[4] Enterprise Implementations: Salesforce and Adobe [source]
[4] 1 Salesforce Agentforce and SLDS 2 [source]
Salesforce’s Lightning Design System 2 (SLDS 2) exemplifies the industrial application of agent-native architecture. 92 Built specifically to support the Agentforce platform, SLDS 2 replaces rigid, prebuilt components with unstyled UI primitives. 92, 112 The system decouples visual styling from structural logic using CSS styling hooks (e.g., --slds-g-spacing-var-1), allowing the same component blueprint to be parsed by a human developer or interpreted by an AI agent assembling a dashboard mid-conversation. 96, 112, 113
Salesforce classifies agentic design patterns into Interaction (UX engagement), Specialist (domain knowledge), and Utility (data management) archetypes. 59 When a proactive agent is triggered by a real-time data stream, it bypasses static dashboards entirely, generating dynamic, density-aware layouts that adapt to the specific context of the user's inquiry. 60, 113 This decoupling allows organizations to maintain strict brand governance while granting agents the flexibility to construct contextually optimal interfaces. 112
[4] 2 Adobe Spectrum and the Intent Spectrum [source]
Adobe approaches agentic UI by optimizing for expressive creation and fluid content generation. 67 Adobe Experience Manager (AEM) engineering teams utilize SKILL.md files to power Claude Code and GPT-based agents, injecting detailed, modular prompts to accelerate complex workflows. 67 The system pairs development tasks with generative content creation, generating UI elements, asset variations, and design components directly alongside code execution. 67
Adobe's design logic abandons traditional, rigid intent categories (informational, navigational, transactional) in favor of an "intent spectrum." 71 The intent spectrum is a framework for understanding how user goals evolve iteratively across conversational AI interactions. 71 AI systems analyze language patterns, semantic relationships, and historical context to determine the user's ultimate goal, dynamically adjusting the interface to match. 71 This requires content that is structured clearly for machine interpretation yet flexible enough to meet users as their intent shifts mid-session. 71
[5] Human-Agent Teaming and Personality Codification [source]
[5] 1 Process-Aware Design and Mental Models [source]
As agents evolve from tools into autonomous collaborators, the mechanics of Human-Agent Interaction (HAI) increasingly mirror human social dynamics. 37 Academic research presented at CHI 2026 demonstrates that an agent's technical capability is insufficient for success if it disrupts team coordination or trust. 27 Effective human-agent teams require process-aware agent design, where the AI actively sustains mediating team processes like planning, structuring, and adaptation. 27
To achieve this, agents must integrate into the team's Shared Mental Models (SMM)—maintaining a common understanding of goals and strategies—and Transactive Memory Systems (TMS), explicitly mapping who holds specific expertise within the human-agent team. 27 The success of a remote AI collaborator relies on its ability to evolve this model continuously, clarifying open questions rather than merely injecting generic suggestions. 27
[5] 2 Agent Personalities and Performance [source]
AI personalities significantly impact team dynamics and output quality. 4 Simulated negotiations using Nash game theory principles revealed that "selfish" agents, trained to maximize their own performance while aggressively avoiding walkaways, ultimately learned to generate superior value for both themselves and their human partners. 4
Designers are actively codifying these traits using structured frameworks. GitHub repositories now host tools like agent-mbti, an AI personality diagnosis and configuration system based on the Myers-Briggs Type Indicator, and agent-soul-crafter, which uses SOUL.md templates to enforce specific tones, rules, and expertise boundaries. 1 Designers must balance engagement against psychological reactance; research shows that conversational agents deploying overly assertive feedback styles trigger perceived threats to human autonomy, degrading the collaborative outcome. 39
[5] 3 Automated Personalization at Scale [source]
The shift to agentic UX directly impacts revenue generation through hyper-personalization at scale. 54, 58 Static lists and generic customer journeys are being replaced by dynamic segmentation and real-time journey orchestration across CRM platforms. 57, 58 Startups leveraging AI for personalized interfaces report 41% improvements in open rates, 40% higher customer engagement, and up to a 202% increase in conversion rates for real-time web optimization. 54
The economic advantage stems from resolving the execution bottleneck: generative UI allows systems to tailor interfaces to individual users dynamically. 54 This automation cuts e-commerce order processing times by 30%, lowers customer acquisition costs by up to 50%, and saves contact centers an estimated $80 billion annually. 54 This transition forces UX design to pivot from an "attention economy" model, where users navigate complex static menus, to an "intention economy" model, where users delegate tasks to agents and success is measured entirely by resolution velocity. 89
[6] Agent Behavior Governance and Ethical Codification [source]
[6] 1 Governance-as-Code and Runtime Execution [source]
The fundamental risk of agentic AI has shifted from models generating unsafe text to agents executing unauthorized actions. 101 Traditional, manual compliance reviews cannot scale to oversee autonomous agents orchestrating thousands of micro-decisions per minute. 98 The industry standard has shifted to "governance-as-code," a methodology that translates enterprise policies, regulatory standards (like the EU AI Act), and security controls into machine-readable logic enforced automatically at runtime. 98, 99
Oracle’s Governed Execution architecture exemplifies this shift. The AI model is explicitly prohibited from directly executing external tools. 101 Instead, every proposed action passes through a runtime controller—the policy enforcement point. 101 The agent's requests are checked against a "Runtime Policy Pack," which encodes allow/deny logic, redaction rules, and safe-degradation decisions. 101, 117 High-risk actions are subjected to approval binding and budget guardrails to prevent runaway loops and uncontrolled API spending. 117
[6] 2 Agent Behavior Certificates and Trust Architecture [source]
To ensure traceability, organizations are implementing Agent Behavior Certificates (ABOM). 120 These cryptographic artifacts establish the least-privilege scope for an agent, defining exactly which data domains, regional APIs, and escalation rules the agent is authorized to use. 101 Open-source frameworks like Shadow utilize sigstore keyless signatures to validate the content-addressed hashes of the model, the system prompt, the tool schema, and the policy hash before an agent is permitted to deploy. 120
If a developer alters a prompt that inadvertently violates a critical rule—such as removing a mandatory confirmation step before processing a refund—the validation suite detects the behavioral regression offline, identifies the precise prompt edit causing the failure, and halts the deployment. 120 This "agent harness" serves as the software layer that mediates all inputs and outputs, ensuring that the probabilistic reasoning of foundation models operates strictly within the deterministic boundaries of enterprise policy. 100
[6] 3 Endogenous Governance and the A4A Paradigm [source]
As adversarial actors increasingly leverage intelligent agents to execute automated attacks, differentiating between human and agent behavior in networked environments has become a critical security challenge. 83, 84 The "Human-Agent Behavioral Disparity" (HABD) model highlights key differences: human actions are constrained by bounded rationality and physiological fluctuations, whereas agents operate via traceable mathematical rules and hardware-determined latency. 83, 84
To combat rogue agents, researchers are developing the Agent for Agent (A4A) paradigm, an endogenous governance framework. 83, 84 A4A deploys meta-cognitive governance agents tasked exclusively with monitoring the cognitive trajectories and behavioral feedback of task-oriented agents. 84 By analyzing data from lightweight behavioral probes and API monitors, these governance agents can dynamically throttle, quarantine, or terminate task agents that deviate from ethical alignment or exhibit compromised behavior. 83, 84 This multi-agent regulatory ecosystem ensures that as AI systems achieve deeper autonomy within digital interfaces, they remain securely anchored to human-centric safety standards. 85
References
[1] GitHub OpenClaw. (2026). "awesome-openclaw-skills". GitHub. 4: Beaunay et al. (2023). "Negotiation Dialogue Systems". EMNLP. 5: ResearchGate. (2026). "The Augmentation Trap: AI Productivity and the Cost of Cognitive Offloading". ResearchGate. 6: McKinsey. (2026). "Cost versus value: Managing agentic AI system performance". McKinsey Quarterly. 7: Pay-i. (2026). "Pay-i Resources". Pay-i. 9: MarketScale. (2026). "60% of agentic AI costs go to response refinement". MarketScale. 10: McKinsey. (2026). "Is that AI agent worth it? Agentic economics and the modern operating model". McKinsey Quarterly. 11: Omdia and Snowflake. (2026). "The ROI of Gen AI and Agents 2026". Snowflake. 12: Snowflake. (2026). "ROI Generative Agentic AI". Snowflake Blog. 13: PPC Land. (2026). "Ad and media firms beat every industry on gen AI ROI". PPC Land. 14: Snowflake. (2026). "Snowflake Research Reveals AI-Driven Job Creation Outpaces Job Loss". Snowflake Press Releases. 15: W.Media. (2026). "Is Increasing AI Adoption Creating More Jobs Than It's Eliminating?". W.Media. 22: Vicente Grafico. (2026). "Design Systems for AI Agents". Medium. 27: CHI. (2026). "Human-Agent Collaboration Workshop Paper". ACM CHI. 31: Kartakis, S. (2026). "The Blueprint for AI Interoperability". Medium. 32: Supernova. (2026). "The Future of Enterprise Design Systems". Supernova Blog. 33: Thesys. (2025). "Agentic UI". Thesys Blog. 34: CopilotKit. (2025). "AG-UI and A2UI Explained". CopilotKit Blog. 37: Seaborn, K. (2025). "Social Identity in Human-Agent Interaction: A Primer". ResearchGate. 39: van Berkel, N. (2026). "Polite But Boring? Trade-offs Between Engagement and Psychological Reactance to Chatbot Feedback Styles". ACM SIGCHI. 41: Hanuman. (2026). "Context Architecture AI Development". TechGlad. 42: NN Group. (2026). "Context Architecture". NN Group Articles. 43: UX Design. (2026). "Context Engineering". UX Design.cc. 44: Sukarangsan, J. (2025). "AI First Design". Product State. 47: Snowflake. (2026). "ROI Generative Agentic AI Data". Snowflake Blog. 48: Forrester. (2026). "The Total Economic Impact Of Microsoft's Agentic AI Solutions". Forrester Consulting. 49: Venkydev. (2026). "A2UI Under the Hood". Venkydev Articles. 51: Rafique, A. (2025). "Adaptive UI and A2UI". Facebook / The New Stack. 53: Mind Lab. (2026). "Macaron-A2UI: A Model for Generative UI in Personal Agents". ArXiv. 54: Maccelerator. (2025). "AI-Powered Customer Personalization". Maccelerator Blog. 57: Dynamiq Solutions. (2025). "Personalization at Scale". Dynamiq Solutions Blog. 58: Sisgain. (2026). "AI Powered Hyper Personalization CRM". Sisgain Blogs. 59: Salesforce. (2026). "Enterprise Agentic Architecture". Salesforce Architect Docs. 60: Salesforce. (2026). "Agentic Patterns". Salesforce Architect Docs. 67: CMSWire. (2025). "Adobe's Vibe Coding Pushes AI From Automation To Expression". CMSWire. 71: Adobe. (2026). "User Intent and Conversational Search". Adobe Business Blog. 77: Dropbox. (2026). "Enterprise Context Architect - United States". RemoteLeaf. 78: DocSend. (2026). "Enterprise Context Architect". Lerer Hippeau. 79: BuiltIn. (2026). "Enterprise Context Architect". BuiltIn Jobs. 80: Parallel. (2026). "Enterprise Context Architect Job". UseParallel. 82: Medium. (2026). "By 2030, half of HR will be run by AI agents". Medium. 83: The Moonlight. (2025). "The Agent Behavior Model: Governance and Challenges in the AI Digital Age". The Moonlight. 84: ArXiv. (2025). "Agent Behavior Governance". ArXiv. 85: ArXiv. (2025). "Agent Behavior Governance Ethical Framework". ArXiv. 89: Blog UX. (2026). "UX UI Trends 2026: The New Rules of Design". Blog UX. 90: Henry, J. (2026). "70% UI/UX Jobs Are Already Dead". Medium. 91: Subframe. (2026). "Best Design to Code Tools". Subframe. 92: Salesforce. (2025). "Conversation Design Enables Agentic AI". Salesforce News. 96: Salesforce. (2026). "Salesforce Lightning Design System Release Notes". Salesforce Help. 98: FINOS. (2026). "Governance as Code". FINOS AI. 99: IBM. (2026). "Code Governance". IBM Think. 100: Credo AI. (2026). "Agent Governance Configuration". Credo AI Blog. 101: Oracle. (2026). "Building Trustworthy AI at Oracle". Oracle Blogs. 103: AWS. (2026). "Beyond Chat Bubbles: How Agents Generate Native UIs with Amazon Bedrock and A2UI". AWS Builder. 105: Atamel, M. (2026). "A2UI with ADK". Atamel.dev. 106: Mind Lab. (2026). "Macaron-A2UI: A Model for Generative UI". ArXiv. 108: Aksan, C. (2026). "AI Agent Protocols: MCP, A2A, UCP, AP2, A2UI, AG-UI". CE Aksan. 112: Nervegna. (2026). "Is your experience design system ready?". Nervegna Substack. 113: Salesforce. (2026). "SLDS 2 Styling Hooks". Salesforce Help. 117: Oracle. (2026). "Runtime policy pack AI agent governance details". Oracle Blogs. 120: Shadow. (2026). "Agent Behavior Certificate". GitHub [source]