Snowflake x Google Cloud AI Agent Lab Comes to New York

Snowflake and Google Cloud are bringing a hands-on AI agent lab to New York on October 13, 2026, with an agenda that follows enterprise data from an agent workflow into an analytics dashboard. The Snowflake x Google Cloud AI Agent Lab is scheduled for 2–5 PM EDT at Google's Chelsea Market office. Major League Hacking hosts the official Luma listing, which currently shows registration closed.

The intended audience includes enterprise software engineers, data architects and technical leaders. The advertised exercise combines Snowflake Cortex, Gemini 3.1 Pro and Looker, giving practitioners a common example through which to examine model behavior, data access and output. The broader significance lies in bringing those engineering responsibilities into the same conversation before a team commits to a deployment.

This is upcoming-event coverage: the planned lab has not yet taken place. Individual keynote speakers and the exact street address are not publicly confirmed in the reviewed listings. Approved guests should rely on the host's access instructions, and readers considering attendance should check the current registration status rather than assume a place is available.

An afternoon organized around a working example

The official Snowflake event page lists October 13 at 18:00–21:00 UTC, equivalent to 2–5 PM EDT. Its agenda puts check-in at 1:30 PM, a keynote at 2 PM, the technical lab at 2:30 PM, and demonstrations and networking at 4 PM. The session closes at 5 PM, so the numeric schedule establishes an afternoon gathering despite an introductory reference to an evening.

Organizers plan a 90-minute exercise involving a Cortex Agent, enterprise datasets, Google's Gemini 3.1 Pro and a Looker dashboard. Attendees are asked to bring a laptop and charger. The sequence creates room to inspect a build while the people responsible for different parts of it can still compare what they think the system is doing.

That matters for an engineering manager deciding what an experiment should establish. A useful exercise can show where data definitions, tool choices and output checks meet, making it easier to identify the next question a team needs to answer. The organizer's ambition to address production-oriented agents should be read as the purpose of the session; any decision to deploy would still depend on the team's own requirements and testing.

The practical appeal is an opportunity to interrogate a concrete example. An engineer can ask how a request reaches the relevant data, while an architect can examine whether the interpretation survives contact with business definitions. Those exchanges have more substance when everyone can refer to the same workflow.

The data layer carries decisions of its own

Snowflake's Cortex Agents documentation describes a platform that can coordinate tools across structured and unstructured data within a governed environment. It includes monitoring and evaluation capabilities, while cautioning that model answers and citations require review. Those details provide useful context for the lab without establishing how its particular exercise will be configured.

For an operator, a natural-language request can hide a considerable amount of business meaning. A question about customers might depend on which records count, which period applies and which user has permission to see the result. The practical discussion begins with those choices, because an elegant answer cannot explain a business definition nobody agreed on.

Google Cloud's Gemini 3.1 Pro documentation labels the model Public Preview. That status belongs in an adoption discussion alongside the advertised integration. Teams evaluating the approach should examine applicable availability, configuration and operating terms before carrying a workshop example into their own environment.

The planned use of Looker adds a second way to examine what an agent produces. A dashboard can organize outputs and operational measures for inspection; choosing which measures deserve attention remains a human decision. Accuracy on a business question, behavior on incomplete inputs and the cost of getting an answer are useful subjects to raise, rather than features the event has promised to demonstrate in full.

Taken together, the agenda gives participants a way to discuss the connection between a model, the tools around it and the data those tools can reach. The commercial question is how much additional work a team would need before that connection fits an actual operating process.

A practitioner room with several responsibilities represented

The audience described by the organizers is unusually relevant to that question: software and AI engineers, data engineers and architects, engineering managers and technical leads. These roles approach an agent from different obligations. Bringing them around one example can expose a handoff that looks obvious within a single discipline and becomes less comfortable when somebody else has to maintain it.

The Snowflake New York user-group page lists Kevin Jong, described as a Principal GTM Engineer, and Ajay Phatak, listed as a Lead Analytics Consultant at Braze, among its organizers. They also appear in the official event page's organizer section. Neither listing establishes an individual keynote role, and Ajay Phatak's company affiliation should not be read as a Braze sponsorship claim.

The Luma page identifies Major League Hacking as host and names Snowflake and Google Cloud in the program. Those are the relationships the public record supports. A named speaker lineup, moderator and participant roster would add specificity, but the available evidence does not support filling those gaps with familiar executives or likely attendees.

Structured networking after the lab could be useful precisely because the working session supplies shared material. Participants can compare how their teams define success, handle permission boundaries or explain uncertain output to a business user. The advertised refreshments make space for that exchange; the value will depend on the questions and experience people actually bring.

What to establish before carrying the example home

The official Luma registration page currently says registration is closed. Earlier language about reviewed applications and limited capacity remains part of the description, but it does not establish current access. Approved guests should confirm arrival instructions and any account or setup requirements directly with the host before the session.

Google's Chelsea Market office is the publicly named venue, while exact address details are gated. That distinction matters for anyone traveling to the event. The coverage identifies the venue and city without substituting an unverified street address for instructions that the organizer controls.

For teams following from outside the room, the agenda offers a useful set of questions to carry into their own evaluation. Who owns the meaning of the data? Which tools can the agent use? How can someone inspect an answer, and what should happen when the available evidence is incomplete? Each answer connects the technical experiment to a colleague who will have to live with its behavior.

At the October 13 lab, Snowflake and Google Cloud plan to put those connections within reach of a working example. The consequential conversation continues when an engineer returns to the team and asks the data owner, the application maintainer and the business user to examine the same output together.