Agentforce: what it is, how it works, and how to adopt Salesforce agentic AI

Agentforce is Salesforce’s agentic AI platform: a set of autonomous agents that understand a request, create a plan and execute tasks in the CRM without relying on a human at every step. Launched in October 2024 as Einstein’s successor, it became Salesforce’s central bet on the agentic enterprise, where people and AI agents work side by side. If Agentforce is the customer-facing AI layer, the internal-team layer also gained a new capability in Slack: see what Claude Tag is.

In Brazil, the topic has moved beyond curiosity and onto the boardroom agenda. The Agentforce World Tour São Paulo, in May 2026, brought together leaders from several industries around a single topic, and Salesforce projects an increase of 67% adoption of AI agents by 2027. For Salesforce users, the question is no longer whether it is worthwhile, but how to adopt it safely and with returns.

This guide directly answers what Agentforce is, how it works internally, how much it costs and how to put it into production. WeeNow is an official Salesforce partner and works precisely at this point: taking agentic AI from pilot to operations.

What is Agentforce?

Agentforce is Salesforce's platform for autonomous artificial intelligence agents. In practice, these agents operate within the Salesforce ecosystem to respond, decide, and act on CRM data across channels such as chat, voice, email, and WhatsApp, always within limits defined by the company.

Agentforce is Salesforce's agentic AI platform that enables you to create and operate autonomous agents within the CRM, capable of planning and executing service, sales and marketing tasks based on company data.

Salesforce treats this capability as digital labor: a workforce that scales without increasing headcount at the same rate. There is also an important technical definition behind the name. For Salesforce, a capability is only considered an agent if it uses the Atlas Reasoning Engine, the reasoning engine that understands the request, decides, and acts. That is what separates a real agent from a fixed-flow chatbot.

Agentforce succeeds Einstein, Salesforce's previous AI layer, and launched in October 2024. Since then, it has evolved rapidly, as the timeline below shows.

How does Agentforce work?

At the heart of Agentforce is the Atlas Reasoning Engine. When it receives a request, it breaks the problem into smaller tasks, builds a plan, executes each step, and evaluates the result before moving forward, in a continuous cycle of reasoning and action. This mechanism gives the agent the autonomy to handle situations that do not follow a predefined script.

An example makes this concrete. A customer asks about an order’s status and requests a delivery-address change. A typical chatbot would answer the first part and stop at the second. An Agentforce agent interprets both intents, looks up the order in the company’s data, checks whether the change is still possible, executes it and confirms it — all in a single conversation.

To act accurately, the agent needs context, and that context comes from Data Cloud, Salesforce's data platform that unifies customer information, history and transactions. Without unified, reliable data, Atlas reasoning loses quality. That is why Data Cloud is often a prerequisite for an Agentforce project, not a detail.

Three pillars support the operation:

• Reasoning: Atlas plans and decides. Since October 2025, it has worked in hybrid mode and can use OpenAI, Anthropic, and Google Gemini models, balancing AI creativity with the predictability of business rules.
• Action: the agent executes through topics and actions connected to Flows, Apex, APIs, and the MuleSoft, making work happen in the company’s real systems.
• Trust: the Trust Layer and guardrails define what the agent can and cannot do, recording every action for auditing.

The result is an agent that retrieves the right data, builds a plan, and completes the task, involving a person only when the case truly requires it.

Agentforce, Agentforce 360 and Agentforce 3: the platform’s evolution

Agentforce has changed substantially in less than two years, and understanding this timeline helps distinguish what is a promise from what is already mature:

• Agentforce (October 2024): the first general release of the agent platform.
• Agentforce 2 (December 2024): A more accurate Atlas and more predictable responses.
• Agentforce 2dx (April 2025): proactive agents and developer tools.
• Agentforce 3 (June 2025): greater interoperability and governance, with Command Center to observe and adjust agents at scale.
• Agentforce 360 (October 2025, at Dreamforce): the platform began to unify apps, data and agents in one place, with configurable Atlas and Agentforce Voice for phone conversations.
• Multi-agent orchestration (June 2026): several specialised agents begin collaborating with each other, coordinated by Atlas, to resolve more complex processes.

The takeaway for decision-makers is clear: the platform has moved beyond the pilot phase and into scaled operations. That changes the kind of project it makes sense to start now.

How much does Agentforce cost?

Agentforce is charged by consumption, not by number of users. There are two main models, and the choice can significantly change the bill at the end of the month.

Model	Billing model	Reference price	Best suited to
Conversations	Per conversation (24-hour window)	US$2 per conversation	complex dialogues, predictable spend per conversation
Flex Credits	Per action executed	US$0.10 per standard action and US$0.15 per voice action (US$500 for every 100,000 credits)	high volume and granular cost control

The difference between the two becomes clear in the usage details. A simple conversation, with three actions, costs around US$0.30 in Flex Credits, well below the US$2 conversation-based model. A long conversation with 25 actions, however, exceeds US$2.50 in Flex and costs more than the fixed price. This is why the choice of model depends on each agent's actual usage profile.

In addition to consumption, there are per-user add-ons starting at US$125 per user per month for teams that use the agent in an assistive capacity. There is also a cost many assessments overlook: Data Cloud, which is needed to feed the agents. To plan your total Salesforce investment, it is also worth reading our guide to how much Salesforce costs.

The amounts above are reference prices in US dollars, according to the Salesforce official price list, and change frequently. Confirm current pricing and contract structure before committing, preferably with support from someone who has already sized similar projects.

What is Agentforce for? Use cases by function

Agentforce is not a tool for only one department. It connects to Salesforce clouds and works wherever there is process and data. The use cases most common:

• Service (Service Cloud): agents can help with triage and resolve defined requests, with permissions, reliable knowledge, and handoff to a human when necessary. To understand the platform that supports this service, see how Salesforce Service Cloud works.
• Sales (Sales Cloud): lead qualification, opportunity updates, next-step recommendations and SDR support, freeing the sales team to negotiate. Quoting has also joined the agents' scope with Agentforce for Revenue, within Revenue Cloud, which succeeds CPQ.
• Marketing: creating and adjusting campaigns, segmentation and responding to interactions at scale.
• Commerce: product recommendations and service within the buying journey.

By industry, there are already mature applications in financial services, retail, healthcare and life sciences, education, and construction. The success pattern is usually the same: start with a high-volume, low-risk case, prove the return, and expand from there.

Is Agentforce the same as a chatbot?

No. A traditional chatbot follows a programmed flow: if the customer says X, it answers Y, and when the conversation goes off-script, it gets stuck or transfers it. Agentforce works from another principle. Because it uses the Atlas Reasoning Engine, it reasons about the request, consults the data and decides the best action, even in situations not anticipated one by one.

In practice, a chatbot responds and an agent resolves. That difference enables end-to-end process automation, not just the first question. To understand the distinction in detail, see our content about how to create a chatbot in Salesforce and the AI models behind these solutions.

Governance, security and LGPD in Agentforce

The more autonomous the agent, the more important governance becomes. In regulated sectors such as finance, healthcare and education, automated decisions must be traceable and auditable. Agentforce provides native controls for this:

• Guardrails and Trust Layer: define the agent's scope and block actions outside it.
• Action logging: every agent step is logged, allowing an audit at any time.
• Data privacy: processing follows LGPD rules, with access control and sensitive-data masking.

It is worth reinforcing a technical point that determines project success: the quality of governance depends on data quality. Unified customer profiles, a clean knowledge base and appropriate processing of personal data are prerequisites for a reliable agent. Atlas reasons well, but it does not make up for poor data.

Agentforce in Brazil: why now

Brazil has become a strategic market for Salesforce agentic AI, and several recent signs show why:

• The Agentforce World Tour São Paulo, in May 2026, was Salesforce’s largest Brazilian event dedicated to agentic AI, built around the theme of humans and agents working side by side.
• Brazilian companies are already expanding their use of agents in sales, service, and data analysis, according to industry news in June 2026.
• Salesforce projects a 67% increase in AI-agent adoption by 2027.

This movement also raises the bar for implementers. A ISG study on the Brazilian Salesforce and Agentforce market assessed 45 providers in 2026, signalling a larger and more competitive partner ecosystem. For companies adopting it, partner selection becomes as important as technology selection.

How to adopt Agentforce in your company

Adopting Agentforce with returns is less about flipping a switch and more about preparing the ground. A path that works in practice:

1. Choose the right use case: start with a high-volume, low-risk process with an easy-to-measure return, such as first-level support.
2. Prepare the data: unify customer data in Data Cloud and clean up the knowledge base. This is the step that most influences the outcome.
3. Define governance: agent scope, guardrails, access control, and LGPD compliance from the start.
4. Measure and expand: track metrics, make adjustments and only then take the agent to other processes.

That is where WeeNow comes in. As an official Salesforce Partner, WeeNow leads the project end to end: scenario assessment, data preparation, implementation, governance definition, and, above all, team adoption — what turns technology into results. If your Salesforce is already in operation, it makes even more sense to build on the existing foundation before moving to agents. Discover the WeeNow Salesforce consulting.

Frequently asked questions about Agentforce

What is Agentforce in Salesforce?

Agentforce is Salesforce’s autonomous AI-agent platform. It enables organizations to create, configure and operate agents that automate tasks and decisions in the CRM across service, sales, marketing and commerce, using company data and respecting administrator-defined limits.

How do Agentforce AI agents work?

The agents use the Atlas Reasoning Engine, which interprets the request, builds a plan, executes each action and evaluates the result. They access context in Data Cloud and act through topics and actions connected to the company's systems, learning to operate within the configured rules and guardrails.

How much does Agentforce cost?

Agentforce is priced by consumption. In the Conversations model, it costs about US$2 per conversation. In the Flex Credits model, it costs about US$0.10 per standard action and US$0.15 per voice action, with credits at US$500 per 100,000. There are also per-user add-ons starting at US$125 per month and the cost of Data Cloud. These figures are indicative and change frequently.

Does Agentforce need Data Cloud?

In practice, yes. Data Cloud unifies the customer data that the agent consults to reason and act. Without this unified, clean foundation, the quality of responses and actions declines. That is why Data Cloud is usually treated as a prerequisite for an Agentforce project.

What is the difference between Agentforce and a chatbot?

A chatbot follows a fixed flow and gets stuck when the conversation goes off script. Agentforce reasons with Atlas, consults data, and decides the best action even in unforeseen situations. The chatbot responds; the agent resolves the process end to end.

Is Agentforce secure and LGPD compliant?

Agentforce includes native governance controls such as guardrails, action logging for audits and masking of sensitive data, with treatment aligned with the LGPD. Effective security also depends on correctly configuring scope, access and data quality during implementation.

How do I start using Agentforce?

The recommended path is to choose a high-volume, low-risk case, prepare the data in Data Cloud, define governance, and measure results before expanding. Working with an official Salesforce Partner accelerates these steps and reduces project risk.

Does WeeNow implement Agentforce?

Yes. WeeNow is an official Salesforce partner and leads Agentforce projects from assessment through adoption, including data preparation, implementation, governance and team training. The focus is on putting agents into production with measurable returns, making use of the Salesforce the company already has.

Conclusion

Agentforce has moved beyond being a trend to become an operational platform. With the Atlas Reasoning Engine, Data Cloud and governance controls, it enables companies to put AI agents to work in service, sales and marketing with scale and traceability. The differentiator is not having the technology, but adopting it for the right use case, with the right data and governance.

WeeNow helps along this path, from assessment to adoption. If your company wants to understand where Agentforce generates returns in your context, speak with a specialist.