How AI agents transformed Snowflake into a different company in just 1.5 years

In the fall of 2024, Snowflake developers gained unrestricted access to the internal coding agent Cortex Code. No one forced employees to use it, nor were there mandatory training sessions. Within six weeks, the company became fully AI-literate without a single top-down directive—people simply saw the benefits and began sharing their discoveries with one another. This accidental experiment proved far more illustrative than any top-down technology implementation strategy.

Productivity increased not by percentages, but by orders of magnitude

To date, Snowflake salespeople without engineering backgrounds create visualizations and entire working applications on their own—simply by describing the task in words. The company’s strongest engineers work 50 to 100 times faster than they did a year and a half ago. This is not a gradual increase in efficiency, but a leap that changes the very structure of development teams: now, one agent writes the code, another critiques it, and a third checks for security—all without direct human involvement at every stage.

Snowflake’s own support service was built on top of Cortex Code in six weeks by a single team. After launch, the queue of inquiries virtually disappeared. A similar story unfolded with the infrastructure reliability engineering team: they rewrote the monitoring and alerting system and automated the analysis of Kubernetes logs, a task that previously took four days.

Software is no longer a “sticky” product

Previously, a company could survive on inertia for years: customers rarely switch from familiar software, even if something better appears. The barrier to entry for development has dropped so much that this logic no longer holds. New products are created faster and cheaper, meaning competition for customers becomes sharper, not softer. The industry is transitioning from a mature market state to one of constant turbulence—a shift that affects not only startups but also major players.

Subscription payments are giving way to payment for results

The stock market valuation of software companies has declined in recent months—the market is revising downward the long-term value of businesses built on the classic subscription model. Snowflake is betting on a pay-per-consumption model: clients pay for the actual value received, not in advance for access. To avoid uncontrolled bills, the company is introducing spending limits—both per user and for the entire account as a whole.

Tokens-based payment, which currently underpins the business of large language model developers, is unlikely to remain the only model for long. Demand for developer productivity is so elastic that the market will seek more balanced schemes: some products will return to fixed subscriptions, while others will combine them with token-based payments.

The main restructuring is happening not in the code, but in the company structure

The true effect of AI is manifested not in the speed of writing code, but in how the organization itself is restructured. In a world where the previous hierarchy of approvals is no longer needed, departments lose their meaning in their old form. At Snowflake, the team that prepared product demonstrations for clients was disbanded—now each salesperson creates demos themselves, and freed-up personnel were reassigned to other tasks where there was previously a shortage of hands.

Not all employees accept these changes easily. The company strives to explain the transition gently but directly: mastering AI tools is no longer just a matter of efficiency, but of preserving one’s own value in the labor market in a few years.

Three qualities that matter more than a plan

Ramaswami went from a Google engineer to the head of an advertising division with revenue exceeding $100 billion, then founded the search startup Neeva, which did not find commercial success and was acquired by Snowflake along with its team. From this experience, he draws three things that are more important than any career plan: readiness for constant hard work, the ability to revise one’s own beliefs as circumstances change, and the refusal to impose self-imposed limitations. In interviews for leadership positions at Snowflake, he directly asks candidates what significant changes they have made in themselves over the past year—and considers this the best indicator of readiness to work in conditions of constant change.