Artificial intelligence
CRM leader Salesforce has integrated generative AI and autonomous agents at the heart of its platform with Einstein GPT (presented in 2023) and Agentforce (unveiled in 2024). Einstein GPT is the first CRM-native content generation engine, capable of producing emails, marketing messages or case summaries based on real-time CRM data. Agentforce, meanwhile, goes one step further: it’s a platform of agents that reason, plan and execute tasks directly in Salesforce, while respecting business rules and strict governance.
Einstein GPT: generates contextualized content (emails, call scripts, support summaries) using customer data and CRM history. The tool enables large-scale personalization and adapted tonality.
Agentforce: offers autonomous agents that understand the context of an opportunity or ticket, make decisions (e.g. apply a credit or offer a product) and execute actions in Salesforce. Agents operate around the clock and can follow complex business rules.
Service Cloud: integration with the support module enables agents to resolve cases, recommend knowledge articles or redirect requests, while maintaining a complete audit of actions.
Governance and security: Salesforce provides granular permissions, activity logging and the ability to control access to data. Agents are supervised and audited, ensuring regulatory compliance.
Beyond content generation and basic actions, Salesforce has introduced several components to facilitate agent development and management. Agentforce Builder offers a unified authoring space in which teams write instructions, define variables and test scenarios. Developers can choose between a low-code approach (wizard and visual interface) and a pro-code approach (scripts and custom functions) to create agents adapted to complex business rules. Agent Script, a specific language, combines deterministic steps (API calls, SQL queries) with the generative logic of Einstein GPT, offering precise control while benefiting from the creativity of LLMs. Agentforce Voice extends these capabilities to telephone and voice channels, enabling agents to guide customers or employees by voice, transcribe conversations and apply appropriate actions.
To feed these agents, Salesforce has introduced anintelligent context layer that extracts and structures information from notes, documents and transcripts, then makes it accessible to agents via APIs. Teams can design transformation pipelines that ingest PDF files, emails or audio recordings, extract key entities and enrich the Data Cloud. This contextualization enables Einstein GPT agents to produce more relevant responses and generate meeting summaries or recommendations for action.
A system of safeguards ensures that agents comply with internal policies. It combines toxicity and sensitive speech filtering with business rules (e.g., not to offer discounts above a certain threshold). Administrators can monitor interactions, review proposed actions and intervene in case of doubt. This architecture prioritizes security and trust, two recurring themes in Salesforce’s strategy.
Salesforce customers are gradually adopting these new capabilities. In sales, agents recommend actions to close an opportunity, generate presentations and send personalized emails. Thanks to Einstein GPT, messages are written in the right style and tone, using CRM information such as purchase history or recent activity. In marketing, teams design targeted campaigns, create varied content (emails, social posts, ads) and adjust messages based on customer reactions. Generative models enable large-scale hyper-personalization by adapting tone, vocabulary and offers to the recipient’s preferences.
In customer service, Einstein GPT and Agentforce summarize exchanges, suggest next steps and create follow-up tickets. Automatic generation of summaries and responses saves considerable time, allowing human agents to concentrate on complex cases. Customers can also receive proactive responses, as the models detect signals of dissatisfaction or intent and send offers or alerts.
Salesforce puts forward the idea that AI agents reinforce human teams rather than replace them. The tools provide suggestions, synthesis and analysis, but the final decision rests with the human. The platform insists on data quality: to guarantee relevant results, it is essential to have complete and up-to-date CRM records and to apply strict governance rules.
In the future, Salesforce plans to deepen integration between Agentforce and Slack, as well as interoperability with other ecosystems via Mulesoft. Growing competition in AI agents (Microsoft Copilot, Google Vertex AI) will prompt Salesforce to step up its investments in proprietary models and workflow optimization. Companies will need to assess their data and governance capabilities to take full advantage of Salesforce’s CRM innovations.
To maximize agent efficiency, Salesforce has also introduced analytics and monitoring tools to track agent behavior and measure indicators such as conversion rate or mean time to resolution. These dashboards enable administrators to identify bottlenecks and adjust rules or generated content. For example, a sales rep can visualize how an agent modified an opportunity and what impact the recommendation had on sales, creating a continuous improvement loop.
Salesforce is also investing in Slack and Mulesoft integration. The Slack connector enables agents to participate in team conversations, suggest responses in real time, trigger approval flows and summarize discussions. Thanks to Mulesoft, agents can interact with third-party systems (ERP, finance, supply chain) via APIs, paving the way for inter-organizational workflows. These innovations demonstrate Salesforce’s intention to make Agentforce a central hub for all digital business interactions, unifying collaboration, actions and analysis under a single interface.
Native integration: agents are directly integrated with CRM and sales, marketing and service modules. No need for middleware: data is immediately available and actionable.
Fine-tuned personalization: Einstein GPT uses real-time customer data to generate hyper-personalized content, while Agentforce applies business rules to make appropriate decisions.
Traceability and governance: every agent action is recorded and can be audited. Administrators can restrict access, define rules and verify results.
Data preparation: the quality and structure of CRM data determine the success of agents. Companies need to cleanse, enrich and organize their data to avoid incorrect answers.
Dependence on the Salesforce ecosystem: Agentforce is designed to work exclusively within Salesforce. Companies not using this CRM will need to adopt the full environment to benefit from it.
Subscription costs: adding Einstein GPT and Agentforce can lead to significant additional costs, especially for licenses and model usage.
| Solution | Strengths | Weaknesses |
|---|---|---|
| OpenAI Frontier | Integrates various agents and data sources, advanced governance | Not CRM-specific, opaque pricing |
| Microsoft Dynamics 365 Copilot | Integrated AI capabilities for Microsoft CRM, generating insights and automations | Less emphasis on autonomous agents and multi-step actions than Agentforce |
| Zendesk AI | Automated support system with intelligent responses and triage | Less integrated with a complete CRM for sales and marketing |
| HubSpot ChatSpot | Conversational assistant for HubSpot CRM, generates e-mails and reports | Less powerful for planning and execution of actions by agents |
What is Einstein GPT?
– A Salesforce-integrated content generation engine that leverages CRM data to create emails, marketing messages and case summaries.
What is Agentforce?
– A Salesforce AI agent platform capable of reasoning, making decisions andexecuting actions in CRM, following business rules and leaving an auditable history.
Benefits?
– Real-time customization, native integration, complete governance and traceability.
Limits?
– Dependence on Salesforce, need for careful data preparation and potentially high costs.
Target audience?
– Companies already using Salesforce for sales, marketing and support, who want to automate their processes with intelligent agents.