Time and productivity set the value of businesses these days. Many of them simply struggle with operational friction or, at worse, costly disruptions.
Today, office workers dedicate nearly 13 hours per week to low-impact activities. Commonly, these often include mundane, repetitive tasks. For sales reps who spend almost 70% of their time on non-selling activities like data entry, personalized emails to buyers, and research, this often creates a productivity lag and diverts focus from deal closure.
What if we say that even a single-LLM AI Agent can greatly change the equation.
Earlier, we described the transformative nature of the Agentic AI technology. So, today we’ll talk about Agentforce, which is Salesforce’s native agentic AI framework and an example of rapid integration of this novel tech into one of the largest CRM systems.
Steep rise of AI agents in workflows automation
After the buzz over GenAI started fading, AI agents swiftly brought back the spotlight on AI. Salesforce picked the perfect timing to roll out Agentforce, claiming its spot in the CRM and autonomous digital workers niche.
Since its launch in September 2024, Agentforce has enlisted over 8,000 customers (including such giant brands like PepsiCo, Heathrow, Indeed, Transcend, etc.) and announced a 3.0 upgrade. The solution goes by leaps and bounds, already highlighting successful automation cases, e.g., Agentforce independently handles 1M support requests coming to the Salesforce Help website, as well as 70% of chat interactions for 1-800Accountant.
Let’s get straight to the “backstage” of Salesforce Agentforce, where we’ll explain what it is and the reasons why Agentforce AI agents are getting rampant across businesses.
What is Agentforce by Salesforce?
Agentforce is an agentic AI platform that is natively integrated into the Salesforce ecosystem. Using it, you can either deploy pre-built AI agents (e.g., Service Agent, SDR Agent, etc.) or customize an AI agent to your processes with low-code/no-code tools by Salesforce (Prompt Builder, Copilot Builder, Flow Builder, Data Cloud and Data Graph Configurator) or with Apex programming language for complex solutions or integrations. To create a domain-specific AI Agent that would cover particular use cases and business needs, you may need a partner’s support with extensive expertise in business analysis, product design, end-to-end software engineering, etc. Sombra, an end-to-end IT services provider, can assist with any request, ensuring consistent delivery and measurable impact.
Who should consider using Agentforce?
Businesses that already rely on Salesforce as their primary CRM system should be the early adopters of Agentforce. Plus, it would be a natural fit for those who seek automation of routine and time-consuming tasks in sales, marketing, commerce, customer service, and IT operations, and consider moving their workflows to Salesforce.
Moreover, we recommend considering Agentforce as it can already handle multiple functions within and outside CRM.  From marketing campaigns, personalization, lead conversion, sales process and optimization, customer requests processing and resolution, to document management, dashboard generation, and more.
What makes an Agentforce AI agent a real digital worker?
While the potential of Agentforce Salesforce is still explored and enhanced, its capabilities are leading us to the era of digital employees:
- They can operate autonomously (like self-service bots) or rely on guided control.
- AI agents are goal- and action-driven, keeping focus on a specific business domain and industry.
- They are designed to catch on complex contexts, function as independent performers of specific tasks, or as partners of human agents.
- Adapt to changing objectives and act upon a given situation as a human would do with the ability to think, plan, interpret information, and act relying on specific guardrails (rules, boundaries).
Is Agentforce ready to meet real-world business demands?
Matching the growing customer demand on time makes companies either expand their human personnel and pay more, or look for alternative options. Organizations that are already using AI agents or plan to do so mainly highlight customer service and support (57%), sales and marketing, IT, and cybersecurity as the most common business functions that need automation of mundane jobs. That is where a Salesforce agent can step in to resolve the persistent staff shortage and provide timely customer feedback.
Source: PwC
But how do we know that Agentforce is ready to match that pressing need in operational automation across industries? Let’s look at a few facts on the behalf of Salesforce’s agentic AI technology readiness for real-world use.
Agentforce relies on well-matured LLMs and LAMs
When LLM models like GPT and BERT barged in on the AI arena in 2018, they soon triggered the development of proprietary LLM models by Anthropic (Claude 3), Mistral (Mistral Large), Cohere (Command R+), Amazon (Titan), and Salesforce (CodeGen), etc. It was CodeGen 2.5 that became the first in-house model for code completion and generation in Agentforce. Then came the xLAM series (not just language but action models), designed for function-calling, handling multi-turn tasks, supporting AI agents’ reasoning, planning, and taking actions without or with minimum human intervention.
Agentforce leverages retrieval-augmented generation (RAG)
With the RAG framework that allows anchoring AI responses in proprietary data, GenAI, as well as AI agents, instead of guessing, can operate with facts, maintain more contextual relevance, and consistency. Moreover, RAG is bonded with Salesforce’s LLMs using Einstein GPT (a proprietary GenAI tool) and Data Cloud, which also advocates for data trustworthiness and accuracy of responses.
Overall technological readiness of a Salesforce system
At some point, Salesforce evolved from a simple Q&A system with standard chatbots and then AI assistants who co-worked and co-created with a human to a true powerhouse of connected core solutions (Customer 360, Data Cloud, Apex, Prompt Builder, etc.). They now support multi-turn, goal-driven AI agents capable of making decisions independently and acting accordingly. With this well-developed and continuously refined ecosystem of tools, Agentforce had a good head start in entering a competitive market. Find out more about Agentforce’s technological readiness below!
Overview of Agentforce Capabilities and Integrations
The AI agent capabilities often stem from the way Agentforce reasons, triggers, and natively integrates with the Salesforce ecosystem.
Source:
The Atlas Reasoning Engine supports Agentforce’s ability to think like a human and make decisions based on user input. It brings together LLMs, LAMs, the RAG module, Rest APIs, and data connectors to access data repositories. Think of Atlas as a brain AI agents use to understand high-level goals, plan and prioritize tasks, next steps, select specific tools, and apply an appropriate sequence of actions.
Customer 360: As a core data platform, it provides a unified customer profile, merging data from Sales, Marketing, Commerce, and varied external databases. Agentforce can sync with Customer 360 through native integration via Data Cloud, using xLAMs’ ability to retrieve relevant data directly from Customer 360, or through proprietary APIs or MuleSoft connectors for real-time data. This way Agentforce accesses the intel on your customer (CRM records, transactional history, contacts, their engagement, purchase journeys, support history, etc.) on demand, depending on its task and access restrictions.
Data Cloud: It can also integrate with different Salesforce clouds to add and update records, trigger sales or service workflows. But above all, it can link to a real-time Data Cloud, which is an advanced data storage and governance tool, merging all types of data (structured or unstructured; static or streaming) from varied sources (internal or third-party lakes, apps, DWHs). Relying both on Customer 360 and Data Cloud, Agentforce builds up its contextual memory (as well as customer and record history, intermediate variables, and past inputs) and dynamically adjusts responses depending on user behavior and records. By the way, Agentforce utilizes RAG and semantic search in Data Cloud to capture meaningful data from both internal and external sources without copying or moving data.
Native MCP support: This one is the hottest feature that the Agentforce 3 iteration brought later this June. The MCP (Model Context Protocol) is an open standard that, in Agentforce, works as a single interface for the agents’ interoperability and integration across different in-house and third-party systems. Using the native Salesforce MCP client, AI agents can access data, tools, or workflows without heavy customizations needed while ensuring compliance and security of all AI agent interactions.
Source: Salesforce
Flow (no-code/low-code) and Apex (Salesforce’s programming language resembling Java for building custom logic, triggers, classes, etc.): Agentforce can directly reach Flow or Apex to initiate automations, orchestrate tasks, and perform actions autonomously. Among such can be record creation and management, sending emails, etc.
Prompt Builder: The development tool for creating prompt templates and deploying LLM prompts with business-specific context. Agentforce can interpret business logic with the help of such prompts.
Einstein Trust Layer: Every interaction an AI agent triggers goes through this layer to ensure access to data is safe and verified while responses are relevant, trustworthy, and compliant.
How Agentforce integrates with internal and external systems
Agentforce goes beyond the limits of Salesforce CRM functionality, since it can connect with Tableau, Slack (both are owned by Salesforce), Google Workspace, Microsoft 365 (Outlook, Word, Excel, Teams, etc.) through native connectors or direct APIs. Though Agentforce is tightly coupled with Salesforce, leveraging its data, architecture, and logic, it can also connect with external systems like HubSpot, SAP, ServiceNow, Jira, PowerBI, DocuSign, etc. with the help of Salesforce’s AppExchange (cloud marketplace) or MuleSoft platform that supports deep two-way integration.
In edge cases (when the connector is costly or unavailable), you can always implement custom integration via Apex, which is also effectively used by Agentforce.
In the case of HubSpot or SAP use, Agentforce cannot operate without Salesforce, as it’s not a standalone framework. Hence, Salesforce will be a primary CRM platform, while HubSpot, for example, would serve as an external data source or an additional campaign system.
Connecting all the dots within a single ecosystem may be intricate. Which is why existing clients highly value Sombra’s services in aligning and integrating external systems with Salesforce.
Automation use cases
Now, let’s consider your options regarding automation use cases and why you should start working with Agentforce even if you lack expertise. It may be only one for the start. When choosing, please mind that most automation use cases are shaped by the type of agent, its role, purpose, goals, and workflows within the system. In the Salesforce environment, you can find pre-built templates of the following AI agent types:
1. Service agent
Its primary function is to assist customers and support customer service and contact center teams by automating routine tasks.
Use Cases:
- Triaging and routing of customer inquiries. This agent can automatically triage and classify cases depending on the topic, sentiment, urgency, etc. It resolves common issues and reroutes complex ones to service reps.
- Updating records and creating follow-up tasks for customer service teams
- Analyzing and updating fragmented data to keep its high quality and accessibility.
2. SDR agent
A sales development rep (SDR) is ideal for leads processing and generation of personalized content.
Use cases:
- Qualification of leads (based on set rules, lead behavior, and history of interaction if any), enhancing lead data in CRM
- Classifying high/low-priority leads
- Scheduling meetings, outbound outreach
- Creating email drafts and responses tailored to the context of a record
- Summarizing and structuring emails, contracts, survey responses, and other documents
- Lead activity logging in CRM
- Analyzing incomplete data and filling in the missing data
Additionally, service agents and SDR agents, as well as custom domain-specific agents, can support and facilitate the approval process. They can apply business rules to suggest possible decisions or identify blockers to pending approvals.
3. Sales coach agent
For sales reps, this agent would become a source of insights and recommendations on how to close the deal with the best outcome.
Use cases:
- Role-playing with a sales rep to improve pitching, learn how to negotiate productively before meeting a customer
- Scoring opportunities
- Analysis of deal health
- Identifying halted deals and ways to reactivate them
4. Merchant agent
The template of this agent was designed for retail and e-commerce to help transform and optimize their operations.
Use cases:
- Notifies about inventory level and automatically refills them
- Continuously analyzes trends and competitors to optimize prices
- Creates descriptions of products or services
- Enables issue triaging and further communication with a customer
- Personalizes promotional campaigns based on customer history and preferences
All the AI agents also provide contextual assistance on demand. They do not act until a user asks for it, e.g., the agent can make a list of overdue tasks or suggest best options to negotiate with a client.
Custom AI Agent Use Cases
Lack of expertise in agentic AI development makes many businesses set aside the idea of AI-powered operational automation. It is especially common when templates of AI Agents cannot address business-specific challenges. Therefore, here, in Sombra, we focus on building custom Agentforce-based AI agents that cover common or unique industry-agnostic use cases. For the past year, Sombra helped its clients to onboard Agentforce AI agents through customized business processes:
- Optimizing lead conversion workflows with a custom integration for the client’s Salesforce environment and all connected systems. The AI agent helps with accurate assessment of lead intent signals, real-time data analysis, personalized communication drafts, and further action recommendations. The AI agent was aligned with complex marketing and sales operations of our client (also in-house marketing automation tools, external data sources, etc). This allowed optimizing lead engagement workflows, ensured instant access to critical data, and lead engagement metrics. 
- Customizing Agentforce for multi-step order management processes. The sales AI agent was directly connected to Salesforce and the client’s inventory, contract management, pricing, and all sales interactions using MuleSoft. The agent was specifically customized to automate quote generation, approval flows, and multi-step order creation. This streamlined operations, enhanced inventory visibility, increased product availability, and improved pricing accuracy.
- Building a bespoke self-service chatbot using Agentforce. Our client sought to decrease the volume of support calls from the external users. Leveraging Agentforce capabilities within Salesforce, we integrated an AI agent customized to the client’s support system and order management processes. It provided the external users with access to targeted solutions through users’ knowledge base articles and FAQs. The agent accelerated customer issue resolution and increased user satisfaction.
The ratio of Agentforce agentic AI benefits and challenges
Sure, like any new technology Agentforce has many promises, yet there are also some limitations and risks we want you to consider. We tried to visualize those that make companies halt before adopting an AI agent.
Best practices for your first Agentforce implementation
To truly benefit from Agentic AI, where should you begin? At Sombra, we prioritize our clients’ needs, focusing on practical outcomes and transparency instead of exaggerated features. Here are some of the best practices for Agentforce implementation that we recommend following if it’s your first AI agent rollout:
1. Go with a single use case
Reasonable enough is to begin with one automation use case leveraging a specific type of AI agent like Service Agent or SDR Agent. It’s best to pick a process with clearly defined goals, well-structured data, minimal compliance risks, and measurable ROI.
For instance, take case triaging, email summarization, or lead priority classification. Such task automation would bring faster productivity gains with well-organized data and clear-cut rules. The agent would operate in a scoped logic and get access to only authorized sensitive data thanks to the Einstein Trust Layer.
2. Prepare your CRM data for the AI agent
Ensure your CRM records are accurate by checking for duplicates, missing or outdated information, and inconsistent formats. Also, ensure your data can be used in real-time by the Agent, who continuously relies on its validity. Properly tag knowledge articles that would help Agentforce to easily retrieve them through RAG. Map out specific data sources your AI agent will need to access without compromising security. Additionally, define data ownership and stewardship rules by designating someone responsible for data maintenance, validation, and updates.
3. Determine AI agent success metrics
We recommend tracking all possible metrics relevant for the selected use case and flagging issues that affect the AI agent performance and reliability. Some of them could be:
- Response time,
- Saved time per each task,
- Frequency of agent intervention,
- Task completion time without human inclusion,
- Escalation and error rate,
- Number of all / reduced follow-ups, etc.
4. Start with PoC and MVP of a custom AI agent
If you select a pre-built AI agent template with a proven use case, it will be easier to deploy. Yet, often businesses come for a specific use case, tailored to their current problems and business domain. In this case, we suggest building a custom AI agent and starting with PoC. This way, you can evaluate the feasibility of an AI agent and its success with the following metrics:
- % of timely and correct responses and decisions in regard to prompts,
- Number of successfully completed tasks,
- % of retrieved fields without the need to transform data,
- Number of prompts with the required data normalization,
- Response time needed to access Salesforce data and internal tools,
- % of successfully fulfilled calls to APIs like MuleSoft or other external systems, etc.
If you are moving forward with an MVP, select a specific workflow or task and utilize a narrow, role-specific context. Test whether your agent triggers necessary actions through Apex or Flow. Only after the MVP verifies the reliability of the Agent’s actions, its reasoning, and its ability to cope with a single scenario can you scale across other use cases.
5. Create and test prompts to save on debugging
The quality of the prompts you write will significantly influence how an AI agent thinks, makes decisions, and the output it generates. To improve this, use Salesforce Prompt Builder to incorporate business terms commonly used by your team. Define the structure of the responses and provide fallback instructions for scenarios where necessary information is missing or errors occur. This approach will help prevent the AI agent from providing misleading information or making incorrect assumptions. Additionally, limit the agent’s scope by using guardrails, such as constraints and rules, to ensure it operates safely and accurately.
6. Exploit Salesforce governance best practices
Like many other teams that develop Agentforce, we also rely on Salesforce’s best practices to ensure security and compliance with key regulations for AI agents. Some effective practices include using named credentials and incorporating event monitoring to continuously track agent activities and behaviors. For multi-step workflows, it is essential to utilize flow orchestration and define escalation paths. In terms of security controls, implement role-based access control (RBAC) and ensure that you log the outputs of an agent.
Conclusion
Agentforce has the potential to penetrate every Salesforce workflow, demonstrating the shift of the platform towards an automation-first culture. In this post, we aim to demonstrate Agentforce’s readiness for implementation in various business scenarios. AI agents are more than just assistants; they are autonomous, fully aware of their context, and driven to take action.
Let us assist you with integrating AI agents into your Salesforce workflows, tailored to meet your specific business needs and use cases. Drop us a line and we’ll be glad to contact you soon!
