|
Snowflake’s strong quarterly report on Wednesday said a lot about what can go right for older software firms that sell AI tools on top of their existing products.
While Anthropic, OpenAI and Cursor get the lion’s share of headlines about AI coding, for instance, Snowflake, whose primary business is selling data management software, demonstrated in its earnings that there’s a much longer list of firms that can provide such products. That’s in part because corporate chief information officers are looking for cost-savings after a fast run-up of AI spending earlier in the year, and they may want coding tools that can access a multitude of models rather than getting locked into one controlled by a single model maker like Anthropic.
Snowflake said Wednesday that the number of customer accounts using CoCo, its coding assistant previously known as Cortex Code, grew at least 28% in the July quarter to 9,100, compared to the previous period. (Enterprises may have multiple accounts for different teams within their firms). CoCo currently relies on models from Anthropic and OpenAI but is adding open-weight and open-source source models such as DeepSeek-V4-Flash and GLM-5.3, a spokesperson said.
“We have heard from many, many customers that they made large commitments to one specific model company and later on are saying, ‘Oh, I should have wanted to [use] a different model,’” said Christian Kleinerman, Snowflake’s executive vice president of product, in an earnings call Wednesday with analysts. “Whereas the commitment to Snowflake gives them that flexibility…. So definitely a very strong advantage for us.”
The comments mirror those made by the CEOs of Palantir, Microsoft and Salesforce: enterprises need a trusted intermediary between themselves and OpenAI or Anthropic.
Snowflake’s results have broader implications. They show how important software that helps companies manage their enterprise data is becoming as AI gets more adept at generating reports and insights from databases. The state-of-the-art AI powering CoCo has made it much easier for older companies to port their data en masse from old systems to newer databases like Snowflake’s, CEO Sridhar Ramaswamy said in the call with analysts.
The earnings report also provided further proof that cloud software firms that primarily charged customers based on how much software they consumed were in better shape for the AI era than those with legacy apps and seat-based pricing tied to the number of employees that use the apps.
In fact, Snowflake’s growth has been so strong—its product revenue accelerated for the third successive period to 37% in the July quarter—that it has allowed the company to defy the pressure other firms like Salesforce have been under to charge companies only when the AI works, known as outcome-based pricing.
In some cases, customers want outcome-based prices to justify increased spending, but some software providers like Salesforce believe it will allow them to generate more revenue.
When it comes to AI coding, market leader Claude Code primarily uses seat-based pricing but increasingly bills based on consumption too, especially for large customers. As their Anthropic bills have taken off, some customers have become increasingly interested in using open source AI coding alternatives or building their own, I reported this summer.
It remains to be seen whether Claude Code will loosen its grip on the market, but other big software firms besides Snowflake are going after the coding opportunity. Atlassian introduced its Jira coding agent in July, while Salesforce launched Slack Code in August. Even OpenAI has made it easy for customers to use non-OpenAI models with its Codex coding assistant. (Technically, you can hook up Anthropic’s Claude Code to run different AI models, but Anthropic doesn’t encourage this).
For Snowflake’s part, CoCo is also helping boost its core business as a database provider: CFO Brian Robins said in the call with analysts that customers using CoCo also used more of Snowflake’s data management products. Snowflake stock soared nearly 17% on Thursday, putting it up more than 60% for the year.
AI Agents Are Targeting Nearly Half of U.S. Occupations
In case it wasn’t obvious, many jobs aren’t yet susceptible to automation by AI, a new study from Cohere’s research group suggests.
The study, conducted in May and published Thursday, looked at the publicly available tools that application developers have created for so-called AI agents to automate different types of work. Cohere found that these tools aim to automate work associated with roughly half the 923 unique occupations listed by the U.S. Department of Labor.
“For anyone asking which jobs agentic AI is coming for, the honest answer from the supply side is that for a great many of them, no one is building anything yet,” the Cohere authors wrote.
On the other hand, some people might be surprised to learn that nearly half of jobs could be in the crosshairs of agents. (The study didn’t cover the proprietary AI agent tools companies are developing behind the scenes to automate various work.)
Cohere’s data suggests some of the jobs more at risk of displacement are, unsurprisingly, graphic designers and web developers, but also some health care professions such as biostatisticians and clinical data managers, because there are tools that have been created by developers for AI agents that can automate the more specialized parts of their jobs.
That could mean a smaller pool of jobs available for those roles, but potentially higher wages, according to Marzieh Fadaee, who runs the Cohere research group.
|
