Salesforce Agentforce Winter ’27 and New Editions: The Complete Guide for Agentic AI Architects (2026)

On September 3, 2026, Salesforce replaced its legacy Enterprise, Unlimited, and Agentforce 1 editions with three new tiers — Core ($195/user/month), Advanced ($395), and Max ($550) — bundling Agentforce, Slack, Tableau Next, enterprise security, and Premier Success Plan into every seat. Five days earlier, on August 31, the company unveiled Winter ’27: fifteen production-ready innovations that shift AI agents from task-level assistants to autonomous workflow executors across sales, service, commerce, marketing, and field operations. Together, these two announcements represent the most aggressive agentic transformation play any enterprise CRM vendor has made to date.

This guide unpacks both announcements through the lens that matters to practitioners: what changed architecturally, what it means for multi-agent orchestration, and how Agentic AI Architects should evaluate Agentforce in a landscape where UiPath, Microsoft, AWS, and Google are all racing to own the enterprise agent layer.

Table of Contents

• The New Editions: Core, Advanced, and Max
• Winter ’27 Top 10 Innovations
• Third-Party Agent Orchestration and A2A
• Agent Skills and Plugins: The Composability Layer
• Agentic Commerce Search: 13% Conversion Lift in Production
• Autonomous Scheduling with Agentforce Voice
• Tableau Knowledge and the Data-Grounding Problem
• Agentforce by the Numbers: Adoption Metrics That Matter
• How Agentic AI Architects Should Evaluate Agentforce
• Agentforce vs the Competition: Where It Fits in 2026
• FAQs
• Key Takeaways
• References

The New Editions: Core, Advanced, and Max

Salesforce’s edition restructure is not a cosmetic rebrand — it is a pricing strategy designed to eliminate the SKU fragmentation that slowed enterprise AI adoption. Previously, an organization wanting Agentforce plus Slack plus Tableau plus enterprise security had to piece together multiple SKUs, each with its own contract and renewal cycle. The new editions collapse all of that into a single per-user-per-month subscription.

Edition Comparison

The Flex Credits model deserves attention from architects. Rather than paying per-agent or per-conversation, credits are consumed based on Agentic Work Units (AWUs) — a composite metric that factors in reasoning complexity, tool calls, and data retrieval. This means a simple FAQ-resolution agent consumes far fewer credits than a multi-step orchestration agent that queries three systems and writes back to two. For capacity planning, this is the number to model: not “how many agents” but “how much agentic compute does each workflow consume.”

The Max tier at $550/user/month holds the same price as the outgoing Agentforce 1 Edition while bundling roughly 60% more capability (Salesforce, Sep 2026). For existing customers on Enterprise or Unlimited editions, this is a consolidation play — fewer contracts, fewer renewal negotiations, and AI capabilities that were previously add-ons now included in the base seat.

Winter ’27 Top 10 Innovations

The Winter ’27 release (GA: October 12, 2026) contains fifteen innovations. Salesforce organized the top ten around a single thesis: agents no longer assist — they execute entire workflows end-to-end, across every channel the enterprise already uses (Salesforce Winter ’27 Release Announcement).

Here is the complete top-ten list with architectural significance for each:

1. Slack Code: Multiplayer Agentic Development

Slack Code moves software development from isolated IDE tabs into dedicated code channels where humans and AI agents write, review, and ship code together. The architectural shift: code execution is now connected to conversational context. A product manager can describe a requirement in Slack, an AI agent writes the implementation, a developer reviews inline, and the merged PR links back to the original thread. This closes the gap between solo AI-assisted coding and team collaboration.

More Read

Microsoft Work IQ and the Work IQ MCP Server: The Complete Guide for Agentic AI Architects (2026)

DigitalOcean Managed Agents: How Firecracker MicroVMs and 16,000+ MCP Tools Are Reshaping Agent Infrastructure (2026)

Robotics Process Automation in Finance and Accounting

2. Adaptive Experiences and Dynamic Plans

For service teams handling complex, evolving cases, an ambient AI agent now listens to case and messaging conversations in real time, reasoning over live context to generate and dynamically update resolution plans as the issue changes. Four customers are already live in production, including PowerSchool with more than 550 users. The broader Service Rep Assistant program has surpassed 100 customers. This is not a demo — it is a deployed capability with measurable adoption.

3. Agentforce Contact Center: 30+ Countries

Agentforce Contact Center is now globally available, bringing CRM, channels (including voice), and AI into a single solution. The key architectural point: voice is built directly into the CRM rather than relying on third-party contact center integrations. This means human and AI agents share the same real-time customer data layer, eliminating the sync lag that plagues bolt-on telephony integrations.

4. Tableau Knowledge: Governed Business Context for Agents

Tableau Knowledge automatically ingests structured and unstructured data, generates living knowledge graphs, and delivers curated context to agents via open standards. The practical impact: agents that previously hallucinated on business-specific questions (product availability, pricing exceptions, policy nuances) now reason over governed enterprise data rather than relying solely on their base model’s training data.

5. Informatica Headless: MCP-Native Data Access

Informatica Headless decouples IDMC’s backend from its UI, letting developers call enterprise-grade data management directly from VS Code, Cursor, Slack, or via universal MCP servers. For Agentic AI Architects, this is significant: it means agents can access governed, clean data without leaving their runtime environment, reducing hallucination loops caused by stale or ungoverned data inputs.

6. Agentic Segmentation and Activation

Marketers describe target audiences in plain language, and a Segmentation agent builds the audience plan while verifying the underlying data. An Activation agent then sets up campaigns conversationally across ad platforms, flagging errors and suggesting optimizations before reaching external partners. This removes the weeks-long bottleneck of manual data modeling and multichannel configuration.

7. Autonomous Scheduling with Agentforce Voice

Cuts 15-minute appointment bookings to seconds. Customers book 24/7 using agents that check real-time availability, find the best technician in the right territory, and complete the booking without human handoff. We cover this in detail below.

8. Agent Skills and Plugins

Over 100 prebuilt skills at launch, with a unified discovery and reuse model across every surface. The Salesforce Development plug-in is already live in the Claude Code marketplace. Full analysis below.

9. Third-Party Agent Orchestration

Native A2A-compliant orchestration for multi-vendor agent environments. Full analysis below.

10. Agentic Commerce Search

A dedicated small language model trained per-merchant on catalog and shopper behavior data. 13% conversion lift and 17% add-to-cart rate increase in production. Full analysis below.

Third-Party Agent Orchestration and A2A

This is the Winter ’27 feature that matters most to Agentic AI Architects building in multi-vendor environments — which, at this point, is nearly every enterprise.

The problem is real: an organization running Agentforce for CRM workflows, Azure AI agents for IT systems, and AWS agents for cloud infrastructure faces fragmented handoffs at every vendor boundary. A workflow that crosses these boundaries historically required a human intermediary or expensive custom integration code.

Agentforce’s Third-Party Orchestration solves this through native support for the Agent-to-Agent (A2A) protocol — the open interoperability standard introduced by Google in April 2025 and now governed by the Linux Foundation’s Agentic AI Foundation with 150+ organizations running it in production (Google Cloud Blog).

How It Works Architecturally

A2A defines Agent Cards — cryptographically signed, machine-readable capability descriptions that orchestrators read to determine which specialist agent is best suited for each sub-task. With Agentforce’s native Third-Party Orchestration:

• Agentforce as orchestrator: A primary Agentforce agent handles CRM-layer work and delegates out-of-CRM tasks (infrastructure provisioning, IT ticket resolution, ERP updates) to specialized agents on AWS, Azure, Google, or any A2A-compliant platform.
• Agentforce as sub-agent: In architectures where an external orchestrator (e.g., AWS Bedrock AgentCore) already exists, Agentforce can participate as a CRM-specialist sub-agent, exposing its capabilities via A2A Agent Cards.

No custom integration code is needed for each delegation path. The Agent Cards make capabilities machine-readable, so orchestration decisions are automated rather than manually configured. This is the same protocol that Microsoft, AWS, SAP, and ServiceNow already run in production environments (Google Developers Blog).

For architects evaluating multi-agent topologies, the practical implication is clear: Agentforce is no longer a walled garden. It can participate in — or orchestrate — heterogeneous agent ecosystems without forcing the entire stack to be Salesforce-native.

Agent Skills and Plugins: The Composability Layer

Every new AI agent starts with zero organizational context, forcing it to reconstruct best practices from scratch — or guess. Agent Skills and Plugins addresses this with a composability model that should be familiar to anyone who has worked with MCP servers or plugin architectures in other agent frameworks.

The Architecture

• Skills are atomic, reusable capabilities — a skill for “qualify a lead against ICP criteria,” “check inventory across three warehouses,” or “verify insurance policy coverage.” Salesforce launched with 100+ prebuilt skills covering common platform workflows.
• Plugins bundle skills together with MCP connectors, hooks, and configuration. A single plugin install brings a complete, governed capability to any surface an agent runs on.
• Registry — both prebuilt and custom skills/plugins are discoverable through a unified registry, meaning agents across different surfaces (Slack, web, voice, mobile) can find and execute the same governed capabilities.

The Salesforce Development plug-in is already live in the Claude Code marketplace (Salesforce), demonstrating that this isn’t a Salesforce-only ecosystem. Developers build reusable capabilities once; agents execute them consistently and securely across every surface the enterprise runs.

For Agentic AI Architects, this composability layer is what transforms Agentforce from a chatbot platform into an enterprise agent operating system. The comparison to draw: UiPath’s package management for RPA workflows, but applied to AI agent capabilities rather than UI automation sequences.

Agentic Commerce Search: 13% Conversion Lift in Production

Agentic Commerce Search is the Winter ’27 feature with the hardest production metrics attached to it. Instead of traditional keyword-match search, a dedicated small language model (SLM) is trained on each merchant’s specific catalog data and shopper behavior patterns.

The results from live deployments (Salesforce):

• 13% conversion lift from intent-driven search
• 17% increase in add-to-cart rates
• Now standard for all Shopper Agent customers

The architectural insight here: Salesforce is not applying a generic LLM to search. Each merchant gets a purpose-trained SLM that understands their specific catalog taxonomy, product relationships, and buyer intent patterns. This is the same approach that has proven effective in enterprise RAG systems — domain-specific fine-tuning outperforms general-purpose models on specialized tasks — but applied to commerce at scale.

For architects evaluating agentic commerce, the question shifts from “can AI improve search?” to “does the vendor train per-tenant models or rely on a shared model?” Salesforce’s per-merchant SLM approach ensures that a sporting goods retailer’s search agent understands “trail runners” differently than a shoe manufacturer’s agent does.

Autonomous Scheduling with Agentforce Voice

Field service scheduling — the kind where a customer calls in, a dispatcher checks technician availability, territory assignments, skill requirements, and drive time, then manually books the appointment — is one of the highest-overhead, most error-prone processes in service operations. Average call time: 15 minutes.

Autonomous Scheduling with Agentforce Voice compresses this to seconds. The agent:

1. Answers the customer’s voice call (24/7, no hold queue)
2. Identifies the service needed from natural conversation
3. Checks real-time technician availability, territory boundaries, and skill match
4. Books the appointment with the best available technician
5. Confirms with the customer — all without human handoff

The business case is straightforward: dispatch operations scale with agent capacity, not headcount. For organizations with seasonal demand spikes (HVAC, plumbing, utilities), this eliminates the need to staff temporary dispatchers during peak periods.

For Agentic AI Architects, the technical takeaway is the integration depth — this feature works because Agentforce Voice has direct access to the CRM’s real-time scheduling engine, territory definitions, and technician skill profiles. A bolted-on voice bot making API calls to a separate scheduling system would face latency and data-freshness issues that make real-time booking impractical.

Tableau Knowledge and the Data-Grounding Problem

The most common failure mode in enterprise AI agent deployments is not model capability — it is data grounding. An agent that can reason brilliantly but hallucinates on company-specific facts (pricing exceptions, inventory counts, policy nuances) is worse than useless; it is actively dangerous.

Tableau Knowledge attacks this problem by:

1. Automatic ingestion of both structured (databases, spreadsheets, CRM records) and unstructured data (documents, PDFs, wikis)
2. Living knowledge graph generation — not a static snapshot, but a continuously updated graph that reflects current business state
3. Context delivery via open standards — agents receive curated, governed context rather than raw data dumps

Business experts stay in control of what enters the knowledge graph and how it is curated. This addresses a known governance gap: most enterprise agent deployments fail not because the model is wrong, but because nobody governed what data the agent was reasoning over.

The combination of Tableau Knowledge (data grounding) + Informatica Headless (MCP-native governed data access) gives Agentforce a data infrastructure story that most competing agent platforms lack. Google’s Vertex AI Agent Builder has strong data grounding through Vertex AI Search, and AWS Bedrock has Knowledge Bases — but neither bundles a governed analytics/knowledge-graph layer directly into the agent seat the way Salesforce now does.

Agentforce by the Numbers: Adoption Metrics That Matter

Vendor claims about AI adoption are easy to inflate. Here are the metrics that survived earnings scrutiny and third-party analysis:

Two numbers stand out for architects: the 15% compound monthly growth in Agentic Work Units tells you that deployed agents are not sitting idle — they are executing progressively more complex workflows month over month. And the 53% reduction in agent creation time suggests the tooling is maturing past the “needs a specialist to build anything” phase.

The $800M ARR at 169% growth puts Agentforce on a trajectory to become a $2B+ standalone product line within 18 months. For enterprise procurement decisions, this level of revenue momentum reduces platform risk — Salesforce is clearly investing in Agentforce as a core business line, not a feature experiment.

How Agentic AI Architects Should Evaluate Agentforce

With the Winter ’27 release and new editions, Agentforce has matured from a “CRM chatbot add-on” into a genuine enterprise agent platform. But maturity does not mean it is the right choice for every architecture. Here is how to evaluate it:

Choose Agentforce When

• Salesforce is already your CRM — the deepest integration value comes from native CRM data access, and the new editions make AI capabilities a bundled benefit rather than an add-on cost
• Your agent workflows are CRM-centric — lead qualification, service case resolution, field service scheduling, commerce search, marketing activation
• You need multi-vendor agent orchestration — A2A-compliant Third-Party Orchestration makes Agentforce the strongest CRM-layer participant in heterogeneous agent architectures
• Data governance is a hard requirement — Tableau Knowledge + Informatica Headless provide governed data grounding that most competitors require you to build yourself
• You want composable agent capabilities — the Skills/Plugins architecture with MCP connectors enables build-once-deploy-everywhere patterns

Look Elsewhere When

• Your workflows are not CRM-centric — infrastructure automation, code generation, data pipeline orchestration, and similar workloads are better served by AWS Bedrock AgentCore, Azure AI Foundry, or open-source frameworks like LangGraph
• You need full control over the agent runtime — Agentforce is a managed platform; if you need to customize the reasoning loop, tool-calling protocol, or memory architecture at a low level, open-source frameworks provide that flexibility
• Cost sensitivity at scale — at $195-$550/user/month, the Flex Credit model needs careful capacity planning. High-volume, low-complexity automations may be more cost-effective on consumption-based platforms
• You are standardized on a competing RPA platform — if your automation backbone is UiPath, Automation Anywhere, or Blue Prism, their native agentic capabilities may integrate more naturally with existing RPA investments

Agentforce vs the Competition: Where It Fits in 2026

The competitive reality in September 2026 is that no single vendor owns the enterprise agent layer. What has changed is that each vendor now has a credible, production-ready offering with real adoption numbers. Agentforce’s differentiation is CRM-native depth and the bundled data-governance stack; its limitation is that it is most valuable when Salesforce is already the CRM backbone.

For the broader landscape of how Microsoft’s Agent Framework 1.0 and other platforms are positioning, the architectural pattern converging across all vendors is the same: A2A for inter-agent communication, MCP for tool/data integration, and governed knowledge layers to prevent hallucination in production.

Also Noteworthy: Claudeforce

On August 26, 2026, Salesforce and Anthropic announced Claudeforce — a deeper integration between Claude and Agentforce (Salesforce Press Release). With the Salesforce Development plug-in already live in the Claude Code marketplace, this partnership signals that Agentforce’s agent runtime is model-agnostic in practice, not just in architecture diagrams. Enterprises can use Claude, GPT, Gemini, or open-source models as the reasoning backbone for different agent specializations within the same Agentforce deployment.

Frequently Asked Questions

What is the difference between Salesforce Agentforce Core, Advanced, and Max editions?

All three editions bundle Agentforce AI agents, Slack, Tableau Next, enterprise security, and Premier Success Plan. The primary differences are Flex Credits (500K / 1M / 2.75M respectively) and pricing ($195 / $395 / $550 per user per month). Max delivers approximately 60% more value than the outgoing Agentforce 1 Edition at the same price point.

Does Agentforce support multi-vendor agent orchestration?

Yes. Winter ’27 introduces native Third-Party Agent Orchestration using the A2A (Agent-to-Agent) protocol. Agentforce can orchestrate agents from AWS, Azure, Google, and any A2A-compliant platform, or participate as a sub-agent in external orchestration architectures.

When is the Salesforce Winter ’27 release generally available?

October 12, 2026. Live deployments will span 30+ countries.

How does Agentforce pricing compare to consumption-based alternatives like AWS Bedrock?

Agentforce uses a per-seat model with Flex Credits consumed by Agentic Work Units. AWS Bedrock AgentCore uses pure consumption-based pricing. For CRM-centric workloads, Agentforce’s bundled approach is often more cost-effective; for high-volume, infrastructure-heavy workloads, consumption-based pricing on AWS or Google Cloud typically costs less.

What production results has Agentic Commerce Search achieved?

Live deployments report a 13% conversion lift and 17% increase in add-to-cart rates using a per-merchant small language model trained on each merchant’s specific catalog and shopper behavior data.

Key Takeaways

• Edition consolidation eliminates SKU sprawl: Core/Advanced/Max bundle Agentforce + Slack + Tableau Next + security + support into single per-seat subscriptions, simplifying procurement and deployment.
• Winter ’27 shifts agents from assistants to autonomous executors: 15 production-ready innovations cover sales, service, commerce, marketing, field ops, and development — with real deployment metrics, not demos.
• A2A-compliant Third-Party Orchestration is the breakthrough for multi-vendor shops: Agentforce can orchestrate or be orchestrated by agents on any A2A-compliant platform, ending the walled-garden era for CRM agents.
• Agent Skills and Plugins create a composable capability layer: Build once, deploy across every surface. MCP connectors and the plugin registry make Agentforce agents reusable and governed.
• Adoption is real, not aspirational: $800M ARR, 18,000+ customers, 11.14 trillion tokens served, and 15% compound monthly growth in agentic work units.
• Data grounding is the differentiator: Tableau Knowledge + Informatica Headless give Agentforce a governed data stack that most competitors require you to assemble yourself.
• Evaluate by workflow fit: Agentforce excels for CRM-centric agent workflows. Infrastructure automation, code generation, and fully custom agent architectures are better served by other platforms.

References

1. Salesforce, “Top Innovations in the Winter ’27 Release: AI Agents Run the Work, Freeing the Enterprise to Work Smarter and Faster,” August 31, 2026. https://www.salesforce.com/news/stories/winter-2027-product-release-announcement/
2. Salesforce, “New Salesforce Editions Bundle Everything Businesses Need for Agentic Transformation,” September 3, 2026. https://www.salesforce.com/news/stories/salesforce-simplifies-editions-2026/
3. Salesforce and Anthropic, “Salesforce and Anthropic Announce Claudeforce,” August 26, 2026. https://www.salesforce.com/news/press-releases/2026/08/26/salesforce-and-anthropic-announce-claudeforce/
4. Futurum Group, “Salesforce Q4 FY 2026 Earnings Show Agentic AI Scaling, Guidance Steadies,” 2026. https://futurumgroup.com/insights/salesforce-q4-fy-2026-earnings-show-agentic-ai-scaling-guidance-steadies/
5. Futurum Group, “Salesforce Winter ’27: Agents Now Run the Workflow,” 2026. https://futurumgroup.com/insights/salesforce-winter-27-agents-now-run-the-workflow/
6. Salesforce, “Agentforce Metrics: Real Impact & Results.” https://www.salesforce.com/agentforce/metrics/
7. Salesforce Ben, “Salesforce Announces 3 Replacement Editions, Bundling AI, Slack, and Security,” 2026. https://www.salesforceben.com/salesforce-announces-3-replacement-editions-bundling-ai-slack-and-security/
8. CIO, “Salesforce Offers More Agentforce Credits to Drive Adoption,” 2026. https://www.cio.com/article/4218804/salesforce-offers-more-agentforce-credits-to-drive-adoption.html
9. TechTimes, “Salesforce Agentforce Winter ’27: AI Agents Now Run Full Enterprise Workflows in Production,” September 1, 2026. https://www.techtimes.com/articles/326164/20260901/salesforce-agentforce-winter-27-ai-agents-now-run-full-enterprise-workflows-production.htm
10. Google Developers Blog, “Announcing the Agent2Agent Protocol (A2A).” https://developers.googleblog.com/en/a2a-a-new-era-of-agent-interoperability/
11. Salesforce, “Salesforce Announces State of Sales Report for 2026.” https://www.salesforce.com/news/stories/state-of-sales-report-announcement-2026/