독서리서치레이더투자 프레임워크
로그인 / 회원가입
← 읽기로 돌아가기
주제DisconfirmAI 편집부업데이트됨 2026-11-22 04:39

Agentic Enterprise Architecture

6 단계1 건5 출처

The concept of "Agentic Enterprise Architecture" was introduced by Salesforce at Dreamforce 2026 to address the gap between probabilistic AI reasoning and deterministic business execution. The core argument is that deploying AI agents in actual business workflows fails not due to model intelligence, but because of a missing connection layer that translates judgment into compliant actions governed by existing permissions, approvals, and data structures.

A Deloitte survey cited in the material indicates that only 5% of U.S. companies consider themselves "very ready" for AI agent integration, with 72% struggling to connect scattered data across systems. To resolve this, Salesforce proposed a four-layer architecture: Data 360 (unified data foundation), Customer 360 (business logic and processes), Agentforce (digital workforce executing tasks), and AIPulse (interface connecting humans and agents). This structure ensures AI agents operate within established security policies and approval workflows rather than acting autonomously without guardrails.

Industry adoption shows mixed results. Anthropic uses Salesforce as its CRM while integrating Claude into Salesforce's interface; Siemens reduced lead response times from days to minutes using SiePortal AI; L'Oréal cut average consultation handling time from 7 minutes to 2.5 minutes. However, challenges remain regarding governance, trust, and continuous monitoring. Kathy Baxter, Salesforce’s Principal Architect of Ethical AI Practice, emphasizes that consequential decisions must route through deterministic layers (like Apex or Flow) to ensure auditability and human oversight, warning that pre-launch testing alone is insufficient for probabilistic systems.

현재까지의 과정

  1. 2026-11-19

    Salesforce introduces 'Agentic Enterprise Architecture' at Dreamforce 2026, defining a four-layer model (Data 360, Customer 360, Agentforce, AIPulse) to bridge probabilistic AI and deterministic business rules.

    byline.network ↗
  2. 2026-11-19

    Deloitte survey reveals only 5% of U.S. enterprises are 'very ready' for AI agents, citing data fragmentation and governance as primary barriers.

로그인 / 회원가입
독서리서치레이더투자 프레임워크
byline.network ↗
  • 2026-11-19

    Anthropic adopts Salesforce as its core CRM and Slack for collaboration, while Salesforce integrates Claude into its Agentforce platform, demonstrating mutual dependency.

    byline.network ↗
  • 2026-11-19

    Siemens implements AI in SiePortal, reducing sales inquiry response times from days to minutes and verifying leads from 132 countries.

    byline.network ↗
  • 2026-11-19

    L'Oréal deploys AI for customer service, cutting average consultation handling time from 7 minutes to 2.5 minutes and promo code issuance from 10 minutes to 1 minute.

    byline.network ↗
  • 2026-10-06

    Kathy Baxter (Salesforce) details the dual-layer approach: generative LLMs handle reasoning, while deterministic layers (Agent Fabric, Flow, Apex) enforce policies and audit trails.

    salesforce.com ↗
  • 아직 해결되지 않은 문제

    • How will enterprises balance the need for AI autonomy with strict regulatory compliance and data residency requirements?
    • What metrics will define success beyond speed, such as error rates in deterministic enforcement or long-term ROI of agentic workflows?
    • Will non-Salesforce vendors adopt similar layered architectures, or will proprietary ecosystems fragment the market further?

    출처

    • byline.networkByline Network↗
    • architect.salesforce.comSalesforce Architect Docs↗
    • salesforce.comSalesforce Blog↗
    • aws.amazon.comAWS Architecture Blog↗
    • idc.comIDC Blog↗

    이 주제 커버리지

    • The Real Bottleneck for Enterprise AI Agents: How Probabilistic Reasoning Connects to Deterministic Business Rules2026-11-19뉴스

      At Dreamforce 2026, Salesforce proposed an 'Agentic Enterprise Architecture,' arguing that the probabilistic reasoning of large language models and a company's deterministic business rules (approvals, permissions, contract terms) belong to different systems. The difficulty in deploying AI agents for actual business execution is not due to insufficient model intelligence, but rather the lack of a connection layer that translates judgment into compliant execution. A Deloitte survey found that only 5% of U.S. companies consider themselves 'very ready' to introduce AI agents.