What Are Agentic Work Units? How Salesforce AgentForce Pricing Works
Salesforce now bills on agentic work units, not seats. Learn what counts as a billable action, how credits work, and what it means for your AI automation costs.
The End of Per-Seat AI Pricing
Salesforce just changed how enterprise software gets billed — and if you’re evaluating AI automation tools, the implications go further than one vendor’s pricing page.
With AgentForce, Salesforce moved away from the seat-based licensing model that’s dominated SaaS for two decades. Instead, you pay per agentic work unit — a unit of billable AI agent activity. Understanding what that means, what counts, and how the math actually plays out is essential before you commit budget to any AI automation platform.
This article breaks down exactly how Salesforce AgentForce pricing works, what constitutes a billable action, how the credit system functions, and what this consumption-based model signals about where enterprise AI is headed.
What Is an Agentic Work Unit?
An agentic work unit is a discrete unit of productive work completed by an AI agent — as opposed to a human sitting at a desk for eight hours.
The term reflects a fundamental shift in how AI software is valued. Traditional SaaS charges you for access (seats). Consumption-based AI charges you for output — what the agent actually did.
Think of it like this:
- Old model: Pay $X/user/month for a CRM. Everyone who logs in counts, whether they close ten deals or zero.
- New model: Pay $X per thing the agent accomplished — a resolved support ticket, a completed data lookup, a finished customer conversation.
Coding agents automate the 5%. Remy runs the 95%.
The bottleneck was never typing the code. It was knowing what to build.
Salesforce uses the word “conversation” as its primary unit. But “agentic work unit” is the broader industry framing — a single, bounded piece of work the agent performs, start to finish.
Why the Terminology Matters
Different vendors use different words for the same concept:
- Salesforce AgentForce: conversations
- Microsoft Copilot: messages (within copilot interactions)
- Anthropic API: tokens
- Many automation platforms: tasks, runs, or executions
The underlying logic is the same: you’re billed for what the AI does, not how many people have access to it. Knowing this helps you compare pricing across platforms more accurately.
How Salesforce AgentForce Pricing Actually Works
AgentForce launched at Salesforce’s Dreamforce conference in late 2024, with a consumption model priced at $2 per conversation.
Salesforce defines a conversation as a complete interaction between an AgentForce agent and a user or customer — from the moment the agent engages to when it resolves (or hands off) the issue. That single $2 unit covers the full back-and-forth within that session, including multiple turns, tool calls, and reasoning steps the agent takes along the way.
What’s Included in One Conversation
A single $2 conversation can include:
- Multiple message exchanges between the agent and the end user
- Agent reasoning steps (planning what to do next)
- Tool calls — like querying Salesforce data, checking order history, or updating a record
- Handoff decisions (escalate to human, close the ticket, etc.)
You’re not billed per message or per API call within a conversation. The meter runs once per session, not once per action inside it.
The Flex Credits System
For customers who already have Salesforce contracts, AgentForce also plugs into the Flex Credits system. Flex Credits are Salesforce’s internal currency for consumption-based products.
One AgentForce conversation typically costs 100 Flex Credits. Organizations with existing credit pools can draw down against those rather than purchasing conversation capacity separately.
This matters for enterprise customers renegotiating contracts — you may already have headroom in your credits allocation that can cover initial AgentForce usage without a new purchase order.
Volume Pricing and Enterprise Agreements
The $2/conversation list price isn’t fixed for large deployments. Enterprise customers negotiating annual agreements can expect lower per-conversation rates at scale. Salesforce has signaled that the model is designed to undercut the cost of human agent time — at $2 per conversation versus roughly $1 per minute for a human support agent, even a five-minute conversation costs the same as roughly ten AI-handled interactions.
The exact volume thresholds aren’t publicly documented, but the structure follows standard enterprise software discounting: commit to higher volumes upfront and your unit cost drops.
What Counts as a Billable Action?
This is where things get nuanced — and where organizations sometimes get surprised by their bills.
Billable: Agent-Initiated Conversations
Any conversation that an AgentForce agent starts or meaningfully participates in counts toward billing. This includes:
- Customer service interactions on web chat, messaging apps, or email
- Agent-assisted sales interactions where the AI engages a prospect
- Internal employee-facing workflows (HR queries, IT helpdesk, policy lookups)
- Outbound agent interactions the AI initiates proactively
Billable: Actions Within Conversations (Depending on Configuration)
Remy doesn't write the code. It manages the agents who do.
Remy runs the project. The specialists do the work. You work with the PM, not the implementers.
While the base $2 covers the conversation itself, some premium agent actions may carry additional costs depending on how your org is configured. For example:
- Accessing third-party data connectors
- Running complex reasoning chains that require extended compute
- Triggering certain Salesforce Flow automations through the agent
Salesforce has not published a comprehensive list of what triggers premium action billing, which is a real pain point for procurement teams trying to forecast costs.
Not Billable: Background Automation Without Agent Interaction
Purely automated workflows that run without an agent session — standard Salesforce Flow, Process Builder, or Apex triggers — continue to operate under existing licensing terms. AgentForce billing only kicks in when an intelligent agent actively participates in the work.
The Gray Area: Escalations and Handoffs
If a conversation gets escalated from AgentForce to a human agent mid-session, the AI portion still counts as one billable conversation. The human interaction then falls under your existing Service Cloud licensing. You don’t get a refund on partial conversations, so agents that frequently escalate before resolving issues can inflate costs without delivering proportional value.
How to Estimate Your AgentForce Costs
Before committing to AgentForce, you need a realistic cost model. Here’s a practical framework.
Step 1: Measure Current Interaction Volume
Pull your existing customer service or sales interaction data:
- How many inbound support tickets per month?
- How many of those could realistically be resolved without human escalation?
- What’s your average handle time per interaction?
AgentForce will only be cost-effective for interactions the AI can actually handle. A resolution rate of 30% means 70% of your conversations still need human handling — so your effective AI cost is spread across a smaller base.
Step 2: Model Your Containment Rate
Containment rate = the percentage of conversations the AI resolves without escalation.
| Containment Rate | Conversations/Month | Monthly Cost |
|---|---|---|
| 40% | 10,000 AI-handled | $20,000 |
| 60% | 15,000 AI-handled | $30,000 |
| 80% | 20,000 AI-handled | $40,000 |
At $2/conversation, the math is simple. The variable is how well your agents are trained to handle your specific use cases.
Step 3: Compare Against Human Agent Cost
A human agent at $15–$20/hour handling 8–10 conversations per hour costs roughly $1.50–$2.50 per conversation on labor alone — before overhead, benefits, and training. At current list pricing, AgentForce sits at rough parity with human labor cost per interaction.
The value case isn’t purely cost reduction; it’s scale (agents handle unlimited concurrent conversations) and availability (24/7 without overtime).
Step 4: Factor in Setup and Training Costs
Building effective AgentForce agents requires:
- Data Cloud integration (often an additional license)
- Agent configuration and testing time
- Ongoing prompt refinement as edge cases emerge
- Quality monitoring
These aren’t billed per conversation, but they’re real costs that belong in your total cost of ownership calculation.
The Broader Shift: Why Consumption Pricing Is Taking Over
Salesforce isn’t alone here. The move from seat-based to consumption-based pricing is happening across the enterprise software stack, driven by a simple economic reality: AI agents don’t need “seats” in any meaningful sense.
A seat implies a human who logs in, uses the software, and logs out. An AI agent can run continuously, serve thousands of customers simultaneously, and never take a lunch break. Charging per seat for something that can clone itself infinitely doesn’t make sense.
What This Means for Budget Owners
Consumption pricing shifts cost behavior in ways that catch finance teams off guard:
Predictability is harder. Usage-based billing means your monthly costs scale with activity. A viral product launch or a seasonal spike sends your AI agent bill up automatically — unlike a flat seat license.
Optimization becomes ongoing work. You’re now incentivized to make each conversation as efficient as possible. Agents that resolve issues quickly cost the same as agents that ramble — but efficient agents handle more volume at the same cost.
Procurement processes need to change. Annual seat licenses are easy to approve. Variable consumption budgets require new approval workflows and budget buffers.
Comparing AgentForce to Other Consumption Models
| Platform | Unit | Base Price |
|---|---|---|
| Salesforce AgentForce | Per conversation | $2.00 |
| Microsoft Copilot (M365) | Per message | Varies by product |
| Intercom Fin AI | Per resolution | $0.99 |
| Zendesk AI | Per automated resolution | Custom |
Intercom’s Fin agent is worth noting directly: at approximately $0.99 per resolution (only charged when the AI fully resolves the issue), it’s more aggressive on price — but limited to customer support use cases. AgentForce’s broader CRM context gives it an advantage for more complex sales and service workflows.
Where MindStudio Fits Into This Picture
If AgentForce’s pricing model raises questions for your team — about cost predictability, vendor lock-in, or building AI workflows without enterprise licensing complexity — MindStudio is worth understanding as an alternative or complement.
MindStudio is a no-code platform for building AI agents and automated workflows. Unlike AgentForce, which is tightly embedded in the Salesforce ecosystem, MindStudio is tool-agnostic — it connects to Salesforce natively alongside 1,000+ other business tools, so you can build agents that work across your entire stack, not just within one CRM.
A few differences worth noting:
- Pricing model: MindStudio plans start at $20/month. Rather than paying per conversation or per action, you’re paying for platform access — more predictable for teams building internal tools.
- Model flexibility: MindStudio gives you access to 200+ AI models (Claude, GPT-4, Gemini, and others) without requiring separate API keys. AgentForce uses Salesforce’s Einstein models, with limited ability to swap in other providers.
- Build time: Most agents in MindStudio take 15 minutes to an hour to build using the visual editor. There’s no equivalent of Salesforce’s Data Cloud setup requirement.
- Use cases: AgentForce is purpose-built for customer-facing CRM workflows. MindStudio covers that territory but also handles internal automation, content workflows, data processing, and custom AI apps — without needing Salesforce at the center.
For teams that want Salesforce integration without full AgentForce commitment, MindStudio’s Salesforce connector lets you build agents that read and write CRM data while avoiding per-conversation billing.
You can try MindStudio free at mindstudio.ai.
Common Mistakes Organizations Make With AgentForce Pricing
Assuming Every Conversation Gets Resolved
The $2 charge applies whether the agent succeeds or escalates. If your agent has a low containment rate, you’re paying for a lot of conversations that still require human follow-up. Test and optimize containment before scaling volume.
Built like a system. Not vibe-coded.
Remy manages the project — every layer architected, not stitched together at the last second.
Ignoring Data Cloud Requirements
AgentForce’s most powerful features — like grounding agents in customer history or using Einstein’s retrieval-augmented generation — require Salesforce Data Cloud, which carries its own licensing cost. Many organizations discover this after signing the AgentForce agreement.
Treating It Like a Set-and-Forget Deployment
Consumption pricing rewards ongoing optimization. Teams that configure their agents once and walk away tend to see costs creep up as edge cases multiply and escalation rates stay high. Budget for quarterly agent tuning as a recurring activity.
Not Negotiating Credit Structures Upfront
If you’re a Salesforce customer already spending six or seven figures annually, you have leverage to negotiate AgentForce credits into your renewal. Buying conversations at list price when you could have bundled them is leaving money on the table.
Frequently Asked Questions
What is an agentic work unit in Salesforce AgentForce?
In AgentForce’s implementation, the primary agentic work unit is a conversation — a complete AI agent interaction priced at $2 per session. The broader term “agentic work unit” refers to any discrete, billable unit of AI agent activity. Salesforce chose conversations as their unit because it maps cleanly to customer service interaction volumes that teams already measure.
Does Salesforce charge per message or per conversation?
Per conversation. A conversation includes all messages, reasoning steps, and tool calls within a single agent session. You’re not billed per turn or per tool call — the $2 covers the entire interaction from start to finish (or handoff).
How is AgentForce different from Einstein Bots?
Einstein Bots were Salesforce’s earlier chatbot product — rule-based, scripted, and primarily useful for simple deflection tasks like FAQs. AgentForce uses large language models with genuine reasoning capability, can take multi-step actions, connects to live data through Data Cloud, and operates across multiple channels. The pricing model is also completely different: Bots were included with Service Cloud licenses, while AgentForce is consumption-billed separately.
Can you cap AgentForce spending to control costs?
Salesforce does offer spending controls, but the implementation varies by contract type. Enterprise agreements typically allow you to set conversation volume limits that trigger alerts or stops. Flex Credit-based deployments let you cap consumption at your credit balance. Either way, you’ll want to configure spend guardrails before launching to production traffic.
Is AgentForce worth it compared to building custom AI agents?
It depends on how deeply embedded you are in the Salesforce ecosystem and what you’re automating. AgentForce has a genuine advantage for organizations that want native CRM context without custom integration work. But for teams that want multi-tool automation, model flexibility, or lower per-task costs, custom agents built on platforms like MindStudio — or directly on model APIs — often deliver better economics and more control.
How do Flex Credits relate to AgentForce conversations?
Everyone else built a construction worker.
We built the contractor.
One file at a time.
UI, API, database, deploy.
Flex Credits are Salesforce’s internal billing currency used across multiple consumption products. One AgentForce conversation typically costs 100 Flex Credits. If your organization already has Flex Credits as part of an enterprise agreement, you can use that existing balance to fund AgentForce usage rather than purchasing conversation capacity separately.
Key Takeaways
- Agentic work units represent a shift from access-based to output-based AI pricing — you pay for what the agent does, not who has a license.
- AgentForce charges $2 per conversation, covering the full agent interaction including all reasoning steps and tool calls within that session.
- Flex Credits (approximately 100 per conversation) let existing Salesforce customers tap existing balances rather than buying new capacity.
- Containment rate — the percentage of conversations the agent fully resolves — is the single most important variable in your cost model. Low containment means high costs for limited value.
- Billable actions include agent-initiated conversations, customer service sessions, and sales interactions. Background automation without agent involvement does not trigger AgentForce billing.
- Data Cloud licensing is often a hidden prerequisite for AgentForce’s most useful features — factor it into total cost of ownership calculations.
For teams that want AI automation across their full tool stack — not just within Salesforce — MindStudio offers a more flexible entry point with predictable pricing and native Salesforce integration alongside 1,000+ other connectors.
