Alejandra Vergara was forty minutes into her first meeting with Aria Attar and Biraj Silwal when she wrote four words in the call notes: “We should do this.”
That isn’t how Ale ended meetings. She’s precise. She weighs things. But this was different. Aria and Biraj had spent months talking to companies before building a single line of production code; fifty-plus conversations, from early-stage startups to enterprises like Chevron. They kept hearing the same two things: “We don’t know how to get started.” “We don’t have the resources.” The founders hadn’t gone looking for pain. The pain kept finding them.
They had also mapped the supply side clearly. Roughly two hundred fifty software engineers exist for every one ML engineer in the average company. The bottom ninety percent of companies had no ML capability at all. Their answer was to build autonomous agents that function as data scientists and ML engineers. Two young technical co-founders, one with a background in scalable cloud infrastructure, one out of the BAIR lab at Berkeley. They had met at an ML hackathon. The conviction crystallized there.
Ale brought them back into Bee Partners. Michael joined the next meeting. Reference calls happened. Each session confirmed what the first one suggested: this team was already living in the future, and passing felt like the mistake.
We said yes at our holiday party. Aria and Biraj flew out for it. They hadn’t incorporated yet. Three weeks after Ale’s “We should do this”, Aria’s answer came back: “Let’s do this.”
That company was called PigeonsAI.
Over the next year, the pitch tightened. “Autonomous data scientists and ML engineers” was real but wide. The sharper version was “autonomous data engineers.” Aria saw early what most of the market is still catching up to: agentic AI would reshape enterprise data infrastructure before it reshaped most other functions, because data pipelines are the prerequisite for everything else. You can’t deploy a model against proprietary data without first getting the data right.
PigeonsAI became TensorStax. About fourteen months after our pre-seed, the company closed its Seed. The product had sharpened with the pitch; prospects were converting to customers. The thesis was proving out.
Then, Snowflake came calling.
Snowflake had been watching the agentic data-engineering space. TensorStax fit, and eight months after the Seed, the deal was closed: TensorStax was now part of Snowflake. The technology now ships inside Cortex Code, Snowflake’s product for agentic systems that reason, verify, and adapt pipelines automatically. For Aria and Biraj, saying yes wasn’t complicated. This was the right acquirer, at the right moment, with the reach to put the technology in front of the customers who needed it most.
We are enormously proud of this team. What they built, what they proved, and how they carried themselves through every stage of it. The founding energy at that holiday party was palpable, and they honored it all the way to close.
Congratulations to Aria, to Biraj, and to the full TensorStax team. Congratulations to Glasswing, who led the Seed and helped set up this outcome. And congratulations to Snowflake; they acquired a helluva founding team.