|
Welcome back! Tech firms ranging from Block to Meta Platforms have announced mass layoffs in the past year, frequently citing AI as allowing them to do more with less people. In a variation on that approach Snowflake says it used layoffs to force software engineers to use AI tools and become more productive. “At the time, my team hated me, but I thought it was a provocative way” to get them to use AI tools, said Snowflake CIO Mike Blandina, speaking during the company’s annual customer conference in San Francisco this week. He said older, senior engineers were more hesitant to use AI coding tools than younger, junior engineers. The executive said he cut roles on his team in January, accelerating a plan to reduce the team slowly throughout the year. “I was given an exit number for the calendar end of ‘26, an exit number of headcount…January 15th of this year, I cut to that number,” the CIO said. The company said in February that it laid off 200 staffers during the three months through Jan. 31, which is a small blip for an enterprise software firm with over 9,000 employees. Blandina said the change was about “proving the concept” that AI tools can create more productivity. “One way to do it is just not have the number of people there, and you’ll know whether the productivity is happening or not,” he said. “I said to the team, if you develop software the old way, we’re all gonna fail…Because we don’t have enough people now.” The move comes as Snowflake has admitted it’s one of many companies including ServiceNow and Uber that are worried over rising spending on AI tools internally. Those worries recently prompted Uber to cap Claude Code usage for its engineers, Bloomberg reported. For its part, Snowflake is reigning in costs by swapping in cheaper models for certain tasks in addition to the layoffs, we reported Tuesday. Snowflake declined to specify which AI coding tools it uses other than its own in-house AI agent, CoCo, which is powered by a number models, primarily Anthropic’s, according to executive vice president Christian Kleinerman. Microsoft’s Pragmatic AI Pitch Tension is growing between AI’s power users and those footing the bill. As employees—especially software engineers—use millions of dollars worth of AI models as they tokenmaxx, executives are getting anxious about whether all that spending is warranted. Companies like Snowflake and Uber have said they’re worried about overspending on AI, while Meta, the originator of the tokenmaxxing trend, has recently told employees to be more selective about their AI use. That dynamic is something Microsoft seemed keenly aware of this week as it rolled out a slew of new AI product announcements at its Build developer conference. Microsoft pitched most of the new products as a more pragmatic and cost-effective option for enterprises. They included “mid-weight” AI models that are cheaper than those of OpenAI and Anthropic and new tools aimed at running more AI locally. Speaking to a room full of app developers, reporters, and Microsoft executives, Microsoft CTO Kevin Scott threw cold water on the idea that accelerating AI progress would neatly translate into value for companies. “We just shouldn‘t have uniform faith that as AI model capabilities improve we’re going to get this crazy fast deployment everywhere, which may be a little bit of a controversial point of view here in Silicon Valley” Scott said. “Just because we can build more complex things than we built before that doesn‘t necessarily mean what we’re building is super valuable, [or] that they’ve increased the top line of the businesses that we're running.” Of course, it behooves Microsoft to make such a pitch given that it can’t currently offer any models as sophisticated as OpenAI’s GPT-5.5 or Anthropic’s Claude Opus 4.8. But as companies get squeezed by rising costs from those vendors, Microsoft’s pitch may start to look more appealing. For instance, Land-O-Lakes has been testing out Microsoft’s new reasoning model, called MAI-Thinking-1, to automate tasks in its butter formulation process. The company used Microsoft’s “tuning” software to customize a copy of the model by feeding it thousands of internal documents, as well as Teams messages and Outlook emails, that human employees had previously written as part of the process. The company found that the customized version of MAI-Thinking-1 was more accurate and ten times more cost-efficient than OpenAI’s GPT-5.5, according to Microsoft senior product manager Tanaya Yadav. (Land-O-Lakes executives themselves weren’t present for the demo; a Land-O-Lakes spokesperson didn’t immediately respond to a request for comment.) Microsoft also debuted new software in its Azure cloud service meant to let companies track their return on investment from AI, including dashboards that show the amount of money agents took to complete given tasks. “You're going to see us investing more in cost and ROI tracking in the coming months,” senior product manager Matt McCleary said. Microsoft isn’t the first AI player to make such a cost-conscious pitch. The AI lab Cohere has similarly focused on selling mid-range models to companies and helping them customize the models while Amazon Web Services has also been pitching its Nova models as affordable and customizable. But unlike those companies, Microsoft has customers that already use its business software for emailing and drafting documents, which could give it a leg up as a natural candidate to customize cheaper models. Whether the pitch sticks with customers for tasks outside of butter formulation remains to be seen.
| 
