By Trent Foley
Most Snowflake Summit recaps rank announcements by technical novelty. Wrong lens. What matters is which capabilities change a business outcome this year: revenue captured faster, risk removed, decisions made in hours instead of weeks.
I watched this keynote from a particular vantage point. Over the last couple months I’ve been promoting Snowflake CoCo and agentic development at Snowflake Data for Breakfast presentations and hands-on labs across the country: Salt Lake City, Philadelphia, Washington DC, Raleigh, Austin, and Dallas, to name a few. At Summit itself, I had the privilege of sharing the stage with Sai Kesanapalli, Chief Data Officer at ADI Global, for our session on ADI Global’s AI-assisted Snowflake development journey. More on that below, because what Sai has built is the best evidence I’ve seen for where this is all going.
Housekeeping first: Snowflake renamed two flagships. Cortex Code is now CoCo. Snowflake Intelligence is now Snowflake Cowork. Here’s the ranking, ordered by business value to the enterprise, not applause volume.
Snowflake Cowork’s Personal Work Engine and Cortex Sense
The biggest announcement isn’t a feature. It’s a shift in who gets to act on data.
Cowork now ships a personal agent with multi-agent orchestration, user memory, personal skills and connectors, scheduled tasks, and shareable governed artifacts. Cortex Sense sits underneath, automatically harvesting context from data and activity already in your account. On Snowflake’s own eval, agent accuracy jumped from 24% to 83% with Sense enabled. One eval set, so treat it as directional, but the direction is the point: context infrastructure, not model quality, is the bottleneck in enterprise AI.
Where this lands: a national wealth management network is building an advisor intelligence hub. Today, an advisor question about client concentration routes through an analyst queue. With a personal work engine, every advisor becomes their own analyst, governance intact. Snowflake’s customer Samsung put roughly 1,000 executives and marketers on this pattern. The business case isn’t headcount, it’s decision latency. When a market signal triggers action in hours instead of weeks, you compete on speed.
Agentic Search: Precise Answers From Unstructured Data
RAG gives you “here are the top five relevant documents.” Agentic Search gives you “47 contracts meet that condition.” It extracts facts from unstructured data, runs an analytical query, and returns a precise number. That’s the gap between a research aid and a system of record.
Where this lands: a national healthcare provider modernizing revenue cycle management. Payer contracts, denial letters, and clinical documentation are unstructured, and the questions that matter are analytical: how many denials cite this clause, which contracts have escalators triggering next quarter. Same pattern for a global private credit manager whose covenant packages consume analyst-days every quarter. If Agentic Search holds up in production, that review collapses from days to hours per portfolio. That’s working capital and risk visibility, not efficiency at the margins.
Snowflake CoCo: Skill Catalog, Desktop, and Autonomous Agents
CoCo got the deepest set of announcements at Summit: a desktop app now generally available, sandboxed execution in Snowsight and the CLI, scheduled autonomous agents through async APIs, new form factors including an Excel plugin, and the one with the longest strategic tail: a skill catalog for sharing, discovering, and reusing skills across the organization.
Every room on that Data for Breakfast and hands-on lab circuit asks the same question: does this hold up in production, at enterprise scale, with real governance? At Summit I answered it by stepping aside and letting a practitioner take the floor.
Thank you, Sai, for joining me on stage for session AI285. The session was generous in the way that matters most: Sai shared the real numbers. His team at ADI Global refactored a 3,000-line legacy stored procedure in under an hour. They shipped roughly 300,000 lines of production code in ten to twelve weeks. They built monitoring dashboards, data quality tooling, and cost tracking with CoCo doing the heavy lifting, and used Cortex Analyst to surface a delinquent-customer pattern their existing reporting had missed. That last one is the detail I keep repeating. It’s not a productivity story. It’s AI catching revenue risk a human workflow walked past.
Sai is one of the most advanced CoCo practitioners on the platform, and ADI Global is proof that agentic development is a production discipline, not a demo category. His framing matched ours: AI amplifies the engineer, which means it can also build tech debt at machine speed. The answer is discipline, specs treated as first-class artifacts, and governance around the agent. That’s exactly what the skill catalog enables at organizational scale: one practitioner’s proven skill becomes the whole company’s standard. Across our engagements, we measure 3 to 5x per-task acceleration with AI-assisted Snowflake development. The catalog turns individual productivity into institutional capability, which is where durable value lives.
Intent-Driven Governance, Agent Identity, and Data Movement Policies
Three governance announcements stack here that matter greatly for our regulated clients. Intent-driven governance: state “protect all PII in this database” and the platform classifies, creates policies, and keeps enforcing them. Agent identity: masking and row policies can behave differently when an agent, not a human, is asking. Data movement policies: tagged data physically cannot leave your perimeter, with violations surfacing in Trust Center. Multi-party approvals mean no single admin can execute a sensitive operation alone.
Where this lands: a large county government carries criminal-justice data whose control obligations map almost one-to-one onto data movement policies. A national digital bank standing up fraud analytics needs the agent identity answer before AI touches member data. Every regulated client has been blocked on the same question: what exactly can the agent see, and can you prove it? The platform now has a native answer. The value is unblocked AI programs compliance previously vetoed.
AIM: AI-Powered Migrations and Teradata Virtualization
Snowflake consolidated migration tooling under AIM, with CoCo doing the conversion, and added Teradata virtualization: run Teradata SQL and BTEQ workloads on Snowflake as-is, convert later on your schedule.
Where this lands: a national mortgage lender is moving a SQL Server estate to Snowflake. A payments services organization serving credit unions just completed a Redshift wave with Teradata and Oracle queued behind it. AIM changes the economics of the remaining waves: AI-assisted conversion compresses the timeline, and virtualization removes the big-bang cutover risk that makes CFOs defer. The pattern across our migration work is consistent: the cost of waiting now exceeds the cost of moving.
Snowflake Data Stream: Managed Streaming, Kafka Compatible
A fully managed, Kafka wire-compatible streaming service built into the platform. Existing Kafka clients point at it and keep working. Topics instantiate directly into Snowflake tables with sub-second latency.
Where this lands: a connected-property security platform lives on device telemetry, which today means a separate streaming stack and a second governance perimeter. Data Stream collapses both into the platform the analytics already live on. The Kafka infrastructure cost disappears and event data becomes analyzable the second it lands. It’s in private preview, so plan for it rather than on it.
OpenFlow: Programmatic APIs, Private Connectivity, and Oracle GA
OpenFlow added a programmatic object model, a data connectivity proxy for private network paths, and a GA Oracle connector. That same payments services organization has an active Teradata and Oracle replication workstream that just moved from pilot risk to production footing. For the county government, where every byte must travel a compliant network path, the proxy addresses the objection that previously forced workarounds. Ingestion is the unglamorous 40% of every modernization budget, and this shrinks it.
The Performance Dividend
Same workloads, lower bill, faster experience. Adaptive Compute hits GA at roughly twice gen-1 warehouse speed. A new interactive compiler showed a 40x compile-time improvement, accelerating one large customer’s workload 3 to 4x. Unistore got an engine rework worth roughly 8x on latency and throughput. And Postgres data mirroring is now a switch: flip it, and an operational Postgres table continuously mirrors into Snowflake. A fintech we’re engaged with gets to delete a CDC pipeline project outright; a major electric utility and a leading cancer center get a compute line item their boards will notice. None of this is exciting. All of it is money.
What We’d Do Monday Morning
Three moves. If you have an unstructured-data backlog, scope an Agentic Search pilot while it’s still a differentiator. If a migration is stalled on risk, re-price it against AIM and virtualization; the math changed. And if you want to know what agentic development looks like when it’s real, study what Sai Kesanapalli and ADI Global have done with CoCo: production code, production governance, production results. Then come find us at a hands-on lab. We’ll show you how to get there.
The platform is no longer the constraint. Execution is. That’s where evolv lives. Reach out to evolv and let’s build something real.
Trent Foley is Chief Technology Officer at evolv Consulting. His career is built around a single standard: excellence, in the craft and in the client outcome. A well-rounded technologist spanning enterprise applications, cloud, and data, Trent brings more than 20 years of technology leadership, including a decade at ORIX building the platforms behind portfolio management operations and consulting leadership at Slalom. Across engagements in banking, healthcare, and financial services, his measure of success has stayed constant: business value the customer can count. Trent is one of the earliest Snowflake CoCo Champions and a frequent presenter on agentic development at Snowflake events across the country. He is based in Dallas-Fort Worth.
