Key Takeaways

  • Snowflake posted $1.49 billion in quarterly revenue, growing 37% year-over-year while rolling out AI products like Koko and Co-work.
  • The breakout builder product Koko ran with fewer than five engineers for most of its build cycle, proving that small teams can deliver massive enterprise products.
  • Company expansion no longer tracks employee headcount; AI systems handle tasks that previously required large operational teams.
  • Token budgets should focus on measurable business output rather than blind spending or artificial caps.
  • Snowflake acquired NATO to anchor Model Context Protocol (MCP) integrations, feeding real-time context from Slack and email directly into active workflows.

Five Engineers, Enterprise Scale

For two decades, tech companies measured their strength by headcount. If you wanted to ship a major enterprise product line, you hired fifty engineers, ten product managers, and four layers of directors. Sridhar Ramaswamy says those days are finished.

During Snowflake's $1.49 billion quarter, which grew 37% year-over-year, Ramaswamy pointed to internal evidence: “Koko for example, it's a breakout success but for much of its existence it never had more than five people and even now the team that works on Koko is tiny but that's the kind of impact that is possible today.”

When a team of five builds software that serves global enterprise clients, organizational design changes completely. You no longer build pyramids of managers to coordinate human labor. You build tight engineering pods that point automated systems at complex task graphs. The bottleneck is no longer how many hands you have on keyboards; it is how clearly your team defines what the software should do.

Spend Tokens, Not Headcount

Many finance departments look at model inference bills with panic. They see hundreds of thousands of dollars going to API credits and try to clamp down on usage. Ramaswamy takes the opposite stance. Token spend is simply the new labor line item, and it is vastly cheaper than adding payroll.

As Ramaswamy explained: “My take overall is that the money that we are spending on AI tokens is well worth the cost we continuously optimize absolutely we don't tell people to token max or do dumb things like that it is about driving real results impact.”

The goal is not to burn compute for the sake of activity. It is about swapping expensive human coordination for automated loops. When an agent handles data preparation, test generation, and customer context retrieval, the compute cost looks trivial next to the cost of an expanded engineering department.

Why Context Beats Model Tuning

Founders often waste months trying to fine-tune open weights or pick the perfect model routing architecture. Snowflake is focusing on a different layer: context integration.

By acquiring NATO, Snowflake doubled down on Model Context Protocol (MCP) to bring external workplace activity directly into runtime environments. As Ramaswamy noted: “We are thrilled that we bought NATO because MCP is increasingly really, really important for Snowflake and all of our customers because it provides the real-time context of everything that's happening in your life whether it's Slack or email.”

Models are commodities. The system with direct access to what happened in your team's Slack channel twenty minutes ago will always produce better work than a generic model running on stale documentation. Connect your agents to live communication channels, give them explicit execution boundaries, and let them work.

What to Do With This

Audit your next planned engineering hire this week. Map the exact tasks intended for that role into an automated agent workflow connected to your team's issue tracker. If an AI pipeline can complete 70% of the rote task graph, cancel the job requisition and redirect that budget into API tokens.