Data 360
Data 360 at Dreamforce 2026: zero copy everywhere, and the context layer beneath agents
Data 360 did not get a Dreamforce release. It got pieces of four announcements: zero copy reaching Aurora, RDS, Glue and BigQuery through an Iceberg catalog, Informatica's MCP servers handing agents the golden record, a semantic layer under Tableau, and the session traces that Agent Optimizer will read. This is what a data team can use now, what is dated for the fall, and the inventory to do first.
Data 360 ships features monthly and does not really have a Dreamforce release. What it had this year was a presence in four other announcements: the AWS one, the Google one, the Trusted Enterprise AI Harness, and Winter ‘27. Read together, they say one thing. The platform is being built as the context layer under agents, and the marketing-CDP features are along for the ride.
The recap covers the event. This post pulls out every Data 360 item, sorts it by date, and ends with the inventory I would run this month.
Zero copy now reaches most of where your data is
The biggest practical change is the length of the zero-copy list. The AWS announcement extends Data 360 federation to AWS Glue-managed Apache Iceberg tables, S3-backed Iceberg tables, Amazon Aurora, Amazon RDS and SageMaker Lakehouse, with the existing Redshift and S3 support continuing. If your lake is Iceberg on S3 with a Glue catalog, or your operational data sits in Aurora or RDS, an agent can ground on it without a nightly copy and without the Redshift detour some teams built.
On the Google side, the BigQuery zero-copy expansion reaches general availability in Fall 2026, with Iceberg REST Catalog adoption, Private Connect support and more Salesforce and Google regions. The BigQuery post covers the current setup; the fall change is mostly about which regions and which network path.
Databricks federation and sharing has its own write-up from this morning, and Winter ‘27 added a Snowflake Zero-Copy V2 connector for data shares, which extends the Data 360 versus Snowflake picture without changing the argument.
The caution is the same for every entry on the list, and I made it in zero copy explained. Federation is a live query against a system you own. Its latency, cost and permission model become part of your agent’s behaviour. A slow Aurora query is a slow agent, and a query that runs on every agent turn is a bill.
The context layer: Intelligent Context, Tableau Semantics, and Trusted Context
The Trusted Enterprise AI Harness, announced on September 10, names six capabilities and puts Data 360 behind the first: Trusted Context, which brings customer data, metadata, semantics, knowledge, signals and memory together, grounded in governed enterprise data. The unified experience is early fiscal FY28. The pieces under it exist now, and I covered the Control Plane that sits over them.
Two of those pieces got stage time. Intelligent Context turns PDFs, contracts, diagrams and transcripts into retrievable context, so an agent can answer from a warranty document rather than a structured field. Tableau Semantics defines business terms once, such as what counts as revenue and which fiscal calendar applies, across the Customer 360 Semantic Data Model, and Google’s announcement adds that Gemini Enterprise can now query Tableau through MCP with row-level security and semantic models enforced per query. That is the agentic analytics direction with a governance seam attached.
The reason the semantic layer matters more than it sounds: if two agents compute revenue differently, users stop trusting both. Defining the metric once is the unglamorous fix, and it is a data team’s job before it is an agent team’s.
Informatica’s MCP servers hand an agent the golden record
Informatica, listed among Salesforce’s own core technologies in the Harness announcement, made five MCP servers generally available in May and expanded them at Dreamforce. Its own write-up names four: CDGC Metadata Search, for asset discovery, schemas, column profiles, glossary definitions and certification status; Data Provisioning, for governed access requests; Job Management, for running integration tasks; and Master Data Management, for retrieving master records and reference data.
An agent on Amazon Bedrock AgentCore reaches them through the AgentCore Gateway with service-to-service OAuth, and the AWS announcement adds Amazon Quick as a client. The practical consequence for a Salesforce data team is that an agent outside Salesforce can now ask which customer record is the golden one and get Informatica’s answer, which is the same answer identity resolution in Data 360 is supposed to reach from the CRM side. If you run both, decide which one is authoritative for which domain before two agents disagree.
The Winter ‘27 items you can use now
Winter ‘27 landed on the first instances on September 4 and reaches the rest on October 2 and 9. The Data 360 items in it are already documented in the release notes, and I have written each up.
Problem Records now land in their own data lake objects, so the rows ingestion quarantined have a place you can query. SQL from Apex gets a first-class API in the sfsqlquery namespace, replacing hand-rolled ConnectApi loops. Data Detect in Shield can scan data lake objects, so sensitive-text discovery reaches the ingested data an agent grounds on. And ingesting through the CRM connector does not consume credits, which is worth re-checking against your Digital Wallet if your pipelines predate that change.
Session traces live here too
The Agentforce keynote’s Agent Optimizer demo traced a spike in escalated refunds to a router gap. The traces it read are the Session Tracing Data Model, which is stored in Data 360. Optimizer goes GA in October, and it has nothing to analyse in an org where session tracing is off or Data 360 is not provisioned. I covered what to switch on before then.
The retention side is yours as well. Once traces and transcripts are synced into Data 360, their lifecycle follows the data-space retention policies you configure, and a right-to-be-forgotten request has to reach that copy too. The transcript storage map has the three stores and the three export paths.
What is dated
Available now: the AWS zero-copy sources, Informatica’s MCP servers, the Winter ‘27 items on upgraded instances, Intelligent Context and Tableau Semantics as they shipped through the year. Fall 2026: BigQuery zero copy GA with the Iceberg REST Catalog and Private Connect. October: Agent Optimizer. Early FY28: the unified Harness experience.
The inventory I would run this month
List every source you currently copy into Data 360 and check it against the new zero-copy list. Anything from Aurora, RDS, Glue or S3 Iceberg is a candidate for federation. Measure the query cost of the federated version before you switch, because you are trading storage credits for query credits.
Write down the three business metrics your agents will be asked about most, revenue, active customer, open case, and define each once in the semantic layer. That work outlives any tool.
Decide which system is authoritative for master data per domain if you run both Informatica and Data 360 identity resolution, and write it into the agent’s grounding configuration.
Turn on session tracing and set the data-space retention for it, so October’s Optimizer has September’s traces and your erasure process knows where they are.
Understanding the basics
What did Data 360 announce at Dreamforce 2026?
Zero copy extended to AWS Glue and S3 Iceberg tables, Aurora, RDS and SageMaker Lakehouse, with BigQuery zero copy reaching GA in Fall 2026 through an Iceberg REST Catalog and Private Connect. Data 360 is also positioned as the Trusted Context capability in Salesforce’s Enterprise AI Harness, and Informatica’s MCP servers are reachable from Amazon Bedrock AgentCore and Amazon Quick.
Which AWS data sources does Data 360 zero copy support now?
AWS Glue-managed Apache Iceberg tables, S3-backed Iceberg tables, Amazon Aurora, Amazon RDS and SageMaker Lakehouse, in addition to the existing Amazon Redshift and S3 support. Federation runs live against the source, so its performance and permissions become part of the agent’s behaviour.
What is Tableau Semantics?
A semantic layer that defines business metrics such as revenue once across the Customer 360 Semantic Data Model, so agents and dashboards compute them the same way. Since Dreamforce 2026, Gemini Enterprise can query Tableau through MCP with row-level security and those semantic models enforced per query.
Do I need Data 360 for Agent Optimizer?
In practice, yes. Agent Optimizer reads the Session Tracing Data Model, which is stored in Data 360, and the Session Trace API alone only reaches back 72 hours. Session tracing and the Trust Layer audit collection must be enabled before the traces exist.
The platform is being built for agents, and priced by the query
Every Data 360 item at Dreamforce reduces the distance between an agent and a system you already own. Federation instead of copies, a semantic layer instead of per-dashboard definitions, an MCP call instead of an integration. That is the right direction, and it moves cost from storage to queries and moves governance onto the sources.
Run the inventory before you enable any of it. The teams that get the most from zero copy are the ones who know what a federated query costs before the first agent makes ten thousand of them.
If you want the zero-copy candidates, the semantic definitions and the trace retention mapped against your own estate before the fall releases land, talk to us, or start with the org health scorecard.
