Cisco's senior vice president and global innovation officer, Guy Diedrich, told Business Insider.

Everything is faster

For some companies, fail fast isn't just a mindset — it represents a new reality driven by AI efficiency gains.

Snowflake CEO Sridhar Ramaswamy said the cloud-data platform company has rolled out numerous pilot programs, often moving from a planning meeting on Monday to a demo by Friday. With its agentic support system, he said, many cases that once took hours to debug can now be resolved in about 20 minutes.

Similarly, something like preparing a customer briefing for an executive used to take hours. Now, Ramaswamy said, it can be done in a minute by asking a sales agent.

John Stecher, the chief technology officer at Blackstone, told Business Insider that the "fail fast" mindset has long been a mainstay in the engineering world, but AI tools have enabled companies to accelerate development timelines.

"Our software engineers, by quantifiable metrics, are about 2x more effective than they were in 2024 with writing code," Stecher said, adding that they now have the ability to deliver faster and higher quality work.

He said the "blocking-and-tackling part of software development and engineering," including tasks like writing unit tests and higher-quality code, allows engineers to "deploy more robust," reliable, and secure code, so that workers spend less time fixing bugs and more time building. That increase in productivity allows teams to build prototypes more quickly and launch them to users faster.

Ramaswamy similarly said that a lot of the heavy lifting on engineering teams for tasks like boilerplate code, tests, and migrations happens much faster. That frees workers to spend more time on tasks that require judgment, architecture, and design, he said.

The downside of failing fast

Moving at warp speed comes with risks, however.

Last year, Salesforce launched an AI agent designed to answer customer queries, and it recommended a competitor's application to a user. The user then wrote about the snafu on LinkedIn, said Joe Inzerillo, president of enterprise and AI at the software company.

Another time, one of Salesforce's newly released AI agents provided a customer with false information gleaned from old web pages that had been removed from its website but were still crawlable by search engines. The error was discovered during a review of conversations the agent was having with users.

Though both incidents exposed early-stage gaps to users, Inzerillo said they also revealed important insights into how users interact with the products and how experiences can be improved. "We're going into every single deployment as an experiment to try to learn something," said Inzerillo.

The consequences of failing fast, however, can be worse if companies overlook important security measures.

Stecher said companies in general shouldn't necessarily deploy more products just because they can now do so more easily. He said firms need a "robust checks and balances" system and have data privacy and security controls in place from the get-go, or they'll pay the price with outages or cyber issues.

Snowflake's Ramaswamy said that while the "fail fast" concept is reappearing, the current version is "less about recklessness and more about shortening the feedback loop." He said the phrase "iterating rapidly" is a more accurate way to describe the company's approach to building with AI.

"We identify areas that we know are going to be generally positively impacted by AI, but set up small, almost weekly, goals," Ramaswamy said. "This way, we can adjust and know when things are working well and when we need to course correct."