Snowflake CIO Says He Used Layoffs to Force Engineers onto AI Tools

Plenty of tech companies have cited AI while cutting staff over the past year, from Block to Meta Platforms, generally arguing the technology lets them do more with fewer people. Snowflake's chief information officer described something more pointed: using layoffs deliberately to push his software engineers onto AI tools.

"At the time, my team hated me, but I thought it was a provocative way" to get engineers using AI tools, Snowflake CIO Mike Blandina said during the company's annual customer conference in San Francisco, as reported by The Information. He said older, senior engineers had been more hesitant to adopt AI coding tools than their younger, more junior colleagues.

Blandina said he made the cuts all at once in January, pulling forward a reduction that had been planned to happen gradually over 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," he said, per The Information. The company disclosed in February that it laid off 200 staffers in the three months through Jan. 31, a small share of a workforce exceeding 9,000 employees.

He framed the move as a forcing function. The cuts were about "proving the concept" that AI tools can drive productivity, he told The Information: "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." His message to the team that remained was blunt: "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 playbook behind the cuts

Speaking separately to theCUBE by SiliconANGLE from the conference floor, Blandina laid out the change-management approach that surrounds that hard edge, and it is more carrot-and-stick than pure stick.

"It's sort of a two sticks and a carrot approach, and eventually you get to one stick and one, a stick and a carrot," he told theCUBE. The sequence matters to him: measure first, encourage later. "Behavior's got to come first because I think everyone's exploring and learning," he said. "You stop measuring and you start encouraging and you start rewarding based on behavior."

He was candid that the resistance skews senior. Some engineers "take the AI red pill really quickly and others not swallow it very well," he told theCUBE, adding that more senior engineers are often "steeped in the dogma of software development life cycles" and "may feel less valuable just writing in English."

Blandina has been modeling the behavior himself. He recounted building a full pricing and billing platform over five weekend sessions of about four hours each, after telling a skeptical team to scrap the old system rather than remodel it. "I literally sat in a room and said, 'Burn it down. You have my permission,'" he told theCUBE. Using Snowflake's in-house coding agent, Coco, he said he produced "a full working system, 60 entities, a full React front end, all the APIs, a full batch process," then challenged his engineers to demo their version next to his as a way to feel what they were experiencing.

The organizational changes have been just as aggressive. Snowflake compressed its agile sprint from three weeks to one, a shift Blandina said "people's heads exploded" over in the first month or two before the team came around. He noted he had run two-week sprints at JP Morgan before joining Snowflake, and pointed to former boss Jamie Dimon's top-down AI mandate as a reference point. "When Jamie speaks, people listen," he told theCUBE, agreeing with host Dave Vellante's characterization of the JPMorgan edict as "Thou shalt embrace AI" or face trouble.

Build-anything, then rein it in

Snowflake's internal-first philosophy runs through all of it. Blandina said his team builds AI capabilities on the company's own Data Cloud, and the approaches that work internally get folded into the product so customers do not have to build them, he told theCUBE. Five years ago he would ask his team to write a product spec; now, he said, he tells them to "type their thoughts into CoCo," which turns the notes into a markdown document teams can build from immediately.

That extended to a company-wide experiment. Snowflake gave most of the company, including its sales team, access to build in Coco, Blandina told theCUBE, and "they built thousands of streamlined dashboards." The company later "reigned in the chaos a little bit," he said, but the exercise taught non-engineers, including finance, what it means to build with AI and made them better partners to engineering.

What his team ships externally

The tools Blandina is forcing internally are also the company's pitch to customers. During the Snowflake Summit 2026 platform keynote, EVP of Product Christian Kleinerman unveiled the rebrands at the center of that strategy: Cortex Code is now Coco, and Snowflake Intelligence is now Snowflake Co-work, positioned as the "control planes" for what the company calls the agentic enterprise.

Among the keynote announcements: a new runtime capability called Cortex Sense that Kleinerman said pushed agent accuracy in one internal eval to 83% from 24%, while cautioning it was a single eval set; the acquisition of automation platform n8n to connect Coco and Co-work to more than 100 business systems; a Coco desktop release so users no longer need the command line; and the addition of SpaceX's AI models to Cortex. Kleinerman closed by framing the platform as a move "from the era of can we to shall we."

The cost anxiety underneath

The productivity drive is unfolding against a competing worry: what all this AI consumption costs. Snowflake has acknowledged it is among the companies, alongside ServiceNow and Uber, concerned about ballooning internal AI spending, according to The Information. Those concerns recently led Uber to cap its engineers' usage of Claude Code, Bloomberg reported, per The Information, and prompted Meta to tell employees to be more selective about AI use after popularizing the "tokenmaxxing" trend. Snowflake is managing its own bill partly by swapping in cheaper models for certain tasks, in addition to the layoffs, The Information reported.

That tension surfaced on the conference floor too. Clover (Fiserv) SVP and CIO Vinayak Kagalkar, interviewed alongside Blandina by theCUBE, said his engineering teams have collapsed software-development cycles that once took two to three weeks down to half a day, but flagged the catch: "the token discussion is triggering now last couple of weeks." Kagalkar said Clover processes roughly $330 billion in gross payment volume across about one million merchants in 12 countries and has partnered with Snowflake for eight years.

The squeeze is creating an opening for cheaper alternatives. At its Build conference the same week, Microsoft pitched "mid-weight" models and local-AI tooling as a more cost-effective option, with CTO Kevin Scott warning that improving AI capability would not automatically translate into business value, according to The Information. The Information reported that Snowflake declined to specify which third-party coding tools it uses beyond Coco, which EVP Christian Kleinerman said is powered by several models, primarily Anthropic's.

The engineers who stayed are the experiment

Blandina's candor about why he cut staff is unusual, but it lands on a contested premise. A Gartner analysis reported by CIO found that 80% of large enterprises reported workforce reductions after launching automation projects, yet identified no correlation between those layoffs and AI return on investment. By Blandina's own account, the productivity gains are still the thing being proved, which leaves the engineers who stayed as the live test of whether removing people really does force the future faster.