The data cloud company spent four days at the Moscone Center arguing that the age of agentic AI has arrived. The evidence, for once, was harder to dismiss than the hype.
Many tech conferences feel like press events aimed at investors rather than engineers. Snowflake Summit 26 stood out as an exception.
When CEO Sridhar Ramaswamy stepped on stage at Moscone Center on June 1, the mood was noticeably different. Snowflake had just announced Q4 FY2026 product revenue of $1.23 billion, up 30 percent from last year. Its stock had surged 37 percent in a single day and was up more than 60 percent since April. After a year spent defending its position against bigger rivals, this financial turnaround was clear to the 20,000 customers, partners, and analysts in attendance.
Ramaswamy built his four-day keynote around a simple theme: “Making AI Real for Business.” But the commitment behind it was strong. Snowflake, known as the home for enterprise data, now wants to be where enterprise AI takes action. This marks a shift in its business, and the summit made that clear.
The Agents Have Names Now
Snowflake showed its ambitions with a rebrand that made its tools’ audiences clear. Snowflake Intelligence, the AI assistant for knowledge workers, is now called CoWork. Cortex Code, the coding and data-engineering agent, is now officially CoCo—a nickname that started internally and is now official. CoWork is for business analysts who want to query data in plain English and act without writing SQL. CoCo is for developers who want to build pipelines, move workloads, and create applications using natural language prompts.
These changes are more than just a new look. CoWork now connects to Salesforce, Microsoft 365, Slack, and Google Workspace via the Model Context Protocol, and it offers a Deep Research mode that turns all kinds of enterprise data into useful insights. CoCo now works on desktop and mobile, integrates with Slack, and supports Anthropic’s Claude Code. Snowflake says CoCo can reduce data migration from months to days. While that claim needs to be tested at scale, early customer demos were impressive.
Ramaswamy described the agents in a way that will connect with anyone who has seen AI projects stall in procurement: “You’re able to turn ideas into working pipelines, applications, or agents using natural language—these are the building blocks of a system of decision.” While the language is ambitious, most people at the event said the products are closer to that goal than usual.
Also Read: Is AI Becoming America’s Next Security Crisis?
The Infrastructure Story Nobody Put on a Banner
Every technology conference has its headline and its fine print. The headline at Summit 26 was agents. The fine print was four infrastructure announcements that will determine whether the agents actually work — and which received a fraction of the stage time their importance warranted.
Cortex Sense is an automated context layer for every agent on the platform. It scans a company’s data and turns raw metadata from thousands of tables into a shared structure that CoWork, CoCo, and third-party agents can use right away. There is no need for custom knowledge graphs or months of setup. Snowflake says this layer gives agents better performance than other setups. It comes with all Cortex AI products at no extra cost, a pricing move that should help adoption.
Iceberg v3, likely the most important announcement of the week, is now generally available and brings a feature the open data community has wanted: bi-directional writes. Now, external engines like Spark, Trino, or any Iceberg-compatible compute can write back to Snowflake-managed tables using open, standards-based controls, with governance and no data duplication. For companies worried about being locked into one data platform, this is a big change. The Polaris governance layer for these open writes is especially important in regulated industries, where interoperability and access control have often been seen as conflicting needs. Snowflake now says they do not have to be.
Snowflake Datastream, now in private preview, is a fully managed streaming service that works with Kafka and is built into the platform. Existing Kafka producers can connect without changing their code, and data flows into Snowflake tables with governance applied as soon as it arrives. This move puts Snowflake in direct competition with Confluent’s streaming business, which the company says is a $128 billion market. It remains to be seen if companies with complex Kafka setups will find the switch worth it, but Snowflake’s direction is clear.
Adaptive Compute, soon to be generally available, removes the need for warehouse sizing and manual tuning by automatically optimizing compute resources in real time. This helps not just data engineers but especially CFOs who have struggled to predict Snowflake’s cloud costs on the bill.
The Anthropic Alliance
While the infrastructure news was important, the Snowflake-Anthropic partnership was the highlight of the event. Anthropic co-founder and president Daniela Amodei joined Ramaswamy on stage for the opening keynote, continuing a Summit tradition—last year, OpenAI’s Sam Altman was the guest. This year, the partnership details were much clearer. Claude models, including Opus 4.8, became available on Snowflake Cortex AI the same day and now power both CoWork and CoCo in Snowflake’s secure environment. Snowflake is one of six launch partners in Anthropic’s Claude Marketplace, which lets customers use their existing Anthropic commitments to deploy Snowflake.
Amodei’s message on stage resonated with enterprise architects who see AI governance as a key advantage: “Trust is an accelerant. Trust is something that helps you go faster.” Snowflake is betting that enterprises want AI that is governed, auditable, and close to their data, especially as they move from early tests to real-world use.
Snowflake also announced plans to acquire Natoma, an agent security startup, to bring its MCP connectivity infrastructure in-house. This move shows that Snowflake wants to control a key capability rather than rely on third-party integrations.
Also Read: How Network-as-a-Service Can Drive Sustainable IT
The Honest Reckoning
None of this is without tension. Roughly 70 percent of Snowflake’s business runs on AWS, and a freshly inked $6 billion committed-spend agreement with Amazon underscores a dependency that the company’s multi-cloud positioning rhetoric cannot fully obscure. The same hyperscalers supplying Snowflake’s compute substrate are building full-stack AI systems from inference silicon to chat interface. Differentiation, in that environment, cannot be positional — it must be continuous.
What Snowflake is betting, with genuine architectural coherence, is that it can systematically simplify what the hyperscalers make complex: Cortex Sense removes the need for an expert ontologist, Adaptive Compute removes the need for an expert warehouse sizer, and Iceberg v3 removes the need to pick a single engine. It is the same instinct that built the company’s original franchise, applied to a harder problem.
Cortex Sense remains in private preview. Datastream has not yet faced the full weight of enterprise Kafka complexity. The automated-context claims have yet to be stress-tested against the chaotic, underdocumented data estates that characterize most real organizations. These are not small caveats.
But the direction, at Summit 26, was unusually clear. Snowflake is no longer asking to be described as a data warehouse that wishes it were a platform. It is asking to be judged as the agentic control plane for the enterprise — and it spent four days in San Francisco making the most credible case for that role it has ever made.
The scrutiny starts now.
