--- title: "Where does a context layer fit with Agentforce and Data Cloud?" url: "https://atlan.com/know/ai-agent/context-layer/where-does-a-context-layer-fit-with-agentforce-and-data-cloud/" description: "A context layer sits under Agentforce and beside Data Cloud (Data 360). See the stack, what each layer owns, and when CRM-only context stops being enough." author: "Emily Winks" author_role: "Data Governance Expert" published: "2026-10-21" updated: "2026-10-21" --- > Atlan is hosting Context Conference, bringing together the leaders and builders at the frontier of giving AI the context it needs to understand their business. It runs online on October 28, 2026, from 11:00 AM to 2:00 PM ET. Atlan co-founder Prukalpa Sankar opens and closes the day. Leaders from AstraZeneca, BNY and Verizon share why they invest in context and what they get from it. Registrants get early access to The AI Context Gap, a new study from MIT Technology Review Insights. Register: https://atlan.com/context-conference/ ## Where does a context layer fit with Agentforce and Data Cloud? A context layer fits underneath Agentforce and beside Data Cloud, which Salesforce now calls Data 360. Data Cloud gives Agentforce customer records, unified profiles, and retrieval inside Salesforce, while the context layer holds certified definitions, lineage, owners, and access policies for data in every system. Agentforce plans the work, Data Cloud grounds it in customer data, and the context layer tells the agent what that data means and whether to trust it. What each layer owns: - **Context layer.** Certified definitions, lineage, owners, quality signals, and access policy across all sources - **Data 360.** Unified customer profiles, retrieval, and semantic models - **Atlas Reasoning Engine.** The plan for each request - **Agents.** Agentforce agents plus agents built on other platforms ## Key takeaways - Data 360 must be enabled for all Agentforce use, per Salesforce's Trailhead module. - Data 360 sees the columns it federates, while definitions written in other tools stay in those tools. - A context layer describes data rather than storing customer profiles. - Mapping against certified definitions beats certifying them after an agent gives a customer a wrong answer. --- A [context layer](https://atlan.com/know/what-is-the-enterprise-context-layer/) fits underneath [Agentforce](https://atlan.com/know/ai-agent/ai-agent-applications/what-is-salesforce-agentforce/) and beside Data Cloud, which Salesforce now calls Data 360. Data Cloud gives Agentforce customer records, unified profiles, and retrieval inside Salesforce, while the context layer holds certified definitions, lineage, owners, and [access policies](https://atlan.com/know/ai-agent-access-control/) for data in every system. Agentforce plans the work, Data Cloud grounds it in customer data, and the context layer tells the agent what that data means and whether to trust it. --- ## What do Agentforce and Data Cloud each do? Agentforce is the agent platform. Salesforce made it generally available on [October 29, 2024](https://investor.salesforce.com/news/news-details/2024/Salesforces-Agentforce-Is-Here-Trusted-Autonomous-AI-Agents-to-Scale-Your-Workforce/default.aspx) and called its Atlas Reasoning Engine "the brain inside Agentforce." Data Cloud is the data platform under it. At Dreamforce on [October 13, 2025](https://www.salesforce.com/news/press-releases/2025/10/13/agentic-enterprise-announcement/), Salesforce launched [Agentforce 360](https://atlan.com/know/ai-agent/ai-agent-applications/what-is-agentforce-360/) with four parts: the Agentforce 360 Platform, Data 360, Customer 360 Apps, and Slack. Data Cloud was rebranded to Data 360 as of October 14, 2025, per Salesforce's [Trailhead module](https://trailhead.salesforce.com/content/learn/modules/data-cloud-powered-agentforce/explore-data-cloud-and-agentforce), and the launch release says Data 360 gives "every agent business context and personalization." The stack reads top to bottom like this: | Layer | Salesforce component | What it does for an agent | |---|---|---| | Agent | Agentforce agents, built from [subagents (formerly topics)](https://www.salesforce.com/agentforce/guide/), instructions, actions, and [guardrails](https://atlan.com/know/ai-agent-risks-guardrails/) ([Salesforce Architects](https://architect.salesforce.com/docs/architect/fundamentals/guide/enterprise-agentic-architecture.html)) | Takes the request and acts on it | | Reasoning | Atlas Reasoning Engine | Plans the task, picks actions, and checks the result | | Applications | Customer 360 apps for Sales, Service, Marketing, and Commerce, plus Slack | Hold the records and processes the agent reads and changes | | Data | Data 360 (formerly Data Cloud) | Unifies customer data, resolves identities, and retrieves grounding content | | Integration | MuleSoft, Flow, [MCP](https://atlan.com/know/what-is-model-context-protocol/), and [A2A](https://atlan.com/know/google-a2a-protocol/) ([Agentforce platform](https://www.salesforce.com/platform/agentforce-platform/)) | Connects agents to outside apps, APIs, and other agents | The data layer is not optional. Salesforce's [Trailhead module](https://trailhead.salesforce.com/content/learn/modules/data-cloud-powered-agentforce/explore-data-cloud-and-agentforce) states that Data 360 must be enabled for all Agentforce use, and that the Agentforce Data Library and the Einstein Trust Layer do not work without it. The Data 360 features that this Trailhead module describes all consume Data 360 credits, so the data tier is part of the agent bill. --- ## What context does Data Cloud already give an agent? Inside Salesforce, a lot. Data 360 builds unified customer profiles through identity resolution and cleans records with data transforms, per Salesforce's [Trailhead unit](https://trailhead.salesforce.com/content/learn/modules/data-cloud-powered-agentforce/explore-data-cloud-and-agentforce). Its [Zero Copy feature](https://atlan.com/know/ai-agent/ai-agent-applications/reverse-etl-for-salesforce-ai-agents/) reads data held elsewhere without duplicating it; Trailhead's example is loyalty data kept in Databricks. The [Dreamforce release](https://www.salesforce.com/news/press-releases/2025/10/13/agentic-enterprise-announcement/) names Intelligent Context as the Data 360 feature that lets agents read [unstructured content](https://atlan.com/know/data-for-ai/unstructured-data-for-ai/) such as PDFs and diagrams. For documents, emails, and tickets, Data 360 also has a native [vector database](https://atlan.com/know/what-is-a-vector-database/), hybrid search that mixes semantic and exact-match keyword search, and no-code retrievers that admins configure with filters and ranking ([Salesforce, November 20, 2024](https://www.salesforce.com/news/stories/how-data-cloud-powers-agentforce/)). Salesforce's engineering team says Salesforce changes reach Snowflake through Zero Copy in about 60 seconds ([Salesforce Engineering, May 5, 2025](https://engineering.salesforce.com/scaling-data-cloud-for-agentforce-and-ai-rethinking-metadata-activation-and-governance-at-hyperscale/)). The same post covers [governance](https://atlan.com/know/ai-agent/ai-agent-applications/salesforce-data-governance/): attribute-based access control that works alongside [role-based access](https://atlan.com/know/ai-agent/context-layer/context-layer-role-based-access-control/), with LLMs curating tags on entities, columns, rows, and content. For business meaning, Salesforce added [Tableau Semantics](https://www.salesforce.com/analytics/tableau-semantics/), a [semantic layer](https://atlan.com/know/semantic-layer/) built into Data 360 that keeps metrics consistent across clouds through the [Customer 360 Semantic Data Model](https://www.salesforce.com/news/press-releases/2025/10/13/agentic-enterprise-announcement/). Salesforce states it is not available without Data 360. Salesforce calls the work of assembling all this for an agent [context engineering](https://help.salesforce.com/s/articleView?id=ai.agent_context_engineering.htm&language=en_US&type=5). Its [guide to context engineering](https://www.salesforce.com/agentforce/guide/) names three levers: subagents that define each specialized agent and its expertise, instructions that guide conversation flow, tone, and intent, and actions and rules that enforce mandatory business logic. Agentforce Script expresses all three in one file, and hybrid reasoning blends [LLM reasoning](https://atlan.com/know/llm-reasoning/) with deterministic, rules-based execution. If every fact an agent needs lives in your Salesforce org, Data 360 covers most of it. What context does to AI accuracy Inside Atlan AI Labs shows what happened to AI accuracy when agents were given business context. Get the 5x Accuracy Ebook --- ## Where does Agentforce's context stop for data outside Salesforce? It stops where Data 360's model of your business stops. Zero Copy can read a table in a warehouse. The meaning of that table usually lives somewhere else: the revenue metric defined in a transformation project, the churn score built by the data science team, the "active account" rule a finance analyst wrote into a BI tool. Data 360 sees the columns it federates, while definitions written in other tools stay in those tools. Three gaps follow. Definitions drift. Tableau Semantics governs metrics modeled in Data 360. When the same metric is also [defined in a warehouse and a dashboard](https://atlan.com/know/ai-agent/context-layer-for-data-analytics-teams/), someone has to decide which version is certified, or the agent answers with whichever one grounded it. Lineage breaks at the boundary. When an upstream job fails and a loyalty field goes stale, the agent still retrieves the field. [Tracing that wrong answer back to a source system](https://atlan.com/know/ai-agent/data-for-ai/data-lineage-for-ai/) takes lineage that spans systems Salesforce does not run. Other agents keep their own context. Salesforce's [architecture guide](https://architect.salesforce.com/docs/architect/fundamentals/guide/enterprise-agentic-architecture.html) plans for multi-vendor setups, where a Salesforce [supervisor agent coordinates agents](https://atlan.com/know/multi-agent-coordination-patterns/) built on Google Vertex or [LangGraph](https://atlan.com/know/ai-agent/ai-agent-memory/what-is-langgraph/) through A2A. Each of those agents needs [the same definition of a customer](https://atlan.com/know/context-management-multi-agent-systems/). [Open protocols](https://atlan.com/know/agent-interoperability-protocols/) help with the plumbing. [MCP](https://www.anthropic.com/news/model-context-protocol), released November 25, 2024, standardizes how an agent connects to a data source, and [A2A](https://www.linuxfoundation.org/press/linux-foundation-launches-the-agent2agent-protocol-project-to-enable-secure-intelligent-communication-between-ai-agents), hosted by the Linux Foundation since June 23, 2025, standardizes how agents hand work to each other. Neither decides whether the definition that comes back is the approved one. On [September 23, 2025](https://www.snowflake.com/en/news/press-releases/snowflake-salesforce-dbt-labs-and-more-revolutionize-data-readiness-for-ai-with-open-semantic-interchange-initiative/), Snowflake, Salesforce, dbt Labs, and RelationalAI, with support from BlackRock, launched the Open Semantic Interchange to reconcile semantic definitions across platforms. That project is now [Apache Ossie (formerly the Open Semantic Interchange)](https://ossie.apache.org/updates/ossie-enters-apache-incubator/), an Apache incubator project since July 10, 2026. For the governance view of this boundary, see [Agentforce data governance coverage and gaps](https://atlan.com/know/ai-agent/ai-agent-applications/agentforce-data-governance/). --- ## Where does a context layer fit in the Agentforce and Data Cloud stack? At the data tier, next to Data 360, and below every agent platform you run. Read the stack from the bottom up: | Position | Layer | What it owns | Who reads it | |---|---|---|---| | 1 | Source systems: your Salesforce org, warehouses, lakehouses, pipelines, BI tools, and apps | The raw data | The layers above | | 2 | Context layer | Certified definitions, lineage, owners, [quality signals](https://atlan.com/know/data-for-ai/data-quality-for-ai-agent/), and access policy across all sources | Agentforce and every other agent; Data 360 teams map from it during setup | | 3 | Data 360 | Unified customer profiles, retrieval, and semantic models | Agentforce | | 4 | Atlas Reasoning Engine | The plan for each request | Agentforce agents | | 5 | Agents | Agentforce agents plus agents built on other platforms | Your users | Context reaches Agentforce by two paths. At build time, Data 360 setup maps connected data to data models ([Trailhead](https://trailhead.salesforce.com/content/learn/modules/data-cloud-powered-agentforce/explore-data-cloud-and-agentforce)). Use the [certified terms and metrics](https://atlan.com/know/ai-agent/how-to-manage-kpi-definitions-and-a-business-glossary-in-a-data-catalog/) from the context layer when you do that mapping, so the definitions Agentforce grounds on match the rest of the company. At run time, agents query the context layer directly. Salesforce's [architecture guide](https://architect.salesforce.com/docs/architect/fundamentals/guide/enterprise-agentic-architecture.html) says agents reach external functionality through the Agentforce MCP Client, and MuleSoft can wrap an API that does not speak MCP. A [workable build order](https://atlan.com/know/how-to-implement-enterprise-context-layer-for-ai/): 1. List the questions your agents must answer and the systems each question touches. 2. Certify the terms those questions depend on, such as customer, active account, and ARR, and give each one an owner. 3. During Data 360 setup, map the certified terms and metrics by hand to Data 360 data models. 4. Expose the context layer through an [MCP server](https://atlan.com/know/ai-agent/how-to-build-mcp-servers-for-enterprise-data/) and register it with the Agentforce MCP Client. 5. Test with real questions, and trace each wrong answer to a definition or a lineage break. For the trade-offs of keeping context inside the platform versus outside it, see [Agentforce versus an independent context layer](https://atlan.com/know/ai-agent/ai-agent-applications/agentforce-vs-independent-context-layer/). Find the gaps in your agent context The Context Gap Calculator estimates where your agents lack the definitions, lineage, and policy they need. Calculate Your Context Gap --- ## When do you need a context layer next to Data Cloud? Not always. If your agents answer questions about Salesforce records, for Salesforce users, inside one agent platform, Data 360 and Tableau Semantics may be enough. The signals below say the boundary has started to cost you. | Signal | What you see | What sits outside Data 360's scope | |---|---|---| | Key data lives outside Salesforce | Agents quote CRM numbers that finance does not recognize | Business logic written in other tools stays in those tools | | The same metric has several owners | The agent and the dashboard disagree on revenue | [Metric definitions](https://atlan.com/know/ai-agent/semantic-layer/semantic-layer-vs-metrics-layer/) held in other BI tools and warehouses | | You run agents on more than one platform | Two agents give two answers to one question | Context built on other agent platforms | | Auditors ask where an answer came from | Nobody can trace a field back to its source pipeline | [Lineage](https://atlan.com/know/ai-readiness/ai-ready-data-lineage/) in systems Salesforce does not run | | Wrong answers trace to stale source data | Fixes happen one agent at a time | [Quality checks](https://atlan.com/know/ai-agent/ai-agent-evals/context-evals-vs-data-quality-testing/) that run in upstream source systems | Timing matters. Data 360 implementation already means mapping each source to data model objects and writing identity resolution rules, per [Trailhead](https://trailhead.salesforce.com/content/learn/modules/data-cloud-powered-agentforce/explore-data-cloud-and-agentforce). Mapping against certified definitions beats certifying them after an agent gives a customer a wrong answer. For how this layer differs from the catalog many teams already run, see [data catalog versus context layer](https://atlan.com/know/data-catalog-vs-context-layer/). --- ## How does Atlan fit with Agentforce and Data Cloud? Atlan [sits underneath Agentforce as an independent context layer](https://atlan.com/know/ai-agent/ai-agent-applications/agentforce-vs-independent-context-layer/), not as a competing agent platform. Atlan's [Salesforce connector](https://docs.atlan.com/apps/connectors/crm/salesforce) catalogs Salesforce objects, fields, and metadata. Atlan connects to [400+ source systems](https://atlan.com/mcp/), so those Salesforce objects sit in one graph with the warehouse tables, pipelines, and BI metrics that define what the data means. Agents read that context through the [Atlan MCP server](https://atlan.com/mcp/), which works with Claude, ChatGPT, Gemini, Microsoft Copilot Studio, and other clients. MCP clients can use the Atlan MCP server [alongside Salesforce's own MCP servers](https://atlan.com/know/ai-agent/mcp-server-for-salesforce/). Workday serves context to AI through the [Atlan MCP server](https://atlan.com/know/what-is-atlan-mcp/). Joe DosSantos, VP of Enterprise Data and Analytics at Workday, says Atlan captures Workday's shared business language and serves it to AI through its MCP server: "As part of Atlan's AI labs, we're co-building the semantic layer that AI needs" ([atlan.com/mcp](https://atlan.com/mcp/)). Atlan was also a [launch partner](https://www.snowflake.com/en/news/press-releases/snowflake-salesforce-dbt-labs-and-more-revolutionize-data-readiness-for-ai-with-open-semantic-interchange-initiative/) of [Apache Ossie (formerly the Open Semantic Interchange)](https://ossie.apache.org/updates/ossie-enters-apache-incubator/), a vendor-neutral specification for semantic definitions. See a context layer working live Watch live demos of Atlan serving definitions and lineage to AI agents. Watch Live Demos Book a Demo --- ## FAQs about where a context layer fits with Agentforce and Data Cloud ### Is Data Cloud the same as Data 360? Yes. Per Salesforce's Trailhead module, Data Cloud was rebranded to Data 360 as of October 14, 2025, the week Salesforce launched Agentforce 360 at Dreamforce. The product, its role as the data layer under Agentforce, and its features carry over under the new name. Documentation and older architecture diagrams still use both names for the same platform, so treat them as one product when you plan. ### Does Agentforce require Data Cloud? Yes. Salesforce's Trailhead module on Data 360 and Agentforce states that Data 360 must be enabled for all Agentforce use. The Agentforce Data Library and the Einstein Trust Layer do not work without it. Enabling it is a lighter step than implementing it, which means connecting sources, mapping them to data models, and setting identity resolution rules for unified profiles. ### Is Tableau Semantics a context layer? Tableau Semantics is a semantic layer inside Data 360. It keeps metric definitions consistent across Salesforce clouds and grounds Agentforce in business terms. A context layer covers more ground: definitions, lineage, owners, quality signals, and access policy across every system, including those Salesforce does not run. The two work together when the team builds Tableau Semantics models by hand from the certified definitions. ### Can a context layer replace Data 360? No. They do different jobs. Data 360 stores and unifies customer data, resolves identities, and runs retrieval for Agentforce, and Agentforce does not run without it. A context layer describes data rather than storing customer profiles. It records what each field means, where it came from, who owns it, and who may see it, and teams configure Data 360 and other agents from those definitions. ### How does Agentforce reach a context layer outside Salesforce? Through open protocols. Salesforce's architecture guide says Agentforce agents call external tools through the Agentforce MCP Client, and MuleSoft can wrap APIs that do not support MCP. A context layer that exposes an MCP server can answer Agentforce at run time. For agent-to-agent handoffs, Agentforce supports A2A, the open protocol now hosted by the Linux Foundation. --- ## Sources 1. [Salesforce's Agentforce is here: trusted, autonomous AI agents to scale your workforce, Salesforce Investor Relations (October 29, 2024)](https://investor.salesforce.com/news/news-details/2024/Salesforces-Agentforce-Is-Here-Trusted-Autonomous-AI-Agents-to-Scale-Your-Workforce/default.aspx) 2. [Salesforce announces the Agentic Enterprise, Salesforce (October 13, 2025)](https://www.salesforce.com/news/press-releases/2025/10/13/agentic-enterprise-announcement/) 3. [Understanding Data Cloud's role in Agentforce, Salesforce Trailhead](https://trailhead.salesforce.com/content/learn/modules/data-cloud-powered-agentforce/explore-data-cloud-and-agentforce) 4. [Agentforce guide to context engineering, Salesforce](https://www.salesforce.com/agentforce/guide/) 5. [Enterprise agentic architecture, Salesforce Architects](https://architect.salesforce.com/docs/architect/fundamentals/guide/enterprise-agentic-architecture.html) 6. [Agentforce 360 Platform, Salesforce](https://www.salesforce.com/platform/agentforce-platform/) 7. [How Data Cloud fuels Agentforce and the next era of AI, Salesforce (November 20, 2024)](https://www.salesforce.com/news/stories/how-data-cloud-powers-agentforce/) 8. [Scaling Data Cloud for Agentforce and AI, Salesforce Engineering (May 5, 2025)](https://engineering.salesforce.com/scaling-data-cloud-for-agentforce-and-ai-rethinking-metadata-activation-and-governance-at-hyperscale/) 9. [Tableau Semantics, Salesforce](https://www.salesforce.com/analytics/tableau-semantics/) 10. [Agentforce context engineering, Salesforce Help](https://help.salesforce.com/s/articleView?id=ai.agent_context_engineering.htm&language=en_US&type=5) 11. [Introducing the Model Context Protocol, Anthropic](https://www.anthropic.com/news/model-context-protocol) 12. [Linux Foundation launches the Agent2Agent Protocol project, Linux Foundation](https://www.linuxfoundation.org/press/linux-foundation-launches-the-agent2agent-protocol-project-to-enable-secure-intelligent-communication-between-ai-agents) 13. [Snowflake, Salesforce, dbt Labs and more launch the Open Semantic Interchange initiative, Snowflake (September 23, 2025)](https://www.snowflake.com/en/news/press-releases/snowflake-salesforce-dbt-labs-and-more-revolutionize-data-readiness-for-ai-with-open-semantic-interchange-initiative/) 14. [Apache Ossie (Incubating): the new name for Open Semantic Interchange, Apache Software Foundation](https://ossie.apache.org/updates/ossie-enters-apache-incubator/)