If AI “works” but isn’t reliable, explainable, or governable, it doesn’t scale. Building on that premise, Salesforce has unveiled a suite of innovations in its Salesforce Platform aimed at addressing the Achilles’ heel of AI projects in large organizations: fragmented data, inconsistent semantics, and weak governance. The company discusses a trust foundation designed to give each AI agent the same context, security, and control, enabling a transition from cautious testing to transformative AI adoption across all workflows.
The urgency is real. A RAND study, cited by Salesforce, indicates that over 80% of AI projects fail to deliver value, often due to Poor data quality,
- Context and accuracy: outputs anchored in unified business data and corporate knowledge.
- Trust, security, and compliance built-in: visibility, controls, and compliance packed into every flow.
- Openness and unification: agents, data, and semantics connected across ecosystems, with no vendor lock-in.
“In today’s era, managing data — not just storing it — is the new foundation for AI readiness, from cloud to core and edge,” summarizes Rob Lee, CTO of the company. “Success depends on having secure data everywhere and accessible from any location, with a unified and consistent experience, in real-time and at scale.”
This framework is called “Agentic Enterprise” at Salesforce: a model where humans and AI agents collaborate seamlessly across flows, decisions, and interactions.
Innovations: an Index of Context, clean rooms without copying, semantics in Tableau, and a “fabric” for governing agents
1) Data Cloud Context Indexing (GA in Winter 2025)
This is an indexing pipeline within Data Cloud that helps agents interpret unstructured content — contracts, diagrams, tables — through a business lens. The promise is to extract precise details from different data sets. For example: a field technician uploads a schematic and the agent guides them step-by-step through a decision tree for resolving a breakdown, turning hours into minutes.
2) Data Cloud Clean Rooms (beta now; GA November 2025)
Data Cloud’s clean rooms enable organizations to share, collaborate, and analyze data without moving or exposing raw data. This approach relies on zero-copy connectivity, eliminates duplicates — reducing security/compliance risks and storage costs — and integrates natively (with privacy enhancements) with AWS Clean Rooms. Example: a group of global banks comparing transaction patterns and detecting fraud rings within hours instead of weeks, without sharing sensitive records.
3) Tableau Semantics (available now; rollout of integrations through FY26)
This is a semantic layer with AI, tightly integrated with Data Cloud, to translate raw data into business language. Tableau offers a ready-to-use Customer 360 Semantic Data Model (SDM), which unifies data and metadata across multicloud environments, simplifies modeling, enforces consistent metric governance, and provides the context needed for AI and BI to generate trustworthy insights. Salesforce is also co-developing the industry’s first open semantic exchange and partnering with Databricks, dbt Labs, and Snowflake to standardize semantics across platforms. Typical use case: teams with different definitions of “ACV” can now align their logic so that every agent and dashboard uses the same vocabulary, ensuring coherent and explainable insights.
4) MuleSoft Agent Fabric (governance already available; Agent Registry, Broker, and Visualizer GA October 2025)
The proliferation of AI agents across teams, platforms, and providers creates a new kind of fragmentation: “agent sprawl” — disconnected flows, redundant automations, compliance blind spots. Agent Fabric offers a single registration, orchestration, and governance hub for all agents—regardless of origin—to coordinate workflows, avoid duplication, close compliance gaps, and trace decisions. Governance is already available, with Registry, Broker, and Visualizer platforms launching in October 2025.
5) Security and compliance embedded in AI
New AI-driven security capabilities will be embedded throughout the platform, with integrations with CrowdStrike and Okta, to advance threat detection and automate compliance management within the operational fabric of the agent-based enterprise.
6) Enterprise metadata with Informatica (pending closure, subject to approvals; expected Q1 FY27)
The planned acquisition of Informatica — cataloging, integration, governance, quality, privacy, and MDM — will add its metadata intelligence to Salesforce’s platform to create a unified data architecture for agent-based AI. The goal: ensure agents operate safely, responsibly, and at scale company-wide.
A common language for the agentic enterprise
The concept of “open and industry-standard semantics” is more than just a slogan. Salesforce aims to prevent each agent from “speaking” a different business dialect. By normalizing definitions — for example, what “active customer” or “churn” or “ACV” means — and governing metrics with Tableau Semantics, the ambiguity that undermines dashboards and AI responses is reduced. The openness (semantic exchange) and alliances (with Databricks, dbt, Snowflake) are designed to ensure semantics follow data across platforms.
“AI should be rooted in deep business context to unlock true multipliers: intelligent automation and actionable insights,” said Rahul Auradkar, EVP & GM of Unified Data Services. “That’s why we’re building a unified foundation to harmonize all data — structured and unstructured — to surface those insights and fuel automation, with the governance and security necessary to operate with confidence.”
Customers: From Roads to Healthcare
AAA Washington has already implemented Salesforce for sales and service. Using Data Cloud and MuleSoft as foundations, their goal is a 360° view of members, breaking silos for faster, more accurate, and relevant responses. “Our partnership with Salesforce helps us transform the roadside experience, insurance, and travel, following our mantra of ‘useful technology, human touch,’” said Jim Ryan, CIO.
In healthcare, UChicago Medicine emphasizes trust: “AI must be built on trust. With a unified foundation for data, each patient interaction is reliable, precise, and meaningful, and staff can spend more time on human care that defines us,” noted Andrew Chang, CMO.
Roadmap (Availability)
- Context Indexing (Data Cloud): Winter 2025 (GA).
- Data Cloud Clean Rooms: beta now; November 2025 (GA).
- MuleSoft Agent Fabric: governance available; Agent Registry, Broker, and Visualizer GA October 2025.
- Tableau Semantics: available; integrations with partners through FY26.
- Informatica Integration: post-completion (expected Q1 FY27), subject to approvals.
Why It Matters: From Fragmented Trials to “Factory-Driven” Governance
The announcement does not present a “miracle model” but rather the pipeline — data, metadata, semantics, governance, and control — that AI needs to operate effectively in regulated and
In a market increasingly filled with copilots and scattered agents, managing “agent sprawl” —the fabric of MuleSoft— and enforcing consistent semantics —Tableau Semantics— could be as crucial as the next language model.
Practical keys for data teams, security, and business leaders
- CIO / CDO: evaluate Context Indexing for critical unstructured content (contracts, blueprints, SOPs); define corporate semantics with Tableau SDM; activate clean rooms for copyless collaboration with partners or consortia.
- CISO / Compliance: integrate CrowdStrike/Okta into agent workflows and automation; document controls and decision traceability in AI (who did what, with which data).
- Architecture / Integration: register all agents in Agent Fabric; orchestrate actions and permissions centrally; reduce parallel or conflicting automations.
- Finance / Operations: unify metrics semantics (like ACV, churn, margin) to ensure all agents and dashboards “speak” the same language.
Conclusion
Salesforce advocates for a core principle in enterprise AI: trust is a design, not an afterthought. By proposing context (Data Cloud + Context Indexing), secure collaboration without copying (Clean Rooms), unified semantics (Tableau Semantics), agent governance (MuleSoft Agent Fabric), and embedded security, the company aims to turn the “promising AI” into “trustworthy AI”—explainable, governed, and repeatable. The real challenge, as always, is in execution: align data, close definitions, document controls, and ensure agents work for the business, not against it.
Frequently Asked Questions
What is Data Cloud Context Indexing, and how does it improve traditional document searches?
It is a streamlined indexing pipeline within Data Cloud that enables agents to interpret unstructured content — contracts, diagrams, tables — through business semantics. Instead of just keyword search, responses are anchored in entities, relationships, and corporate rules to provide step-by-step instructions or resolutions: for example, guiding a technician through a technical schematic using a decision tree.
How do the Data Cloud Clean Rooms “zero copy” features work, and what use cases are they best suited for?
They allow collaborative analysis without moving or exposing raw data. Zero-copy connectivity prevents duplication of sensitive records, reduces security/compliance risks, and lowers storage costs. Typical use cases include fraud prevention among banks, retail partnership analytics, or campaign measurement with third parties, all without sharing personal data.
What does MuleSoft Agent Fabric solve regarding “agent sprawl,” and how is it deployed?
It addresses agent proliferation by providing a central registry, orchestration, and governance platform for all agents—independent of origin. It enables workflow coordination, avoids duplication, closes compliance blind spots, and traces decisions. Governance is live, with Registry, Broker, and Visualizer available, and these tools launching in October 2025.
What is Tableau Semantics, and how does it unify metrics like “ACV” across an organization?
Tableau Semantics is an AI-driven semantic layer integrated into Data Cloud that translates raw data into business language. It includes a ready-to-use Customer 360 SDM and collaborates with partners like Databricks, dbt Labs, and Snowflake to create an open semantic exchange. This allows different teams to define metrics like “ACV” consistently, so insights are coherent and explainable across dashboards and agents.
via: salesforce