Salesforce Insights for Regulated Industries | Vantage Point

Salesforce Enterprise AI Harness: What Changed, What to Do Now

Written by David Cockrum | Nov 21, 2026, 1:00:00 PM

Quick Answer

 

Salesforce's Trusted Enterprise AI Harness, announced the week of Dreamforce 2026, is a governance and control-plane architecture for enterprise AI: six "trusted" capabilities (context, agency, action, governance, security, and models) plus a new AI Control Plane where you register agents, set identity and policy, observe behavior, and control cost — across Salesforce and third-party AI alike. The headline isn't a feature; it's a category shift: governance is becoming a product surface you buy and configure, not just a consulting engagement you commission. Many foundation technologies are available today; new capabilities and the unified experience begin rolling out in early Salesforce FY28. This post covers what changed, who's affected, and what to do now — exactly the kind of architecture work Vantage Point designs and implements.

Key Takeaways (TL;DR)

  • What it is: Salesforce's Trusted Enterprise AI Harness — six reusable capabilities that give AI agents context, reasoning, action, governance, security, and model flexibility in one composable architecture.
  • The real news: a new AI Control Plane gives enterprises one place to register agents, establish identity and policy, manage lifecycle, observe outcomes, and control cost — including third-party AI.
  • Why it matters: governance is shifting from bespoke consulting deliverables to a productized layer of the stack. Budgets and ownership should shift with it.
  • What it doesn't do: the Control Plane enforces policy; it doesn't write it. Data quality, permission models, use-case risk, and vendor contracts remain yours.
  • Timeline: foundations are available today; new capabilities begin rolling out early FY28, with upgrade paths for existing customers.

What Changed at Dreamforce

Announced just ahead of Dreamforce 2026 and featured at the show, the Enterprise AI Harness answers a question every enterprise AI program eventually hits: once agents stop being demos and start doing real work, how do you give them what they need reliably and securely, without rebuilding it for every agent?

Salesforce's answer: make the surrounding infrastructure a shared layer rather than a per-agent project. The Harness combines six capabilities drawn from Data 360, Informatica, MuleSoft and Agent Fabric, Tableau, Agentforce, Salesforce Guardian, and the core Salesforce Platform:

Capability What it gives AI Practical example
Trusted Context Data, metadata, semantics, knowledge, signals, and memory, grounded in governed enterprise data An agent answering "can we fulfill this order?" understands the customer, their contract, and what "available" means for them
Trusted Agency Reasoning, planning, state, and orchestration, with deterministic controls where certainty is required The agent decides whether the order can be fulfilled while following rules that can't be bent
Trusted Action Secure connections to applications, APIs, workflows, and processes The agent reserves inventory, updates the order, or hands off to a person
Trusted Governance Lineage, quality, guardrails, and controls over the data and policies AI relies on The answer rests on trusted information and policy-compliant handling
Trusted Security Identity, permissions, privacy, data protection, and runtime security The agent sees and does only what it's authorized to
Trusted Models Routing to the right model per job by accuracy, performance, cost, and requirements Work routes to the right intelligence — and can change models as technology evolves

Two design decisions matter more than any single capability. First, the Harness is composable: use the six together or take only what you need, with Salesforce technology, your existing technology, or both — including third-party models, agents, and systems. Second, it's built headlessly, accessible through MCP, APIs, Skills, and Plug-ins, extending into Claude, Slack, Microsoft Teams, Agentforce, and other AI experiences — the "AIforce" strategy of bringing Salesforce to any AI, agent, app, or surface.

The AI Control Plane: Governance Becomes a Product

The most strategically significant piece is the new AI Control Plane — one place to see, manage, and control agents as they spread across teams, applications, models, and systems. Per Salesforce's announcement, it lets companies:

  • Discover and register agents and AI capabilities
  • Establish identity and policy
  • Manage lifecycle
  • Evaluate performance
  • Observe behavior and outcomes
  • Control cost — across Salesforce and third-party AI

Read that list from a consulting perspective. Agent inventory, identity, policy enforcement, audit and observability, cost governance: until now, those were deliverables. Firms like ours built them as governance frameworks and custom tooling on top of whatever the platform provided. Salesforce is now productizing the layer itself.

That's a pattern this decade keeps repeating in the Salesforce ecosystem: platform absorbs scaffolding. Integration monitoring became a product. DevOps became a product. Now AI governance becomes a product — and consulting value moves up a level, from building the plumbing to deciding what flows through it.

What the Control Plane Covers — and What You Still Own

The fastest way to misread this announcement is to assume a productized control plane means governance is "handled." It isn't. The Control Plane is an enforcement and visibility layer. The substance of governance — the policies, the data, the risk judgments — doesn't ship in a box.

Governance responsibility AI Control Plane Your organization
Agent discovery and registration ✅ Registry across Salesforce and third-party AI Deciding which agents should exist and what they're for
Agent identity and permissions plumbing ✅ Identity establishment and enforcement hooks The permission model itself — who agents act as, what data they can touch
Policy enforcement ✅ Applies policies consistently at runtime Writing the policies: acceptable use, data handling, escalation rules
Audit and observability ✅ Behavior and outcome monitoring Reviewing what you observe, acting on anomalies, evidencing compliance
Cost control ✅ Visibility and controls for AI spend Budgets, unit economics, and ROI accountability per use case
Data quality and lineage ✅ Tooling for lineage, quality, and guardrails Fixing the data — deduplication, stewardship, semantic definitions
Model routing ✅ Routing by accuracy, performance, cost Choosing acceptable models and providers under your regulatory posture
Vendor and contract risk ❌ Entirely yours: DPAs, residency, training-use terms, subprocessor review

This division of labor should look familiar if you read our analysis of Claude inside the Salesforce Trust Boundary: the platform answers "where does the model run, and what enforces the rules?" while you still answer "what are the rules, and can the agent see that?" The same logic applies to cost — the Control Plane will show you agent spend, but as we covered in the token economy trap, visibility without an operating model is how AI bills climb while token prices fall.

Who's Affected

CIOs and platform owners get a consolidation path: one governance layer across Salesforce and third-party AI instead of a bespoke stack per initiative. The planning question is which current investments map onto the six capabilities — and which become redundant.

CISOs, CCOs, and risk leaders get the enforcement machinery they've been asking for — identity, policy, observability — but inherit the harder half: writing policy an enforcement engine can actually apply. "Use good judgment" doesn't compile.

Data and analytics leaders are closest to the foundation. Trusted Context and Trusted Governance are only as strong as your metadata, semantics, and data quality — hence the Data 360 and Informatica lineage.

CFOs and procurement should note the cost-control mandate and the deliberate vagueness on packaging: pricing and upgrade details come closer to general availability. Budget for the capability, not yet for a line item.

Existing Salesforce customers get an explicit commitment: current Data 360, MuleSoft, Tableau, and Agentforce investments form the foundation, with upgrade paths as new capabilities ship.

What You Should Do Now

You don't need to wait for FY28. The foundations are available today, and the governance work sits on your side of the table regardless of when the unified experience ships.

  1. Inventory your agents and AI experiments. You can't register what you haven't found. Include the unofficial ones — individual AI licenses, automations with LLM steps, embedded AI features turned on by default.
  2. Assign ownership. Who is accountable for agent identity, policy, and spend? If the honest answer is "nobody," that's the gap the Control Plane will make visible — fix it before the tooling does it for you.
  3. Write enforceable policy. Convert AI acceptable-use guidance into rules a control plane can apply: which data classes agents may touch, which actions require human approval, what gets logged and retained.
  4. Audit permissions for agent-scale access. Agents retrieve at machine scale. Permission models configured for humans reading one record at a time become exposure the moment an agent exercises them ten thousand times a day.
  5. Strengthen the data foundation. Context is the durable differentiator in Salesforce's framing — and context quality is data quality. Prioritize deduplication, metadata, and semantic-layer work.
  6. Map your stack to the six capabilities. Identify where existing investments already cover a capability and where the Harness will consolidate or collide.
  7. Get pricing intelligence early. Packaging is unannounced. Ask your account team how upgrade paths apply to your editions before FY28 budgets lock.

Timeline and Availability

  • Today: many foundation technologies are already available — Data 360, MuleSoft and Agent Fabric, Tableau, Agentforce, Salesforce Guardian.
  • Early FY28: new capabilities and the unified experience begin rolling out. (Salesforce's fiscal year runs February through January, so FY28 begins in early 2027.)
  • Closer to GA: additional details on availability, packaging, pricing, and upgrade paths.
  • Existing customers: eligible Salesforce investments can be upgraded to unlock new capabilities as they ship.

The standard caveat applies: availability may vary by region, and purchasing decisions should be based on what's currently available. Treat FY28 as a planning horizon, not a dependency.

How Vantage Point Helps

This announcement sits squarely in the intersection where Vantage Point works: Salesforce architecture, AI governance, integration, and the organizational readiness that turns platform capabilities into business outcomes. Through our Salesforce implementation and advisory and compliance and security solutions practices, senior consultants help you inventory agents, write enforceable policy, audit permissions for agent-scale access, and map your stack to the six Harness capabilities — so that when the Control Plane arrives, you have something worth enforcing. And because the Harness is explicitly open to third-party models and systems, our system integration and data migration work keeps the MuleSoft and API foundations ready for a headless, composable architecture.

Senior consultants only — no junior handoffs. The experts you meet are the experts who deliver.

Ready to Get Ahead of the Control Plane?

 

Governance is becoming a product, but the policies, data, and decisions that make governance real are still yours. Vantage Point's senior consultants can run an AI governance readiness assessment — agent inventory, policy gap analysis, permissions audit, and capability mapping — on a practical timeline. Contact Vantage Point to schedule a readiness session and connect AI strategy to CRM operations.

Frequently Asked Questions

What is Salesforce's Trusted Enterprise AI Harness?

It's a governance and control-plane architecture for enterprise AI, announced at Dreamforce 2026. Six capabilities — Trusted Context, Agency, Action, Governance, Security, and Models — give AI agents shared context, secure action, and enterprise controls, built on Data 360, Informatica, MuleSoft, Tableau, Agentforce, Guardian, and the Salesforce Platform.

What is the Salesforce AI Control Plane?

A single place to see, manage, and control agents and AI across the enterprise: discover and register agents, establish identity and policy, manage lifecycle, evaluate performance, observe behavior, and control cost — across Salesforce and third-party AI.

When will the Enterprise AI Harness be available?

Many foundation technologies are available today. New capabilities and the unified experience are planned to begin rolling out in early Salesforce FY28, which starts in early 2027. Pricing, packaging, and upgrade-path details will be announced closer to general availability.

Does the AI Control Plane replace our AI governance program?

No. The Control Plane is an enforcement and visibility layer — it applies policies, tracks agents, and monitors cost and behavior. Your organization still owns the substance: writing enforceable policies, fixing data quality, defining permission models, judging use-case risk, and managing vendor contracts.

Does the Harness work with non-Salesforce AI like Claude or ChatGPT?

Yes — openness is a stated design goal. The Harness supports third-party models, agents, and systems; it's built headlessly with MCP, APIs, Skills, and Plug-ins, and extends into Claude, Slack, Microsoft Teams, and other AI experiences. The Control Plane explicitly covers third-party AI alongside Salesforce's.

What will the Enterprise AI Harness cost?

Salesforce hasn't announced pricing or packaging. Those details, along with upgrade paths for existing Data 360, MuleSoft, Tableau, and Agentforce investments, come closer to general availability. Ask your account team how upgrade eligibility applies to your editions before FY28 budget planning.

What should we do before the Control Plane ships?

Focus on the half of governance the platform won't do for you: inventory your agents (including unofficial ones), assign ownership for identity and spend, convert AI policy into enforceable rules, audit permissions for agent-scale access, and strengthen data quality. That work pays off regardless of which control plane you adopt.

Sources

Vantage Point is a boutique CRM consulting firm helping businesses transform with Salesforce, HubSpot, and AI. Learn more at vantagepoint.io.