Salesforce Agentforce Topics

How might we assess potential AI use cases?

- Business Value - Ease of Implementation -- Technical feasibility -- Operational readiness -- Data readiness -- Risks -- Executive buy-in - Build a prioritized backlog - Choose Your First AI Project - Evolve Your AI Strategy

What are Key to AI Success?

- Define an AI Strategy - Create Your Data Strategy - Assemble Strategic Stakeholders - Establish the Committee's Responsibilities - Define Your AI Vision - Establish Business Goals - Align on AI Ambitions

How can we inventory our organization's AI usage?

- Identify AI technologies - Document use cases - Map data flows - Establish ownership - Update regularly

What is a data inventory?

A data inventory helps you manage diverse data assets and identify potential issues. Identify what data you need in your project. Identify where the data is stored. Answer some questions about your data. - Is the data type structured, unstructured, or semi-structured? (Learn more about data classification in Data Fundamentals for AI.) - How often is your data refreshed? - Is the data updated in real-time, hourly, daily, monthly, or static?How can the data be accessed? - Have governance standards been implemented for the data - What are some data considerations that can cause challenges in your project?

What is a field generation prompt template?

A field generation prompt template is a specific type of prompt template that's designed to return a generated response from an LLM to a specific field on a record. Field generation prompt templates have a direct relationship to record fields.

What is a prompt template?

A prompt template is a reusable prompt. Prompt templates include placeholders for specific details about customers, products, and more

What is a two-pronged approach to AI strategy?

Bottom-up: Democratize access to AI, allowing employees to experiment. It encourages enthusiasm and familiarity with the technology and sparks potential use cases. You can centralize and scale the best ideas. Top-down: Provide direction from leadership to define the overall AI strategy and transform vision into value.

How do autonomous agents differ from chatbots?

Chatbots follow predefined rules. Autonomous agents can execute tasks independently.

What are the key components of autonomous agents?

Data - Data is the foundation of an autonomous agent's functionality. Decision-Making - When an autonomous agent analyzes data, it uses advanced decision-making algorithms to prioritize and execute tasks efficiently. Action Execution - After making data-driven decisions, the agent seamlessly transitions to executing the planned actions. Learning and Adaptation - Over time, the agent continuously learns from each interaction and adapts to improve future performance.

What are the three main categories in which data can be classified?

Data can be classified into three main categories: structured, unstructured, and semi-structured. Structured data is organized and formatted in a specific way, such as in tables or spreadsheets. Unstructured data is not formatted in a specific way and can include text documents, images, audios, and videos. Semi-structured data is a combination of structured and unstructured data. It has some defined structure, but it may also contain unstructured elements. Examples of semi-structured data include XML (Extensible Markup Language) or JSON (JavaScript Object Notation) files.

What are the 7 stages of the data lifecycle in AI?

Data collection: In this stage, data is collected from various sources, such as sensors, surveys, and online sources. Data storage: Once data is collected, it must be stored securely. Data processing: In this stage, data is processed to extract insights and patterns. This may include using machine learning algorithms or other data analysis techniques. Data use: Once the data has been processed, it can be used for its intended purpose, such as making decisions or informing policy. Data sharing: At times, it may be necessary to share data with other organizations or individuals. Data retention: Data retention refers to the length of time that data is kept. Data disposal: Once data is no longer needed, it needs to be disposed of securely. This may involve securely deleting digital data or destroying physical media.

What is data masking?

Data masking involves tokenizing each value, so that each value is replaced with a placeholder based on what it represents. This means that the LLM can maintain the context of a conversation and still generate a relevant response. Salesforce uses a blend of pattern matching and advanced machine learning techniques to intelligently mask information behind the scenes

What are some precautions for safe and trusted Autonomous Agents?

Define Clear Boundaries Implement Robust Security Measures Monitor and Audit Integrate Human Oversight Ensure Transparency Test Thoroughly Continuous Learning and Improvement

What are characteristics of the Plan stage of an AI project?

Define the problem to solve with AI and how to measure success. Assess the project's technical and data requirements. Identify the features and customization to solve the problem. Prepare your data. Build a trust strategy. Share your plan with your project stakeholders.

What is Fine Tuning?

Fine-tuning is the process of further training a pre-trained model on a new dataset that is smaller and more specific than the original training dataset. After the main versions are released, sometimes there are specialized versions fine-tuned for specific tasks.

What are some limitations to Generative AI?

Generative AI may generate biased or inappropriate responses based on the data it was trained on. It may struggle with understanding user input context or generating coherent responses.

What does the feedback framework do?

It allows users to provide qualitative feedback in the form of a thumbs up or thumbs down, and if the response wasn't helpful, specify a reason why.

How does dynamic grounding improve a prompt's context?

It pulls authorized, secure, relevant data directly from the customer's org, and includes it in the prompt template.

What is a Large Language Model (LLM)?

LLMs are advanced computer models designed to understand and generate humanlike text. They're trained on vast amounts of text data to learn patterns, language structures, and relationships between words and sentences.

What are parameters in an LLM?

LLMs vary by size, but they typically contain billions of parameters. Parameters are the factors that the model learns during its training process, building the model's understanding of language. The more parameters, the more capacity the model has to learn and capture intricate patterns in the data, improving its ability to produce humanlike text.

What are some details you can add to your prompt that clarify or constrain how the LLM responds?

Limits To prevent hallucinating, give the model guardrails that it must stay within. Language Tell the model what language to generate output in. Style & Tone Give the model style and tone guidelines to follow.

What is Machine Learning (ML)?

Machine learning (ML)

What are five ingredients of a prompt?

Participants Describe who's sending and receiving the model's output. Setting Give the model contextual information. Goal Describe what you hope to achieve with the model's output. Relationships Describe the relationship between the participants involved. Also mention how the model's output relates to the participants. Data Give the model data to work with.

What are the stages of an AI project?

Plan Build Launch

Hallucination

Predictions from generative AI that diverge from an expected response, grounded in facts, are known as hallucinations. They happen for a few reasons, like if the training data was incomplete or biased, or if the model was not designed well.

What are some examples of ethical issues in data collection and analysis?

Privacy violations: Collecting and analyzing personal information without consent, or using personal information for purposes other than those for which it was collected. Data breaches: Unauthorized access to or release of sensitive data, which can result in financial or reputational harm to individuals or organizations. Bias: The presence of systematic errors or inaccuracies in data, algorithms, or decision-making processes that can cause unfair or discriminatory outcomes.

What permission must be assigned to a user for them to create prompt templates?

Prompt Template Manager Permission Set

What do the Trust Layer's prompt defense guardrails help protect against?

Prompt injection attacks

What are the two Types of Data?

Quantitative data is numerical and can be measured and analyzed statistically. Examples of quantitative data include sales figures, customer counts based on geographical location, and website traffic. Qualitative data, is non-numerical and includes text, images, and videos. In many cases, qualitative data can be more difficult to analyze, but it can provide valuable insights into customer preferences and opinions. Examples of qualitative data include customer reviews, social media posts, and survey responses.

What is Prompt Resolution?

Replaces each merge field in the prompt template with CRM data to help refine AI prompts

What are the two Prompt Builder features that allow users to preview prompt responses?

Resolution and Response

What are some examples of Salesforce prompt templates?

Sales Email prompt templates help your team draft truly personalized emails for your customers, products, and events based on record data. Field Generation prompt templates bring generative AI-assisted workflows to custom fields within a Salesforce record. Record Summary prompt templates creates a rich-text summary for a Salesforce record based on the record's data. Flex prompt templates generate content for business purposes not covered by other prompt template types, allowing you to define your own resources.

What are three examples of Salesforce LLMs?

Salesforce LLMs include CodeGen, CodeT5+, and CodeTF.

What are types of Guardrails?

Security Guardrails These guardrails ensure that the project complies with laws and regulations, and that private data and human rights are protected. Common tools here include secure data retrieval, data masking, and zero-data retention. Technical Guardrails These guardrails protect the project from technical attacks by hackers, such as prompt injection and jailbreaking, or other methods to force the model to expose sensitive information. Ethical Guardrails These guardrails keep your project aligned with human values. This includes screening for toxicity and bias.

What is semantic search?

Semantic search uses machine learning and search methods to find relevant information in other data sources that can be automatically included in the prompt.

What are aspects of the Build phase of an AI project?

Set up, customize, or build the solution. Conduct a pilot and collect feedback. Refine your solution.

What are three types of Machine learning (ML)?

Supervised learning: In this machine learning approach, a model learns from labeled data, making predictions based on patterns it finds. The model can then make predictions or classify new, unseen data based on the patterns it has learned during training. Unsupervised learning: Here, the model learns from unlabeled data, finding patterns and relationships without predefined outputs. The model learns to identify similarities, group similar data points, or find underlying hidden patterns in the dataset. Reinforcement learning: This type of learning involves an agent learning through trial and error, taking actions to maximize rewards received from an environment. Reinforcement learning is often used in scenarios where an optimal decision-making strategy needs to be learned through trial and error, such as in robotics, game playing, and autonomous systems. The agent explores different actions and learns from the consequences of its actions to optimize its decision-making process.

What are four techniques for collecting data?

Surveys collect data from a group of people using a set of questions. They can be conducted online or in-person, and are often used to collect data on customer preferences and opinions. Interviews collect data from individuals through one-on-one conversations. They can provide more detailed data than surveys, but they can also be time-consuming. Observation collects data by watching and listening to people or events. This can provide valuable data on customer behavior and product interactions. Web scraping collects data from websites using software tools. It can be used to collect data on competitors, market trends, and customer reviews.

What are examples of Data Format?

Tabular data is structured data that is organized in rows and columns, such as in a spreadsheet. Text data includes unstructured data in the form of text documents, such as emails or reports. Image data can include visual information in the form of a brand logo, charts, and infographics. Geospatial data refers to geographic coordinates and the shape of country maps, representing essential information about the Earth's surface. Time-series data refers to data that can contain information over a period of time, for example, daily stock prices over the past year.

What type of flow do you use with a field generation prompt template?

Template-Triggered Prompt Flow

What is the Agentforce Reasoning Engine?

The Agentforce Reasoning Engine is designed to enhance user interaction through faster, more capable, and more multi-turn conversations. - Multi-turn chat - Topic classification - Instructions and actions - Knowledge Retrieval - Searchable public data

What are four examples of data protection laws and regulations?

The California Consumer Privacy Act (CCPA): A set of regulations that apply to companies that do business in California and collect the personal data of California residents. The Health Insurance Portability and Accountability Act (HIPAA): A set of regulations that apply to healthcare organizations and govern the use and disclosure of protected health information in the United States. The General Data Protection Regulation (GDPR): A set of regulations that apply to all companies that process the personal data of European Union citizens. European Union Artificial Intelligence Act (EU AI Act): Comprehensive AI regulations banning systems with unacceptable risk and giving specific legal requirements for high-risk applications.

What is the Einstein Trust Layer?

The Einstein Trust Layer ensures generative AI is secure by using data and privacy controls that are seamlessly integrated in the Salesforce end-user experience. These controls let Einstein deliver AI that securely uses retrieval augmented generation (RAG) to ground your customer and company data, without introducing potential security risks

What is the Einstein Trust Layer?

The Trust Layer is a sequence of gateways and retrieval mechanisms that together enable trusted and open generative AI.

Channels

The applications where an agent gets work done. This can be your website, CRM, mobile app, Slack, and more.

Knowledge

The data an agent needs to be successful. This could include company knowledge articles, CRM data, external data via Data Cloud, public websites, and so on.

The prompt template for generating an email should describe the intended recipient. Who else should the template mention?

The email sender that the LLM is representing.

Actions

The goals an agent can fulfill. This is the predefined task an agent can execute to do its job based on a trigger or instruction. For example, it could run a flow, prompt template, or Apex.

Guardrails

The guidelines an agent can operate under. These can be natural-language instructions telling the agent what it can and can't do, when to escalate to a human, or could come from built-in security features in the Einstein Trust Layer.

Topic classification

The reasoning engine classifies user utterances into topics based on predefined descriptions, ensuring relevant responses.

What is the brain behind every agent?

The reasoning engine enables agents to think deeply and understand human intent, and take action within the flow of a conversation to call different topics and actions as the conversation shifts. Salesforce relies on the Atlas Reasoning Engine to do this work.

Multi-turn chat

The reasoning engine facilitates interactive communication with users by considering and adapting to added conversational context, enhancing the accuracy of the service provided.

Knowledge Retrieval

The reasoning engine uses multiple techniques including advanced retrieval augmentation generation (RAG) which selectively uses multiple language models to iteratively refine the quality of queries, retrieving the most relevant knowledge chunks, while also evaluating the quality of the response.

What is a Prediction?

The text that a generative AI generates is really just another form of prediction. It predicts a sequence of words that are likely to have meaning and relevance to the reader.

How do specificity and context impact prompt responses?

They enhance the output by tailoring it to your requirements.

What is the audit trail?

Timestamped metadata is collected into an audit trail. This includes the prompt, the original unfiltered response, any toxic language scores, and feedback collected along the way.

What type of Salesforce license do you need to create a sales email prompt template in Prompt Builder?

To create a sales email prompt template in Prompt Builder, you need the Einstein for Sales, Einstein for Platform, or Einstein for Service add-on.

Why is it important to inventory your usage of AI tools and technologies?

To get a complete view of your organization's AI risks.

Why is it important to include direct instructions for the LLM to generate only the expected type of content?

To prevent the LLM from creating content about the process of creating content.

Where can Topics and Actions be created and customized?

Topics and actions can be created and customized using Agent Builder.

What are Topics?

Topics define the range of jobs your agents can handle.

What is LLM training?

Training an LLM is like teaching a robot how to understand and use human language. Training is like a mix of reading practice, quizzes, and special lessons until the robot becomes a language expert.

Why Is Fine-Tuning Important?

Transfer learning: Pre-trained models have already learned a lot of generic features from their extensive training datasets. Fine-tuning allows these models to transfer that general knowledge to specific tasks with relatively small datasets. Efficiency: Training a deep learning model from scratch requires a lot of data and computational resources. With fine-tuning, you're starting with a model that already knows a lot, so you can achieve good performance with less data and time. Better Performance: Models fine-tuned on specific tasks often outperform models trained from scratch on those tasks, as they benefit from the broader knowledge captured during their initial training.

True or False: Data is the fuel driving machine learning.

True

True or False: Effective prompts include both ingredients and instructions.

True

True or False: Natural language-the way we actually speak-is unstructured data.

True

True or False: Salesforce has zero data retention policy with the third-party LLMs.

True

True or False: The standard agents you see in your org are based on your Salesforce licenses.

True

True or false: Grounding is a technique used to provide context and specificity to prompt responses.

True

True or False: The first step in creating an agent is the guided setup.

True - This setup process walks through creating an agent, associating topics, and more.

True or False: Salesforce Supports Third-Party LLMs

True: As part of Salesforce's commitment to an open ecosystem, Einstein is designed to host LLMs from Amazon, Anthropic, Cohere, and others—entirely within the Salesforce infrastructure.

What makes Flex templates different?

Unlike other prompt templates, Flex template inputs aren't predefined; instead, you choose them during template creation. In Prompt Builder, you have the option to add up to five inputs to a Flex template.

What are examples of elements of natural language in English?

Vocabulary: The words we use Grammar: The rules governing sentence structure Syntax: How words are combined to form sentences according to grammar Semantics: The meaning of words, phrases, and sentences Pragmatics: The context and intent behind cultural or geographic language use Discourse and dialogue: Units larger than a single phrase or sentence, including documents and conversations Phonetics and phonology: The sounds we make when we communicate Morphology: How parts of words can be combined or uncombined to make new words

What are four big questions which help in prompt template creation?

Who is involved, and how are they related? [Key Ingredients: Participants, relationships, data] What are you trying to accomplish? [Key Ingredients: Goal, instructions] What is the context? [Key Ingredients: Setting, tone & style, language] What are the constraints? [Key Ingredients: Limits, instructions]

How can you avoid toxicity in the output from your prompt template?

You need to add limits and guardrails to prompt templates.

Agentforce Agent

Autonomous, proactive applications designed to execute specialized tasks to help employees and customers

What are the individual tasks an agent is configured to do?

Actions

What is an Action?

Actions are the tools Topics can use to get the jobs done.

What are a few of the standard agents you can configure?

Agentforce Agent (Default): This is an agent you can customize for your employees to help them access information, summarize relationships, forecast revenue, all in the flow of work. Agentforce Sales Development Rep (SDR): If you're looking to boost your business's bottom line, this agent can keep leads engaged around the clock, fielding questions and even objections while streamlining the tasks needed to keep reps operating at their best. Examples include using CRM and external data to schedule meetings so that sellers can keep their focus on closing deals. Need to tackle some admin? Agents can help there as well. Agentforce for Setup: Can help find documentation and customize your org with a low-code promptable partner. Agentforce Service Agent: Can be used to provide your customers with personalized interactions, common answers, and a support path for escalation.

What is Agentforce for Sales Coaching?

Agentforce Sales Coach provides deal-specific, personalized feedback to sales reps so they're always ready for great customer conversations. Sales reps can get tailored feedback whether they want to perfect their sales pitch or conduct a role play to nail a negotiation..

What is Agentforce for Sales?

Agentforce helps with sales development by handling personalized outreach to leads, responding to questions, and even scheduling introduction meetings for sales reps.

Searchable public data

Agents can now securely access public data through the Einstein Trust Layer, expanding its knowledge base.

How does an agent take action?

Agents take action and adhere to guardrails using natural language descriptions that outline the tasks and operational boundaries. Here's a summary of how they operate. The agent first receives a trigger, which can be a conversation with an employee or customer, a change in data, or an automation. The agent uses the LLM and natural language descriptions to identify the context and select a topic that best fits the job to be done, including the scope, data required, and necessary conditions. Depending on the task, an agent selects and chains actions. Those actions are executed via flows, apex classes, APIs, or direct prompts. Agents dynamically plan and execute tasks while strictly following predefined guardrails. They also have built-in mechanisms for harm and toxicity detection, using the Einstein Trust Layer, ensuring they avoid engaging in inappropriate or harmful activities.

Role

An agent's purpose. This defines the job to be done and the broader goals the agent should achieve on your team.

What are aspects of the Launch phase of an AI project?

Announce the change to the organization. Deliver training. Take a baseline measurement of your metrics. Roll out to all end users. Collect feedback. Evaluate the project's success.

What are autonomous agents?

Autonomous agents understand and respond to requests, and then act without human intervention. Give the agent a goal, and it generates tasks for itself, completes them, and moves on to the next one until the goal is achieved.

What are some ways we can address bias and promote fairness?

Diversifying data sources: One of the key ways to address bias is to ensure that data is collected from a diverse range of sources. This can help to ensure that the data is representative of the target population and that any biases that may be present in one source are balanced out by other sources. Improving data quality: Another key strategy for addressing bias is to improve data quality. This includes ensuring that the data is accurate, complete, and representative of the target population. It may also include identifying and correcting any errors or biases that may be present in the data. Conducting bias audits: Regularly reviewing data and algorithms to identify and address any biases that may be present is also an important strategy for addressing bias. This may include analyzing the data to identify any patterns or trends that may be indicative of bias and taking corrective action to address them. Incorporating fairness metrics: Another important strategy for promoting fairness is to incorporate fairness metrics into the design of algorithms and decision-making processes. This may include measuring the impact of certain decisions on different groups of people and taking steps to ensure that the decisions are fair and unbiased. Promoting transparency: Promoting transparency is another key strategy for addressing bias and promoting fairness. This may include making data and algorithms available to the public and providing explanations for how decisions are made. It may also include soliciting feedback from stakeholders and incorporating their input into decision-making processes.

What is dynamic grounding?

Dynamic grounding links the conversation to a prompt template and begins replacing the placeholder fields with page context, merge fields, and relevant knowledge articles from the customer record. Dynamic grounding is what makes prompt templates reusable so they can be scaled across an entire organization.

Instructions and actions

Each topic includes specific instructions and actions, such as verifying order details or obtaining further information, to assist users accurately and efficiently.

What are three strategies that can help promote data privacy and confidentiality?

Encryption: Protecting sensitive data by encrypting it so that it can only be accessed by authorized users. Anonymization: Removing personally identifiable information from data so that it can't be linked back to specific individuals. Access controls: Limiting access to sensitive data to authorized users, and ensuring that data is only used for its intended purpose.

True or False: The Einstein Trust Layer lets customers get the benefit of generative AI by compromising their data security and privacy controls.

False

What are some examples of technologies that encompass AI?

Machine learning uses various mathematical algorithms to get insights from data and make predictions. Deep learning uses a specific type of algorithm called a neural network to find associations between a set of inputs and outputs. Deep learning becomes more efficient as the amount of data increases. Natural language processing is a technology that enables machines to take human language as an input and perform actions accordingly. Large language models are advanced computer models designed to understand and generate humanlike text. Computer vision is technology that enables machines to interpret visual information. Robotics is a technology that enables machines to perform physical tasks.

What are merge fields?

Merge fields are what Salesforce uses to connect your prompt templates to Salesforce record fields, such as from sales or service records. When you send your prompt to the LLM, the merge fields are replaced with your specific business context and customer data.

What is natural language processing (NLP)?

Natural language processing (NLP), is a field of artificial intelligence (AI) that combines computer science and linguistics to give apps and AI assistants the ability to understand, interpret, and generate human language in a way that's meaningful and useful to humans.

What is natural language?

Natural language refers to the way humans communicate with each other using words and sentences. It's the language we use in conversations, when we read, write, or listen.

What are the two natural language processing subfields?

Natural language understanding (NLU) and natural language generation (NLG). Data processed from unstructured to structured is called natural language understanding (NLU). Data processed the reverse way-from structured to unstructured-is called natural language generation (NLG)

What are three limitations to the effectiveness of Machine Learning?

Overfitting occurs when the model is too complex and fits the training data too closely, resulting in poor generalization. Underfitting occurs when the model is too simple and does not capture the underlying patterns in the data. Bias occurs when the model is trained on data that is not representative of the real-world population.

What are the two types of parsing in NLP?

Parsing includes syntactic parsing, where elements of natural language are analyzed to identify the underlying grammatical structure, and semantic parsing which derives meaning.