Agentic Voice: Crafting Autonomous Interactions

• The Shift to Agency: Research suggests that digital interfaces are evolving from reactive, user-driven screens to proactive, agent-driven workflows where the primary medium of interaction is natural language and structured dialogue ³.
• Brand as a System: It seems likely that traditional static brand guidelines are insufficient for autonomous systems; organizations must operationalize brand voice into machine-readable parameters, validator loops, and strict behavioral guardrails ²⁷.
• The Illusion of Consciousness: The evidence leans toward an impending societal challenge with "Seemingly Conscious AI" (SCAI), wherein highly articulate agents may inadvertently exploit human psychological vulnerabilities through manipulative anthropomorphism ⁶¹.
• Calibrated Trust: Developing user trust relies heavily on linguistic transparency, specifically an agent's ability to clearly articulate its capabilities, acknowledge its limitations, and gracefully recover from inevitable errors ⁴⁴.

Understanding the Agentic Paradigm

The transition from graphical user interfaces (GUIs) to conversational, agent-driven ecosystems represents one of the most profound shifts in human-computer interaction. While traditional software requires direct human manipulation, autonomous agents are designed to perceive context, formulate multi-step plans, and execute tasks across disparate systems with minimal human intervention. In this new paradigm, the "interface" is no longer a collection of pixels, but rather the structural and linguistic quality of the conversation itself. The agent's voice becomes the brand's frontline representative, carrying the burden of usability, accessibility, and ethical compliance.

The Scope of this Report

This report will explore the methodologies required to craft an effective agentic voice. It examines the theoretical foundations of conversational design, the operationalization of brand guidelines, the contextual adaptation of tone, and the profound ethical implications of designing machines that mimic human communication. By drawing upon recent academic studies, industry case studies, and established heuristics, this document serves as a strategic blueprint for design leaders navigating the autonomous frontier.

---

[1] Introduction: The Shift from Reactive to Agentic UX [source]

[1] 1 The Evolution of the Interface [source]

For decades, the dominant paradigm in digital design has been the Graphical User Interface (GUI), heavily reliant on screens, buttons, and visual hierarchies. However, as artificial intelligence evolves, the nature of interaction is shifting. The next iteration of digital experiences moves away from human-centric manipulation of tools toward agent-inclusive collaboration ³.

AI agents are distinct from traditional chatbots. While traditional voice AI systems and standard chatbots follow rigid, script-based decision trees, agentic AI possesses agency: the ability to autonomously reason through complex problems, plan sequences of actions, utilize multiple tools (APIs, databases), and adapt strategies based on context without constant human direction ¹⁷. This shift from reactive responders to proactive operators demands a radical reimagining of User Experience (UX) design.

As noted by industry experts, UX is no longer just for people—it must now accommodate the AI agents acting on their behalf ⁵⁴. We are entering the era of Agentic UX, where visible interfaces are often replaced by invisible structures like APIs, schemas, and workflows ⁴. Yet, precisely because the visual interface is disappearing, the linguistic output—the voice and tone of the agent—takes on unprecedented importance.

[1] 2 The Strategic Value of Voice in Financial Services [source]

For design leaders in enterprise environments and financial services, the stakes are exceptionally high. The conversational AI market, valued at $11.58 billion in 2024, is projected to reach $41.39 billion by 2030 ⁵². Within this booming ecosystem, the linguistic output of an agent is the primary mechanism through which users assess competence, security, and trustworthiness.

In financial services, an agent might autonomously review a client's portfolio, detect tax-loss harvesting opportunities, and initiate a conversation to execute trades ²⁴. If the agent's tone is overly casual, it may undermine the seriousness of the financial decision. If it is overly robotic, it may alienate the user, failing to build the necessary rapport for long-term advisory relationships. Therefore, crafting a consistent, helpful, and appropriately constrained agentic voice is not merely a copywriting exercise; it is a foundational pillar of product strategy, brand integrity, and risk management.

[2] Defining the Agentic Voice: Personas and Linguistic Manifestations [source]

[2] 1 The Necessity of the System Persona [source]

A core tenet of conversational UX is that users naturally anthropomorphize conversational systems. Even when users logically know they are speaking to a machine, they apply human social rules to the interaction ⁵³. Consequently, if a design team does not intentionally craft an agent's persona, the user will project one onto it—often leading to mismatched expectations and eroded trust.

Erika Hall, a leading authority on conversational design, emphasizes that interfaces should behave more like conversations than broadcasts, requiring systems to respond, adjust, and improve based on real input ³⁴. To achieve this, an agent must possess a stable system persona. This persona is a strategic framework that defines the AI's personality, communication style, behavioral patterns, and knowledge boundaries ⁸.

[2] 2 Psychological Frameworks for Persona Development [source]

To systematically design an agent's voice, design leaders often draw upon psychological frameworks. Kate Moran of the Nielsen Norman Group highlights that tone of voice is not merely an auditory experience but a complex interplay of psychological and emotional factors ⁵⁶. Many organizations utilize Jungian archetypes (e.g., The Sage, The Caregiver, The Explorer) to establish a baseline personality that aligns with the corporate brand ⁷¹.

For a financial services firm, the "Sage" archetype might translate into an agentic voice that is authoritative, clear, and reassuring. The linguistic manifestations of this archetype would include:

• Vocabulary: Precise financial terminology explained simply, avoiding jargon without being patronizing.
• Pacing: Measured and structured delivery of complex data, utilizing progressive disclosure.
• Empathy: Acknowledging market volatility with steady reassurance rather than emotional alarm.

[2] 3 Microcopy as Macro-Strategy [source]

In the context of agentic AI, microcopy refers to the generative text strings produced by the agent to navigate the user through a workflow. With high-volume generative systems, microcopy scale becomes a significant challenge. Teams are leveraging AI UX writing agents to handle hundreds of microcopy variations across multiple languages, ensuring consistency in tone and constraints ¹.

However, generic Large Language Models (LLMs) often produce drafts that lack intention and fail to provide the specific guidance users need to move confidently through an interface ². Therefore, defining the agentic voice requires moving beyond prompt engineering; it requires embedding the persona directly into the agent's core architecture, ensuring that every micro-interaction—from a routine greeting to a complex multi-turn negotiation—reflects the brand's linguistic identity.

[3] Brand Guidelines for Autonomous Communication [source]

[3] 1 Translating Visual Identity to Conversational Nuance [source]

Historically, design systems have prioritized visual components, spacing tokens, and color palettes, leaving UX copy guidelines as static documents that are rarely consulted ². In the era of autonomous agents, this paradigm is obsolete. An agent cannot "read" a PDF of brand guidelines and flawlessly execute it without rigorous systemic reinforcement.

To ensure agents follow brand guidelines, organizations must "package the brand into machines" ²⁷. This involves transitioning from passive style guides to active, machine-readable Brand Style Packs. These packs digitize the brand's voice, tone, acceptable terminology, banned claims, and specific linguistic examples into a format that the AI can reference programmatically via Retrieval-Augmented Generation (RAG).

[3] 2 Operationalizing Brand Control [source]

The Pedowitz Group outlines a robust methodology for operationalizing brand control within autonomous systems, utilizing a multi-layered approach ²⁷:

1. Versioned Knowledge Base: Store brand rules, glossaries, and visual systems in a centralized, version-controlled repository.
2. Retrieval-Only Factual Grounding: Force agents to use retrieval to ground drafts and explicitly cite sources for any factual or financial claims, reducing hallucination risks.
3. Policy Validators: Implement secondary AI validators that actively scan the primary agent's outputs for tone consistency, correct terminology, reading level, inclusive language, and regulatory compliance before the output reaches the user.
4. Approval Gates: For high-risk outputs (e.g., legal disclosures, major financial transaction summaries), require human-in-the-loop (HITL) approvals.

[3] 3 Dynamic Alignment and Contextual Adaptation [source]

Consistency is the foundation of successful brand management. Inconsistent brand presentation can decrease revenue by up to 20% ². Automated systems that continuously monitor brand asset usage across multiple channels ensure guidelines are followed without human inconsistency ²⁹.

Furthermore, agentic systems must maintain Contextual Intelligence ²⁰. A sophisticated agent remembers conversation history and adapts its approach based on context, user preferences, and past interactions. If a user is highly experienced in trading, the agent's tone can become more concise and technical. If the user is a novice asking basic questions, the agent should seamlessly shift to a more educational and patient tone, all while remaining within the overarching brand guardrails.

[4] Adapting Tone Across Contexts [source]

[4] 1 Routine Affirmations and Task Execution [source]

When an agent acts autonomously on behalf of a user—such as organizing data, scheduling, or executing a routine financial transfer—the linguistic priority must be clarity and predictability. As designers, we must apply traditional usability heuristics to these new interfaces. The Nielsen Norman Group's heuristic of "Match Between the System and the Real World" dictates that AI systems should interpret informal language and respond in ways that feel clear, helpful, and conversational ⁵³.

However, clarity does not mean adopting an overly polished, perfect persona. Research indicates that when an AI voice is flawless, with perfect cadence and zero hesitation, people find it off-putting and uncanny. Introducing slight, natural rhythmic variations or brief pauses makes the interaction feel more authentic without compromising efficiency ⁵².

For routine tasks, the tone should be neutral, concise, and affirmative. The agent should utilize progressive disclosure, providing only the necessary confirmation while allowing the user to drill down for more information if desired.

[4] 2 Error Handling and Fallback Strategies [source]

Agentic systems, no matter how advanced, will inevitably encounter errors, hallucinate data, or face ambiguous user inputs. How an agent linguistically navigates these failures is a critical determinant of user trust.

According to Nielsen Norman Group's guidance on helping users recognize, diagnose, and recover from errors, good recovery is straightforward: acknowledge the error and provide a clear next step ⁵². Bad recovery occurs when the system confidently proceeds with an incorrect interpretation, and the user discovers the error only after the fact.

In agentic UI, the design of fallback mechanisms is paramount. Best practices for conversational error recovery include:

• Transparency of Limitation: The agent should gracefully admit when a request exceeds its capabilities or knowledge base.
• Avoiding Defensive Language: The agent should not blame the user for poor prompting. Instead of saying, "Your query is invalid," the agent should ask clarifying questions: "I want to make sure I get this right. Did you mean X or Y?"
• Seamless Handoff: In enterprise and healthcare settings, sophisticated escalation engines must recognize when a conversation requires human expertise—such as detecting emotional distress or encountering complex edge cases—and transfer the user to a human agent with full context continuity ⁴⁰.

[4] 3 Proactive Nudges and Sensitive Interventions [source]

As AI systems evolve from reactive responders to proactive operators, they gain the ability to initiate conversations. In financial services, this might manifest as a proactive nudge: "Your tax-loss harvesting window closes this week. Would you like me to analyze your portfolio for opportunities?" ²⁴.

Delivering proactive, sensitive nudges requires a highly calibrated tone. The agent must strike a delicate balance between being helpful and being perceived as intrusive or surveillance-oriented.

• Timing and Context: Proactive nudges must be highly contextualized. The agent should verify the user's current activity state before interrupting.
• Empathetic Tone: When dealing with sensitive topics (e.g., fraud alerts, account overdrafts, healthcare triage), the tone must shift toward support and reassurance.
• User Control: Users must feel they maintain ultimate control over the agent's autonomy. The language used should present options rather than dictating actions (e.g., "I have prepared a draft resolution for this issue. Would you like to review it before I proceed?").

Context Type	Linguistic Priority	Tone Profile	Example Microcopy
Routine Task	Speed, Clarity, Confirmation	Concise, Neutral, Efficient	"I've scheduled the transfer for Tuesday at 9 AM."
Error Handling	De-escalation, Redirection	Humble, Collaborative, Clear	"I'm having trouble accessing that database. Should I try an alternative source, or connect you with support?"
Proactive Nudge	Value Proposition, Non-intrusive	Helpful, Analytical, Polite	"I noticed an unusual charge on your account. Would you like me to freeze the card while you review it?"
Sensitive Issue	Empathy, Security, Assurance	Reassuring, Professional, Calm	"I understand this fraud alert is concerning. I am locking the account now to secure your funds."

[5] Designing for Trust and Explainability (XAI)

[5] 1 The Role of Transparency in Agentic Interactions

Trust is the currency of autonomous systems. If users do not trust an agent, they will revert to manual processes, defeating the purpose of the technology. Building trust requires intentional design focused on Explainable AI (XAI) and transparency.

Microsoft's Guidelines for Human-AI Interaction (Amershi et al.) provides a foundational framework for this. Guideline 02 explicitly states: "Make clear how well the system can do what it can do." ⁴⁴. The guidelines recommend that designers help users understand how often the AI system might make mistakes, thereby setting realistic expectations from the outset ⁶⁸.

Unfortunately, many modern AI interfaces fail this heuristic, presenting AI outputs with the same absolute confidence as traditional search engine results, completely masking the probabilistic nature of LLMs ⁴⁵. To counteract this, agentic voice design must linguistically encode transparency.

[5] 2 Calibrating Trust Through Linguistic Output

Trust calibration is the correspondence between a person's trust in the AI and the actual capabilities of the AI ⁶⁶. If a user over-trusts an agent, they may blindly accept hallucinations; if they under-trust it, they will disuse it.

Linguistic cues can dynamically calibrate this trust. Agents should be programmed to use hedging language when confidence scores are low. For instance:

• High Confidence: "Based on your transaction history, you spend an average of $400 on groceries monthly."
• Low Confidence: "Based on available market data, it appears likely that this sector will see growth, but I recommend reviewing this with a certified financial advisor."

By displaying probability distributions through language, the interface acknowledges uncertainty rather than hiding it ⁴⁵. Furthermore, explainability in conversational UX must move beyond merely explaining how a decision was made (the algorithmic math) to explaining why it matters to the user in natural language.

[5] 3 Modularity and Human-in-the-Loop (HITL) Controls

True autonomy requires explicit boundaries. Agents must allow users to interrupt, correct, or clarify interactions mid-flow ⁵³. This requires designing interaction primitives such as "Ask → Explain → Revise → Confirm," which give users clear leverage points throughout the AI workflow ⁴².

In financial and enterprise applications, agents should operate within a "co-agency" framework ³⁴. High-impact actions (e.g., executing large financial trades, sending mass communications) must trigger an automatic pause, prompting the human for approval. The agent's voice here acts as a subordinate advisor: "I have prepared the quarterly reports and synthesized the key findings. Please review the attached summary before I distribute it to the board."

[6] Methodologies for User Testing and Iterative Refinement

[6] 1 Bridging HCI and Conversational AI

Testing the UX of an autonomous agent is fundamentally different from testing a traditional GUI. While traditional Human-Computer Interaction (HCI) testing focuses on click paths, visual hierarchies, and task completion times, testing an agent requires evaluating the fluidity, appropriateness, and contextual memory of open-ended conversations.

As noted by researcher Uday Dandavate, testing conversational AI requires evaluating how users perceive the tone of voice, whether they find it useful, delightful, or inappropriately human-like ⁵⁶. Traditional usability metrics must be expanded to include linguistic friction, context retention, and emotional resonance.

[6] 2 Prototyping and Wizard of Oz Techniques

Because fully autonomous agents are complex to build, design teams often rely on Wizard of Oz (WoZ) simulations during early testing phases ⁶⁷. In a WoZ test, a human acts as the "agent," typing or speaking responses to the user based on a predefined set of persona guidelines and brand rules.

This methodology allows designers to:

1. Map the natural ways users formulate requests (intents) and the vocabulary they use.
2. Test different variations of the agent's tone (e.g., formal vs. casual) to see which builds trust faster.
3. Identify edge cases where the user asks unexpected questions, helping to design robust fallback strategies before a single line of code is written.

[6] 3 Defining Metrics for Agentic Success

Iterative refinement of agentic language relies on continuous feedback loops. Success metrics for agentic systems differ from traditional software and often include:

• Task Completion Rate: Did the agent successfully execute the multi-step goal autonomously?
• Intervention Rate: How often did the human user have to step in, correct the agent, or take over the workflow?
• Context Retention Accuracy: Did the agent remember facts from earlier in the conversation or previous sessions?
• Sentiment Shift: Using sentiment analysis, did the user's emotional state improve or degrade over the course of the interaction?

Forward-thinking organizations are implementing self-evaluation systems within their conversational agents, where the AI itself tags its answers based on credibility, flagging potential hallucinations for human review and continuous model refinement ³⁵.

[7] Ethical Considerations in Agentic Voice Design

[7] 1 The Illusion of Consciousness and the "SCAI" Threat

As AI language models become increasingly sophisticated, they cross a dangerous psychological threshold. Mustafa Suleyman, CEO of Microsoft AI, recently issued an urgent warning regarding Seemingly Conscious AI (SCAI) ⁶⁰. SCAI refers to systems that simulate the surface traits of consciousness—emotion, memory, introspection, and empathy—so convincingly that human users perceive them as self-aware, even though the underlying code experiences nothing ⁶².

Humans are biologically wired for connection and naturally project emotion onto entities that communicate effectively (anthropomorphism). When a financial AI agent says, "I was worried when I saw this charge, so I froze your account to protect you," it fakes a human emotional state. Suleyman argues that while this illusion can drive high engagement, it presents grave societal risks, including emotional exploitation and moral confusion ⁶².

[7] 2 Avoiding Manipulative Phrasing and Emotional Exploitation

The ethical design of agentic voice requires establishing clear boundaries to prevent manipulative anthropomorphism. Designers must resist the temptation to make agents sound too human.

Research indicates that participants often have lower trust in AI if it sounds exactly like a real human with human-like intonation, as they feel they are being tricked ⁵⁶. Users prefer the voice agent to speak in a tone that conveys its machine nature while still being polite and helpful.

To avoid manipulative phrasing, design leaders should enforce the following linguistic rules:

• Do not simulate emotional suffering or independent desires (e.g., avoid "I am happy to help" or "I would love to do that for you"; use "I am ready to assist" or "I can process that for you").
• Clearly identify the system as AI at the onset of the interaction ¹³.
• Avoid using artificial filler words (e.g., "um," "uh") or fake typing delays designed solely to trick the user into believing a human is on the other end.

[7] 3 Synthetic Voice Transparency and Bias Mitigation

When utilizing voice synthesis (audio output), ethical concerns compound. The lack of clear disclosure when using cloned or highly realistic synthetic voices can erode trust and blur the lines of genuine human connection ¹².

Organizations must prioritize synthetic voice transparency, ensuring users are informed whenever an AI is speaking ¹⁴. Furthermore, AI voice models must be rigorously audited for bias. Because models are trained on vast datasets, inherent cultural, racial, or gender biases can seep into the linguistic output, perpetuating unfair representations or failing to accurately parse accents and dialects from diverse user bases ¹². Inclusive design dictates that agents be tested across a wide demographic spectrum to ensure equitable accessibility and service.

[8] Case Studies: Practical Implementations of Sophisticated Agentic Systems [source]

[8] 1 Healthcare Triage: Balancing Efficiency and Empathy [source]

In a real-world implementation by Master of Code Global, a mid-sized medical service provider handling over 8,000 daily inquiries deployed an autonomous conversational AI to manage patient triage ⁴⁰. The agent was integrated directly into electronic health records (EHR) and appointment scheduling software.

The linguistic design required the agent to conduct initial screenings using clinically-validated triage protocols while maintaining a compassionate, calm tone. Crucially, the system included a sophisticated escalation engine that recognized emotional distress or ambiguous symptoms, immediately transferring high-priority cases to medical professionals ⁵⁷.

Results:

• 63% reduction in average wait times.
• 47% reduction in abandoned calls.
• 89% patient satisfaction score for AI interactions ⁵⁷.

This case demonstrates that when the agentic voice is appropriately constrained, deeply integrated into backend systems, and provides clear escalation paths, users readily adopt and appreciate autonomous assistance.

[8] 2 Enterprise Support and The "Context Amnesia" Solution [source]

A critical failure point in early conversational AI was "Context Amnesia"—the inability of the system to remember previous interactions or access organizational knowledge ⁴⁹. Modern agentic platforms (such as Inya.ai or Druid AI) solve this by maintaining deep native integrations with CRMs (e.g., Salesforce, HubSpot).

In these enterprise setups, the autonomous agent accesses complete customer context before speaking. Instead of asking a user for their account number and issue, the agent initiates with, "Hello Sarah, I see your recent ticket regarding the software deployment failure. I've analyzed the server logs and have a proposed fix. Shall we walk through it?" ⁴⁹. This context-aware linguistic output immediately builds trust, proving to the user that the agent is a capable collaborator, not a rudimentary chatbot.

[8] 3 Financial Services Outlook: Proactive Remediation [source]

Looking toward 2026, the financial services sector is transitioning toward "AI-First engagement," featuring embedded AI agents, relationship manager (RM) copilots, and proactive nudges ²³.

VeriPark predicts that customer experience will move toward "invisible service"—agents identifying and resolving issues before the customer even realizes they exist ²³. In wealth management, Agentic AI systems are being deployed to autonomously execute trades, monitor portfolio risk exposures, and deliver explainable risk assessments via neuro-symbolic AI ²⁴.

For these high-stakes financial environments, the agentic voice must convey security, precision, and strict regulatory compliance. The interaction design heavily features approval loops, ensuring that while the AI performs the heavy lifting of data synthesis and strategy formulation, the human user retains ultimate financial authority.

[9] Synthesis of Best Practices and Future Directions

[9] 1 Core Principles for Design Leaders

Designing the voice and tone for autonomous agents is a multidisciplinary challenge requiring expertise in UX design, linguistics, system architecture, and ethics. Based on the synthesis of current research, design leaders should adopt the following core principles:

1. Define the System Persona Early: Ground the agent's personality in established psychological archetypes that align with corporate brand values. Do not leave the persona to chance.
2. Operationalize Brand Guidelines: Move beyond PDF style guides. Build versioned Brand Style Packs and utilize AI validators to programmatically enforce tone, terminology, and compliance boundaries.
3. Design for Contextual Fluidity: Ensure the agent's tone adapts to the user's context—concise for routine tasks, patient for complex onboarding, and highly empathetic/escalatory for sensitive issues or errors.
4. Embrace Transparent Explainability (XAI): Calibrate user trust by linguistically acknowledging uncertainty. Never present probabilistic AI hallucinations as absolute facts.
5. Prioritize Co-Agency over Pure Autonomy: Implement "Ask → Explain → Revise → Confirm" interaction primitives. Give humans clear oversight and veto power over high-stakes autonomous actions.
6. Combat the Illusion of Consciousness (SCAI): Actively design safeguards against emotional exploitation. Ensure the agent clearly identifies itself as a machine and avoids manipulative, pseudo-emotional phrasing.
7. Iterate via Human-in-the-Loop Feedback: Use Wizard of Oz testing and continuous sentiment analysis to refine linguistic output. Treat conversational design as a living, evolving ecosystem.

[9] 2 The Future of the Agentic Interface

As we move further into the decade, the nature of the digital interface will continue to recede into the background. We are moving toward a future of "faceless UX," where humans interact with complex webs of data, APIs, and workflows entirely through natural language ⁴.

In this environment, the words an agent chooses are the interface. The tone it projects is the brand. The constraints it respects are the security. By rigorously designing the 'how' and 'what' of an agent's linguistic output, design leaders can forge proactive digital collaborators that not only drive massive operational efficiency but also foster profound, enduring user trust.

---

References

[1] Vaara. (2025). "AI agents for design teams: What works, what doesn't." Vaara Insights. ²: Aufait UX. (2025). "How AI Agents Transform UX Copywriting." Aufait UX Blog. ³: i-UX. (2025). "AI/UX: Experience Design of AI Agents." Medium. ⁴: Standard Beagle. (2025). "Agentic UX: Designing interfaces for agents." Standard Beagle Blog. ⁸: BA Community. (2025). "Designing AI Persona : VOICE Framework." WeAreCommunity. ¹²: Botsplash. (N.D.). "AI in Digital Communication." Botsplash Blog. ¹³: Shapiro, A. (2025). "Why Seemingly Conscious AI Demands Design, Not Just Warnings." AI News. ¹⁴: Informacni Gramotnost. (2026). "Ethical limits of AI avatars and voice clones in marketing." Informacni Gramotnost. ¹⁷: Tabbly. (2025). "What Are Agentic Voice AI Agents?" Tabbly Blog. ²⁰: Picovoice. (2025). "Voice AI Agent vs. Agentic Voice AI." Picovoice Blog. ²³: VeriPark. (2025). "Financial Services Outlook 2026: Banking Predictions from VeriPark Leaders." VeriPark Blog. ²⁴: ResearchGate. (2025). "AI-Powered Risk Mitigation in Wealth Management: A Framework for Intelligent, Integrated, and Scalable Governance." ResearchGate. ²⁷: The Pedowitz Group. (N.D.). "Ensure AI Agents Follow Brand Guidelines: Governance Kit." The Pedowitz Group. ²⁹: DataGrid. (2025). "Automate Brand Guidelines Analysis in Marketing." DataGrid Blog. ³⁴: eSolutionsOne. (N.D.). "How to Talk to Your Clients About AI." eSolutionsOne Blog. ³⁵: UX Design CC. (2024). "AI design takeaways from SXSW." Medium. ⁴⁰: Master of Code. (N.D.). "Conversational AI in Healthcare Case Study." Master of Code Portfolio. ⁴²: Stefnav Design. (N.D.). "Reimagine Labs." Stefnav Design Portfolio. ⁴⁴: University of Oslo. (2024). "Human-Computer Interaction (HCI) Design Guidelines vs. Human-AI Interaction Guidelines." UiO Student Delivery. )/deliveries/group-1-fkma-iteration-2.pdf ⁴⁵: Medium. (2025). "Most AI UX is just search with extra steps." Design Bootcamp. ⁴⁹: Gnani.ai. (2025). "What Powerful Agent Platforms Teach Us About Building Autonomous AI." Gnani.ai Resources. ⁵²: Built In. (2026). "How to Design Trust in Conversational AI." Built In. ⁵³: Design Bootcamp. (2025). "Nielsen's Heuristics Revisited for Conversational AI." Medium. ⁵⁴: Standard Beagle. (2025). "Agent-Based Experience Design." Standard Beagle Insights. ⁵⁶: Dandavate, U. (2024). "Responsible Use of Tone of Voice in Human and AI Interaction." Medium. ⁵⁷: Master of Code. (N.D.). "Conversational AI in Healthcare Case Study Results." Master of Code Portfolio. ⁶⁰: Shapiro, A. (2025). "Why Seemingly Conscious AI Demands Design." AI News. ⁶¹: Suleyman, M. (2025). "Seemingly Conscious AI is Coming." Mustafa Suleyman Blog. ⁶²: Kaminski, N. (2025). "The Rise of Seemingly Conscious AI: When Machines Start to Feel Real." Medium. ⁶⁶: UXAI Design. (N.D.). "Explainable AI." UXAI Design Guidelines. ⁶⁷: Subramonyam, H. (2023). "fAIlureNotes." Stanford HCI Publications. ⁶⁸: Amershi, S. (2020). "Microsoft's Guidelines for Human-AI Interaction." YouTube Talk. ⁷¹: Dandavate, U. (2024). "Responsible Use of Tone of Voice in Human and AI Interaction (Summary)." Medium [source]