Salesforce AI (Joyce Kay Avila) (z-library.sk, 1lib.sk, z-lib.sk)

【One-Line Pitch】 A structured, product-by-product map of how Salesforce's AI stack—Data 360, AI Model Builder, Tableau Next, Agentforce, Slack, and MuleSoft—fits together into a governed "agentic enterprise." Best for Salesforce architects, admins, developers, and consultants who need the big picture and a shared vocabulary before diving into individual tools. 【Book Arc】 - **Opening (~0%–10%)**: Frames the "agentic enterprise"—humans and AI agents working side by side inside governed systems—and explains why the book spans the whole stack rather than Agentforce alone. Defines audience and goals. - **Early (~10%–36%)**: Establishes prerequisites and scope: who benefits (architects, developers, admins, consultants, plus strategists and decision-makers), what prior knowledge is assumed (general CRM and AI familiarity), and what the reader will be able to do by the end. - **Middle (~36%–52%)**: Lays out the learning path chapter by chapter—data and context (Data 360, AI Model Builder, Semantic Model Builder, Tableau Next), conversational self-service (Einstein Bot Builder), then Agentforce internals (Agent Script, Agent Graph, Atlas reasoning engine) in both no-code and pro-code tracks. - **Late (~52%–64%)**: Moves into domain applications (Sales, Service, Marketing), then beyond classic CRM into Slack as a conversational front door and MuleSoft Agent Fabric for multi-agent orchestration and integration. - **Ending (~64%+)**: Consolidates everything into a layered architecture model—twelve distinct "systems" (record, reference, context, insight, work, engagement, collaboration, agency, orchestration, integration, control, coordination)—and maps each Salesforce product to the layers it serves. 【Key Takeaways】 - **The unit of analysis is the "agentic enterprise," not a single product** (Opening): the book's central argument is that agents only work when data, context, orchestration, and governance are in place together—so it deliberately covers the whole stack. This is the frame that makes the rest of the tool chapters cohere. - **Twelve layers separate concerns that are easy to conflate** (Ending): record, reference, context, insight, work, engagement, collaboration, agency, orchestration, integration, control, coordination. Use this table as a diagnostic: when an AI initiative stalls, ask which layer is missing rather than which tool is wrong. - **A layered platform beats a pile of tools** (Ending): Salesforce's claimed advantage is a composable, governed fabric sharing one data model, APIs, and metadata—not feature count. Relevant when weighing build-vs-buy or assembling best-of-breed alternatives. - **Context is manufactured, not assumed** (Middle): predictions from AI Model Builder, semantic models from Semantic Model Builder, and visualizations in Tableau Next are all presented as ways to feed context to both humans and agents. Treat context engineering as a deliverable, not a side effect. - **Chatbots and agents are complementary, not competing** (Middle): Einstein Bot Builder suits low-risk deflection and can borrow agentic features instead of rigid dialog trees; Agentforce handles deeper agency. Match the mechanism to the risk profile of the interaction. - **Agentforce has a real internal architecture worth learning** (Middle): Agent Script (a 2025 domain-specific language mixing deterministic logic and prompt instructions via subagent blocks), Agent Graph, and the Atlas reasoning engine. Understanding these explains why some use cases resist no-code solutions. - **No-code first, pro-code when complexity demands it** (Middle): the book explicitly sequences standard Agentforce agents before developer tooling, acknowledging cases that cannot be built declaratively. - **Slack and MuleSoft extend the enterprise beyond CRM** (Late): Slack as the conversational front door and collaboration surface; MuleSoft Agent Fabric for integration and multi-agent orchestration across ecosystems (Salesforce, SAP, Oracle, Workday) and with human workers. 【Reading Tips】 - **Read the architecture chapters twice.** The twelve-layer table and the product-to-layer mapping are the book's spine; skim them first, then return after the tool chapters so the mapping lands. - **Skim the audience and conventions material.** The preface and front matter repeat scope and role lists across several chunks—extract the "by the end you will" list and move on. - **Deep-read Agentforce (Chapters 6–7) if you build agents.** The reasoning-engine comparison and Agent Script subagent blocks are the densest technical content; budget extra time and expect to consult Salesforce docs alongside. - **Treat this as a reference, not a tutorial.** The author states there are no coding examples—only screenshots of code, YAML, and JSON. Pair it with hands-on org access or Trailhead for actual implementation. - **Use the certification angle deliberately.** The book positions itself as preparation for newer exams (Tableau Next, Marketing Cloud Next); if that's your goal, read with the exam objectives open beside you. 【Coverage Limits】 The available excerpts concentrate on front matter, the preface, the chapter roadmap, and the opening architecture chapter; the detailed tool chapters (AI Model Builder, Tableau Next, Agentforce internals, Slack, MuleSoft) are described but their actual content is not sampled here. Specific procedures, screenshots, and best practices in those chapters are therefore not summarized in this guide.