Snowflake's earnings report on Wednesday clearly demonstrated the positive outcomes that traditional software companies can achieve by stacking AI tools onto their existing product offerings. For example, the headlines in AI coding are mostly dominated by Anthropic, OpenAI, and Cursor; however, Snowflake, a data management software company, proved in its earnings that a large number of other vendors can also offer such products.

This is partly because, after Anthropic increased its pricing power earlier this year, enterprise chief information officers have begun looking for ways to control their billing expenses from both Anthropic and OpenAI. Enterprises want coding tools that can connect to multiple large language models, rather than being locked into a single model vendor like Anthropic.

Snowflake stated that in the quarter ending in July, the number of customer accounts using its coding assistant CoCo (formerly known as Cortex Code) grew at least 28% quarter-over-quarter, reaching 9,100 accounts. (Different internal teams within an enterprise may open multiple accounts.) According to a Snowflake spokesperson, CoCo is currently powered by models from Anthropic and OpenAI in its backend, while it is also integrating open-weight and open-source large models such as DeepSeek-V4-Flash and GLM-5.3.

During Wednesday's analyst earnings call, Snowflake's Executive Vice President of Products, Christian Kleinerman, stated: "Many, many customers have told us that after making significant purchase commitments to a particular model vendor, they later realized, 'I should have been able to use other models.' Choosing Snowflake gives customers that flexibility... This is undoubtedly one of our core advantages."

This statement aligns with the views of CEOs at Palantir, Microsoft, and Salesforce: enterprises need a trusted intermediary to connect their business with OpenAI and Anthropic. Snowflake's earnings report sends a broader signal: as AI becomes increasingly adept at generating reports and business insights from databases, the importance of enterprise data management software continues to rise.

CEO Sridhar Ramaswamy stated on the analyst call that the top-tier AI capabilities embedded in CoCo have significantly reduced the difficulty for traditional enterprises to migrate their existing data from legacy systems to modern databases like Snowflake on a large scale.

This earnings report further confirms that in the AI era, cloud software vendors that charge based on actual resource consumption are in a more favorable operating position compared to companies that rely on legacy applications and charge per user seat. In fact, Snowflake has shown very strong growth momentum, with product revenue accelerating for the third consecutive period in the July quarter, growing 37%, allowing it to avoid the pressure that peers like Salesforce are experiencing—namely, the industry-wide discussion around outcome-based pricing.

Some customers want to adopt outcome-based pricing to justify their increased technology spending; while some software suppliers like Salesforce believe this model can help enterprises achieve higher revenues. Claude Code, the leader in AI coding, primarily uses a subscription model with seat-based pricing, but for large customers, it is increasingly incorporating usage-based billing.

Earlier this summer, it was reported that as Anthropic's billing amounts continued to rise, some customers began evaluating alternative solutions or building their own coding tools. Whether Claude Code can maintain its market dominance remains to be seen, but besides Snowflake, other large software companies are also tapping into the AI coding space: Atlassian launched its Jira coding agent in July, and Salesforce launched Slack Code in August. Even OpenAI has lowered the barrier, allowing customers to use non-OpenAI models within its own Codex coding assistant. (Although it is technically possible to connect Anthropic's Claude Code to other AI models, Anthropic does not encourage this practice.)

For Snowflake, CoCo is also driving its core database business: CFO Brian Robbins mentioned on the call that customers using CoCo tend to purchase more of Snowflake's data management products. Snowflake's stock surged nearly 17% on Thursday, bringing its year-to-date gains to over 60%.

AI agents are targeting nearly half of US occupations. New research from the Cohere research division indicates that many jobs remain difficult to automate with AI. The study, titled "The Early Footprint of Automation," completed in May, surveyed public tools developed by developers for AI agents designed to automate various tasks. Cohere found that existing tools target work associated with approximately half of the 923 occupations listed by the US Department of Labor.

The report's authors wrote: "If you ask which occupations will be impacted by agentic AI, looking at it from the supply side, frankly speaking: there are a large number of jobs for which no automation tools have been developed yet." On the other hand, the fact that nearly half of all occupations have been targeted by AI agents may surprise many. (The study did not include internal, non-public AI agent tools developed by enterprises themselves.)

Unsurprisingly, the study points out that occupations with higher risk of displacement include graphic designers and web developers; it also includes certain medical roles such as biostatisticians and clinical data managers—developers have already built tools for AI agents that can handle highly specialized aspects of these jobs. According to the head of Cohere Research, Marzieh Fadaei, this suggests that employment numbers in such roles may contract, but wages for workers in these positions could potentially rise.