We cover the following topics:
Whether you’re exploring your first AI use case or planning a wider rollout, we’ll guide you through what’s possible and where to start.
Get in touch +44 (0) 203 7438014 or email info@infomentum.co.uk
Agentforce - the what, the why and the how
Agentforce is Salesforce's autonomous AI platform that goes beyond basic chatbots. These intelligent agents handle complex, multi-step business processes independently while knowing when human judgment is needed. Powered by advanced reasoning capabilities and natively integrated with Salesforce, they work with real-time data and execute actions across platforms within enterprise-grade security. Your team gains digital employees who work 24/7 on routine tasks, freeing them to focus on strategic decisions and relationships that matter most.
An AI agent is a software-based digital assistant that can interpret data, take actions or make decisions to support human users. AI agents are programmed to assist, but not fully replace human teammates. Common types include:
AI agents can handle almost any question you configure them for, from simple enquiries to complex, multi-step processes. The key is how the agent is set up. You decide what it knows, which data it can access and what actions it’s allowed to take. For example, when someone asks, “What’s the status of my order?”, a well-configured agent won’t guess. It securely pulls the customer’s order history from Salesforce, provides the latest updates, and can even process changes or requests without human intervention. In short, an AI agent is only as capable as the data sources, business rules and instructions you connect to it.
Agentforce agents are built to do more than follow scripts. They can reason and make decisions. Powered by the Atlas Reasoning Engine (the AI “brain” behind Agentforce), these agents can fetch live data, assess the situation, choose the next step, take action and then adjust if needed. That means they’re capable of managing complex tasks and multi-step workflows without constant supervision. Instead of asking for instructions at every step, they work things out, thus reducing doubt, increasing reliability and building real trust in AI-driven support.
Salesforce offers ready-made AI agents through its AgentExchange marketplace, including agents for service, sales, marketing and e-commerce. These can be used straight away and lightly adapted to standard needs. But the real power of Agentforce lies in building your own agents, tailored to your organisation. With Agent Builder, you can create agents that reflect your processes, language and priorities. It is a low-code environment, so you do not need to be a developer. With the right guidance, business teams can build their own agents.
Custom AI agents are built in Agent Builder, Salesforce’s dedicated workspace for designing, managing and training AI agents. You start by defining the agent’s purpose — what it needs to do — and then connect it to your organisation’s data so it can respond with the right context.
From there, you shape how the agent behaves using a set of Salesforce tools. Prompt Builder controls how the agent communicates, while Agentforce Flow defines the steps and decisions it should follow. For more advanced logic, developers can use Apex, Salesforce’s programming language. Einstein AI enables the agent to understand natural language and choose appropriate actions. Together, these tools enable you to build intelligent, secure AI agents aligned with your needs without requiring deep technical expertise.
AI agents are ideal for handling high-volume tasks that require consistency and quick decisions that are based on the available knowledge base. They are not limited to fixed rules. Modern agents can assess context, choose a suitable action and adapt their response within the boundaries you set. They are perfect candidates for tasks such as triaging enquiries, routing reports to the right team, retrieving and updating records, summarising case notes, or sending routine updates. What they do not replace is human judgment in sensitive or complex situations. Instead, they take on the time-consuming work around it, freeing staff to focus on decisions, care and public service where human expertise is essential.
A chatbot is the simplest form of conversational technology. It follows pre-defined scripts to answer basic questions such as opening hours or password resets. It cannot move beyond what it has been programmed to say.
An AI assistant is more capable. It understands natural language, can hold a conversation and can help with tasks such as drafting messages or finding information. On Salesforce, Einstein Copilot is an example. It can recommend actions or create content, but it still asks for your approval before anything happens.
An AI agent goes further. It does not stop at suggestions. Once configured, it acts on your behalf, continually learning and improving. It can update records, route reports, produce summaries or move tasks forward in the background. In practice, it works more like a digital team member, able to make decisions within the rules you set.
Agentforce is the full platform for creating AI agents that can reason, plan and carry out tasks on their own. These agents can retrieve data, make decisions and take multiple actions to complete a process without needing constant human input.
Prompt Builder, on the other hand, is a Salesforce tool for creating specific instructions for AI. It produces single responses or outputs but does not act on its own or manage a full workflow. A helpful way to think about it is this: Agentforce is the digital employee, while Prompt Builder guides and briefs them, defining how they should respond and what they are allowed to do.
Agentforce uses prompts from Prompt Builder as part of its decision-making process. These prompts act like guidance notes, helping the agent decide how to respond or what information to retrieve during a task. While a prompt can handle a single reply or instruction, an Agentforce AI agent combines many prompts with reasoning and live data, allowing it to carry out multi-step tasks independently.
Set up, scale and deployment
Agentforce is built directly into Salesforce, so it works as a natural extension of the platform rather than a separate system. It uses the same data, records and permissions already in place, so there is no need to rebuild processes or move information elsewhere. Because it runs alongside tools such as Service Cloud and Einstein, AI agents can slot into existing workflows and act within the same trusted environment your teams use every day.
Getting started with Agentforce is relatively quick. Basic AI features can be set up in just a few days. A focused rollout, with one or two use cases and a small group of users, typically takes around four to six weeks. It includes planning, configuration, testing, training and go-live. Larger, multi-cloud deployments may take two to three months.
Success depends on starting with clearly defined tasks and gradually expanding as the agent proves its value. The core team usually includes a Salesforce administrator to manage access and flows, a developer for advanced logic, and a consultant or delivery partner to guide strategy and governance.
Turning on Agentforce is simple. It can be enabled with a few configuration steps inside Salesforce. The real work lies in defining what you want the agent to do. You do not need deep technical skills, but you do need a clear process, reliable data and a well-framed use case. Once those foundations are in place, Agentforce lets you build, test and refine agents quickly, starting small and scaling as confidence grows. In short, it is easy to begin and powerful when designed with purpose.
You don’t need to be a developer to get started. Agent Builder allows you to describe what you need in plain language, and Salesforce will generate the agent’s structure, including suggested actions, flows and permissions. However, for more advanced automations or custom data logic, familiarity with Flow and basic Apex can be helpful.
Equally important is understanding data governance and access control. Knowing which data the agent can use, who grants permissions and how information is protected ensures that the agent operates safely and responsibly within your organisation’s rules.
There are no additional infrastructure requirements beyond Salesforce. Agentforce is fully cloud-native and runs within your existing Salesforce environment, so there is no need for extra servers, hosting or third-party tools. Everything operates securely within the platform, using your current data, permissions and controls. You will only need the appropriate Salesforce licences with Agentforce features enabled.
Yes. Agentforce works in modern web browsers, including on mobile devices, so that it can be accessed anywhere. Salesforce also provides a dedicated mobile app that lets teams use AI agents in the field or while working remotely. Core capabilities, such as viewing updates, triggering actions, or receiving summaries, remain fully supported, ensuring no loss of functionality outside the desktop environment.
Agentforce is designed for enterprise scale. It can support large, distributed, or multi-region organisations by routing tasks based on location, applying the correct permissions to each team, while maintaining consistent rules across every region. Control and oversight remain centralised, while delivery can be tailored locally. This approach allows AI agents to operate securely and reliably, no matter the size or structure of your organisation.
Strictly speaking, no, you do not need Data Cloud to start using Agentforce. AI agents can work with the data already held in your Salesforce org. However, Agentforce performs better with Data Cloud, as it provides real-time, unified records that give agents a richer context.
You do not need a full implementation, even the basic tier of Data Cloud enables features such as identity matching, cross-system insights and real-time personalisation. This allows agents to make better decisions and deliver more intelligent outcomes.
Not at all. Agentforce is just as valuable for smaller and mid-sized teams, especially those already using Salesforce. It allows organisations to improve efficiency and service without increasing headcount by deploying AI agents to handle routine work. You can start with a single use case and scale as value is proven. Licensing and configuration can also be tailored to suit different organisation sizes, so you only adopt what you need.
AI agents in Agentforce are highly customisable. You are not limited to preset templates. Using Salesforce tools such as Flow, Prompt Builder and Apex, you can shape how an agent behaves, what data it uses and how it communicates. If required, agents can also connect to external systems through APIs, allowing them to work beyond Salesforce.
Agents can grow over time. You can assign them new actions, expand their responsibilities and refine their decision-making as your needs evolve. This makes them adaptable, not static — capable of becoming a true extension of your team.
Yes, Agentforce agents can improve over time, but not through uncontrolled self-learning. Today, they refine their performance through human guidance. You review how they handle tasks, adjust their prompts, logic or flows and redeploy. This ensures they become more accurate and capable with each iteration, while remaining fully under your control.
Salesforce is also developing features such as feedback loops and adaptive reasoning. These will allow agents to learn from real outcomes and make more informed decisions as they go, while still operating within the guardrails you set. The direction of travel is clear: Agentforce agents will become increasingly self-improving, but always with safety, security and governance at the core.
The most common challenges arrive not from the technology itself, but from preparation and ongoing oversight. Poor data quality is often the first hurdle. Incomplete or inconsistent records can confuse an AI agent, so it is worth investing in clean, reliable data before launch. Another risk is trying to do too much too soon. Starting with a focused use case and expanding gradually helps teams build confidence without pressure. Finally, agents still require human supervision.
Setting clear performance measures and monitoring their outputs ensures they continue to deliver value and behave as expected. With the right foundations and governance, Agentforce can scale safely and effectively.
Agentforce should be treated much like a new member of staff. It needs clear boundaries, supervision and regular review. Begin by defining what the agent is allowed to do, when it should escalate to a person and which actions require approval. Salesforce provides full oversight through audit logs and the Einstein Trust Layer, so every decision and action is traceable and explainable.
Human-in-the-loop controls can also be built in, allowing you to add checkpoints, require sign-offs for critical steps or restrict access to sensitive data. With the right governance in place, your organisation stays firmly in control while the agent handles routine work with confidence.
Salesforce has made it clear that Agentforce is central to its future AI strategy, and the platform is evolving quickly.
Agentforce is rapidly evolving from a supportive co-pilot to a fully capable operator designed to take on real work.
Security, controls and governance
Yes. Agentforce is protected by Salesforce’s Einstein Trust Layer, which provides encryption, access controls and full audit trails. When external AI models are used, data is routed securely, and organisations can choose to keep prompts within regions such as the UK or EU to meet GDPR and data-residency needs. In short, Agentforce follows the same security and governance standards trusted across the Salesforce platform.
Agentforce relies on Salesforce’s security framework, including Salesforce Shield, to protect data at every step. It uses encryption, role-based access controls and audit logs to ensure only authorised users can access sensitive information. All activity is monitored to detect unusual behaviour early, reducing the risk of leaks or insider misuse. Importantly, AI agents cannot bypass existing permissions as they follow the same access rules as your users, keeping full control in your hands.
Agentforce is designed so that AI agents only access the data they are allowed to use. Through the Einstein Trust Layer, you can control permissions, restrict sensitive fields and apply data masking to hide confidential details. Every action is logged and fully auditable, so you can see exactly how information is handled. With these safeguards in place, automation operates safely without compromising privacy or control.
If an agent produces an incorrect output, it can be reviewed, corrected and refined, just as you would with a staff member following guidance. You remain in full control, with the ability to pause, adjust or retrain the agent at any time. AI assists, but it does not replace human accountability.
Agentforce includes a central command centre that gives full visibility across all AI agents. From here, you can monitor performance, review activity logs and track how each agent is being used. Permission controls and approval workflows ensure departments follow consistent rules, while dashboards and audit reports maintain transparency and accountability at scale. In essence, you retain oversight while allowing agents to operate safely across the organisation.
Agentforce builds on Salesforce’s enterprise-grade protections. It encrypts data, tracks consent, keeps audit logs, and can redact sensitive information. When using external AI models like OpenAI, geo-routing keeps data in the right region (for example, UK data stays in the UK). This ensures you meet privacy regulations without needing extra infrastructure.
Yes. Organisations remain fully in control of how AI agents operate. You can set boundaries around what an agent is allowed to do, decide when human approval is required and monitor how it behaves over time. Every decision is logged through the Einstein Trust Layer, so actions can be reviewed, adjusted or reversed if needed. Admins can refine prompts, update rules or limit permissions, ensuring AI agents remain transparent, accountable and compliant at all times.
Insights, reporting and performance
Agentforce provides dashboards and reports that let you track how well your AI agents are working. You can measure performance using metrics such as resolution
time, accuracy, user satisfaction and adoption across teams. By reviewing these insights, you can fine-tune agents' behaviour and expand their responsibilities with confidence.
Tools like Utterance Analysis in Agent Builder show users' interactions with agents, including the questions they ask, agents' responses and where improvements are needed. With these insights, you can refine prompts, adjust logic and improve reliability over time.
Yes. AI agents operate under human control. Suggestions can always be accepted, ignored or corrected by your team. If an agent is unsure, it can be designed to hand the task back to a person for review. Feedback on incorrect suggestions can then be used to refine prompts and improve performance over time. In every case, humans remain the final decision-makers for accuracy and accountability.
Yes. Agentforce allows you to track the performance of each AI agent using Salesforce’s built-in dashboards and reports. You can view how often an agent is used, how accurate its responses are and what outcomes it delivers, such as tasks completed or cases resolved.
For deeper insight, tools like the Agent Performance Dashboard and usage logs let managers and admins review trends over time and identify areas for improvement. For example, you can compare two agents to see which one handles more requests or delivers faster resolutions. This level of visibility ensures each agent remains effective, accountable and aligned with your goals.
Agentforce provides dashboards and reports that show how AI agents are improving day-to-day work. You can measure reductions in handling time, quicker response rates, and the number of repetitive tasks taken off your team's plate. These insights help you quantify the value of automation and highlight where further optimisation may be possible. Over time, you can clearly see how agents are freeing staff to focus on higher-value work.
AI agents can only perform as well as the data they rely on. When information is accurate, consistent and up to date, agents make better decisions, deliver clearer responses and handle tasks with confidence. Poor or incomplete data, however, can lead to hesitation, incorrect routing or unclear answers.
By monitoring performance data, such as errors, hand-offs, or repeated queries, you can identify where data needs improvement. This creates a guided learning cycle: the cleaner the data, the smarter and more reliable the agent becomes. In short, improving data quality directly improves AI performance.
Cost and business value
Agentforce is offered as an add-on to Salesforce, and pricing varies based on how many AI agents you deploy, the features you enable and your usage. There is no single flat fee, as each organisation sets up Agentforce differently.
Most teams start small with a pilot to control costs and prove value before scaling. We can help you choose the right licences, estimate usage and build a cost model that aligns with clear outcomes, not guesswork.
Salesforce uses a usage-based Flex Credit model for Agentforce, similar to a mobile data plan. Each time an AI agent performs an action or handles a conversation, a small number of credits is used. You only pay for what you consume, which lets you scale gradually and keep costs aligned with actual usage.
Agentforce usage is measured through Actions and Conversations, which represent real work carried out by the AI — such as retrieving information, updating records or completing a task. Each of these consumes Flex Credits. This usage-based model means you only pay for what the agent actually does, aligning cost directly with the value delivered.
Yes. Agentforce impact can be measured using Salesforce reports and dashboards that track key metrics such as resolution time, adoption rates, productivity gains and the accuracy of AI-driven actions. Over time, you can compare results before and after deployment to see how agents are reducing manual work, speeding up responses and improving service outcomes. These insights help validate return on investment and guide future optimisation.
No. You don’t need a full AI strategy to begin. With Agentforce, you can start small, for example, by launching a single agent to assist with replies or summaries, and build from there as your team gains confidence. Early pilots help prove value quickly and naturally shape your AI direction. Our Agentforce Jumpstart programme is designed to support exactly this approach, helping you launch your first agent while laying the foundations for a wider strategy over time.
Yes. Agentforce helps reduce pressure by removing repetitive, time-consuming tasks, allowing staff to focus on work that requires judgment, empathy and expertise. It also provides real-time guidance, boosting confidence and reducing stress in high-demand environments.
Over time, this not only improves morale and well-being but can also be measured through operational gains such as faster case resolution, reduced handling time and fewer backlogs. In short, supporting your people and improving performance go hand in hand.
Agentforce delivers value by increasing productivity, reducing handling time and improving service without adding headcount. Early adopters report faster routing, lower cost-to-serve and higher employee satisfaction from removing repetitive work. The return comes not from replacing people, but from giving teams leverage, automating what slows them down while keeping judgment and empathy with humans.
The strongest ROI comes from starting with clear, measurable use cases. Small wins compound over time.
Integration and ecosystem
Agentforce is built directly into the Salesforce platform, using the same data, permissions, and interface your teams already use. It operates across Service Cloud, Experience Cloud and custom Salesforce apps without needing separate tools.
For external or legacy systems, Agentforce leverages MuleSoft, Salesforce’s integration platform, to create secure API-based connections. This allows AI agents to access trusted data from other systems and act within wider business processes while maintaining governance, security and control within Salesforce.
Agentforce can connect to other platforms, such as Microsoft Teams, telephony systems and external CRMs via MuleSoft, Salesforce’s integrated API platform. This allows AI agents to fetch data, trigger actions and surface suggestions directly within the tools your teams already use. By using MuleSoft, Agentforce can securely interact with internal and legacy systems while remaining centrally governed within Salesforce.
Agentforce supports multilingual use by allowing AI prompts to be configured in different languages and by integrating with trusted translation services. This enables agents to communicate clearly with users across regions while maintaining consistent service standards. Global teams can run localised workflows, ensuring the same level of service, whether delivered in English, Spanish or any other supported language.
Einstein AI uses the data already stored in Salesforce to deliver insights, predictions and recommended actions. It analyses past cases, knowledge articles and customer interactions to generate helpful responses or suggestions in real time. This helps teams work faster, make informed decisions and respond with greater accuracy.
Agentforce and Einstein are designed to work together. Einstein provides the insights, predictions and natural language understanding, while Agentforce uses that intelligence to act. This means AI agents can move beyond simple responses and make informed, data-driven decisions like updating records, routing tasks or guiding users across the Salesforce platform. Together, they help teams work faster, with greater accuracy and confidence.
Team enablement and usability
Yes. Agentforce is built into the familiar Salesforce interface, so there’s no need to learn a new system. Most features are guided and intuitive, making it easy for teams to adopt. Users interact with AI agents just as they would with any other Salesforce feature through buttons, prompts and suggestions. It keeps training simple and confidence high.
Users don’t need any technical skills to use Agentforce. AI suggestions and automations appear directly in their Salesforce workspace, ready to apply with a single click. Everything is built to feel familiar, so teams can benefit immediately without learning new tools or commands.
Very little. Agentforce sits inside the familiar Salesforce interface, so most users can start working with it right away. Admins can support adoption with short walkthroughs, quick tips or simple guidance articles. Salesforce also includes in-app prompts and guidance to help new users learn by doing, keeping onboarding light and intuitive.
Adoption grows best when you start small. Launch with a pilot group, demonstrate clear time savings and share quick wins. Use dashboards to show real impact and let users see how AI reduces effort, not control. Highlight success stories, encourage feedback and expand gradually as confidence builds, turning early advocates into champions across the organisation.
Yes. Agentforce can be configured to support each team in the way that works best for them. Prompts, workflows and layouts can be tailored by department, region or role. A front-line agent might receive suggested replies or case summaries, while a manager could see performance insights and trends. This flexibility ensures everyone benefits from AI that fits their responsibilities and priorities.
Agentforce doesn’t require teams to be restructured. Instead, it helps rebalance workloads by taking on repetitive tasks and freeing staff to focus on more complex or high-value work. Over time, this may shift responsibilities, but it does not mean reducing roles. Only in large-scale transformations might organisational structure need to adapt, and even then, it is guided by strategy, not by the AI itself.
