# Jobs to Be Done in AI Product Design and Evaluation

The advent of large language models and generative artificial intelligence has fundamentally altered the landscape of software product development. Traditional software engineering operates on deterministic principles, where predefined inputs consistently yield predictable outputs [cite: 1, 2]. In contrast, modern artificial intelligence systems are probabilistic and non-deterministic, generating highly variable outputs dependent on nuanced contextual parameters, prompt structures, and underlying sampling mechanisms [cite: 2, 3]. This paradigm shift has disrupted traditional product management methodologies, rendering feature-centric roadmaps and standard usability metrics insufficient for building reliable, user-centric applications. 

To navigate this complexity, researchers and product strategists are increasingly turning to the Jobs to Be Done (JTBD) framework. Originally popularized to explain disruptive innovation, the framework shifts the focus of product design from user demographics and technological capabilities to the underlying progress a user is trying to make in a specific circumstance [cite: 4, 5]. When applied to artificial intelligence, the framework serves as a vital stabilizing mechanism. It prevents engineering teams from building "solutions in search of a problem" and provides a robust lens through which non-deterministic algorithmic features can be architected, bounded, and evaluated for real-world efficacy [cite: 6, 7].

## Theoretical Foundations of Artificial Intelligence Design

### The Shift from Feature-Led to Outcome-Driven Development

In the rapid commercialization of generative artificial intelligence, a pervasive anti-pattern has emerged: the integration of foundational models into existing interfaces as bolt-on features rather than purposeful tools [cite: 8]. The software industry has witnessed a proliferation of generic chatbots and text-generation widgets added to enterprise platforms without a clear understanding of the user's ultimate goal. This phenomenon represents a classic "solution in search of a problem," where technological novelty precedes human utility, resulting in functional but rarely adopted products [cite: 5, 7].

The JTBD framework counteracts this technology-first bias by demanding that innovators understand the functional, emotional, and social dimensions of a user's goal [cite: 5, 9]. Individuals do not adopt artificial intelligence products simply to interact with a neural network; they "hire" these systems to make tangible progress in their workflows [cite: 4, 10]. For example, a user does not hire a language model merely to "generate text." They may hire it to "draft a difficult email to a dissatisfied client without sounding defensive," which addresses both a functional and emotional job, or to "summarize a 50-page technical report to prepare for a board meeting in five minutes," which represents a specific functional job driven by strict temporal constraints [cite: 4, 9].

By structuring product strategy around specific user pain points and latent needs, organizations can ensure that technological deployment aligns with human reality [cite: 11]. The Reforge framework for generative artificial intelligence product strategy emphasizes this alignment, advocating for a user strategy that prioritizes solving critical customer problems over pushing technological frontiers simply for the sake of utilizing the newest model parameters [cite: 11].

| Dimension | Feature-Led Artificial Intelligence Design | Jobs to Be Done-Led Artificial Intelligence Design |
| :--- | :--- | :--- |
| **Primary Focus** | Model capabilities (e.g., token limits, multimodal inputs, reasoning speed). | User progress and desired outcomes (e.g., reducing time-to-decision, minimizing cognitive load). |
| **Development Origin** | Originates with a new foundational model or API, prompting the question, "What can be built with this?" | Originates with an unmet user need or persistent workflow friction, prompting the question, "Can probabilistic algorithms resolve this?" |
| **Interface Paradigm** | Open-ended chat interfaces or generic prompt inputs requiring the user to articulate complex system commands. | Purpose-built, context-aware interfaces that guide the user directly to the outcome with minimal cognitive overhead. |
| **Success Metric** | Number of features shipped, API calls made, or basic isolated task completion rates. | Behavioral changes post-interaction (e.g., whether the user immediately utilized the output or abandoned the workflow). |
| **Operational Risk** | High risk of creating products with initial novelty but low long-term retention. | Higher probability of product-market fit due to deep integration with existing human workflows and incentives. |

### Core Principles of Outcome-Driven Innovation

To operationalize the JTBD framework in algorithmic development, organizations frequently utilize Outcome-Driven Innovation (ODI) principles. ODI ties customer-defined metrics, known as desired outcomes, directly to the customer's job to be done, effectively rendering the unpredictable process of innovation highly measurable [cite: 12]. In rigorous ODI research, customers may evaluate the importance and current satisfaction levels of up to 150 specific outcome statements related to a complex job [cite: 12]. 

Applying an opportunity algorithm to these satisfaction and importance ratings allows product teams to calculate an opportunity score for each outcome. This quantitative approach reveals priority areas where current solutions, including existing non-AI software, fail to help the user get the job done effectively [cite: 12]. In the context of artificial intelligence, this predictive, data-driven approach prevents data science teams from proposing innovations that customers do not want, and simultaneously prevents product teams from requesting capabilities that existing artificial intelligence models cannot reliably deliver [cite: 13]. It anchors the non-deterministic output of the machine to the highly deterministic success criteria of the user.

### Human-Centered Artificial Intelligence and Task Augmentation

The integration of outcome-driven frameworks into artificial intelligence development is closely tied to the principles of Human-Centered Artificial Intelligence (HCAI). HCAI advocates for the creation of systems that augment human expertise, decision-making, and creativity, acting as collaborators rather than focusing solely on cost-reduction through full workforce automation [cite: 14, 15]. The JTBD framework inherently supports this augmentation mindset because it treats the human user as the primary agent attempting to fulfill a job, with the artificial intelligence acting as a supportive tool hired to eliminate specific frictions [cite: 16].

Research analyzing over 200,000 anonymized Bing Copilot conversations mapped user goals and artificial intelligence actions to the O*NET database, which classifies United States occupations by their core activities [cite: 17, 18]. The data revealed critical insights into how intelligent systems are actively hired in the workforce. Users frequently hire generative systems for tasks involving complex language processing, research, summarization, and communication [cite: 19]. Notably, in 40% of the analyzed cases, the user's primary goal and the algorithm's executed activity involved different task sets. For example, when a user sought to gather information, the artificial intelligence acted similarly to a reference librarian, demonstrating that the system acts primarily as an advisor or coach rather than a direct replacement for the human worker [cite: 17].

This distinction between overarching occupations and granular tasks is crucial for system design. An occupation is not a single job to be done; it is a complex bundle of interconnected jobs [cite: 20, 21]. By applying the JTBD lens, researchers and designers can decompose rigid job titles into specific, augmentable tasks. Data indicates that interpreters, historians, and technical writers exhibit high artificial intelligence applicability scores, reaching up to 0.49, because their daily workflows consist of discrete jobs that align perfectly with the probabilistic strengths of large language models [cite: 17, 18, 19]. Conversely, roles requiring physical effort and on-the-spot human judgment, such as machine operators, repair workers, and caregivers, exhibit applicability scores as low as 0.05, representing areas where current technology offers minimal utility [cite: 17, 18, 19].

## Mapping Artificial Intelligence Capabilities to User Jobs

### Categorization of Generative System Capabilities

To effectively align product strategy with user needs, designers must maintain a clear taxonomy of what foundational models can reliably achieve. Research conducted by the Human-Computer Interaction Institute at Carnegie Mellon University analyzed 85 distinct research applications utilizing generative artificial intelligence to map its core capabilities [cite: 22]. This research extracted 294 specific capabilities, grouping them into 33 clusters, which were further categorized into 13 action capabilities and three primary themes: Generate New Content, Transform Content, and Understand Content [cite: 22]. 

These three capability themes represent the fundamental mechanisms by which artificial intelligence can be hired to complete user jobs. When a user needs to synthesize extensive market research, they hire the system's "Understand Content" and "Transform Content" capabilities. Recognizing these parameters allows teams to identify the intersection between user needs, company goals, and the actual technological trajectory of the models, preventing the deployment of systems for jobs they are statistically likely to fail [cite: 22, 23].

### Emergent Use Cases in Enterprise Workflows

By moving beyond generic conversational interfaces and focusing on specific business outcomes, organizations are discovering highly effective, emergent use cases for large language models. These applications prioritize knowledge management, decision support, compliance, and operations—areas where workflow friction is high and the cost of inefficiency is significant [cite: 24].

Rather than relying on theoretical capabilities, successful enterprise deployments utilize the JTBD framework to restructure organizational data into formats that algorithms can easily process and retrieve.

| Enterprise Use Case | Primary Job to Be Done | Implementation Mechanism | Measurable Outcome |
| :--- | :--- | :--- | :--- |
| **Knowledge Management** [cite: 24] | Quickly retrieve contextualized historical insights from sprawling institutional repositories. | Retrieval-Augmented Generation (RAG) integrated into proprietary document databases. | Consulting firms report 30% time savings on research tasks, with high weekly utilization rates among staff. |
| **Supply Chain Resilience** [cite: 24] | Proactively detect and mitigate external disruptions to logistics and material availability. | Artificial intelligence-driven monitoring systems analyzing global weather, news, and logistics data streams. | Transition from reactive scrambling to proactive route adjustment, maintaining continuity of critical deliveries prior to weather events. |
| **Equipment Troubleshooting** [cite: 25] | Accelerate the diagnostic accuracy and onboarding speed of field repair technicians. | Specialized large language models trained on technical manuals guiding agents through complex hardware diagnostics. | Reduction in agent onboarding and training periods from ten weeks to two weeks. |
| **Marketing Search Optimization** [cite: 26] | Restructure technical documentation to capture algorithmic search intent and generative summaries. | Transitioning long-form blog posts into direct question-and-answer structures utilizing specific FAQ metadata schema. | Immediate 43% increases in organic traffic as AI agents easily extract direct answers for generative search overviews. |

These implementations demonstrate that the most valuable applications of large language models are rarely standalone chatbots; rather, they are embedded systems that quietly eliminate friction, accelerate decision-making, and reduce operational risk within a highly specific job context [cite: 24].

## Architectural Implications for Agentic Systems

### The Limitations of Conversational Interfaces

Early implementations of generative artificial intelligence relied almost exclusively on the conversational chatbot as the primary user interface. However, academic and industry critiques increasingly emphasize that "chatting" with an algorithm is frequently a severe workflow bottleneck rather than a streamlined solution [cite: 8, 27]. When a user is forced to act as a translation layer between their intent and the machine's output—constantly prompting, copying, reformatting, and pasting across disparate applications—the artificial intelligence is merely accelerating individual micro-tasks without completing the overarching job [cite: 8].

Through an outcome-driven lens, the conversation itself is rarely the user's goal; the conversation represents user friction. End-users desire systems that execute complex coordinations seamlessly [cite: 8]. A language model confined within a chat interface operates without the ability to execute actions in the user's broader environment, severely limiting its utility for complex professional jobs [cite: 8].

Furthermore, research into human-computer interaction highlights the cognitive limitations of linear chat interfaces for complex data analysis. The Tableau SyncSense research project identifies that while conversational interfaces capture the chronological history of an analysis, they obscure the structure, branching, and iteration that characterize real-world data workflows [cite: 27]. By flattening dynamic analytical processes into a single continuous scroll, these interfaces make it difficult for users to recover context or follow analytical threads. SyncSense proposes structuring these interactions into interpretable components—threads, speech acts, artifacts, and insights—allowing users to navigate analytical memory rather than merely generating isolated answers [cite: 27].

### Ephemeral Interfaces and Workflow Encapsulation

The friction inherent in chat-based interactions is driving a paradigm shift in system design toward "workflow encapsulation" and "ephemeral interfaces" [cite: 28, 29].

[image delta #1, 0 bytes]

 Instead of relying on static user flows or forcing users to command systems via open text boxes, intelligent systems are increasingly designed to dynamically generate temporary, context-aware interface elements [cite: 28]. These ephemeral UIs appear precisely when needed to facilitate a specific step in a user's job—such as reviewing a generated plan or approving a risk threshold—and vanish once the user's intent is fulfilled [cite: 28].

In Human-Agent Centered Design, task analysis no longer exists primarily to choreograph predetermined interaction steps. Instead, it is utilized to surface implicit knowledge about human intentions, allowing designers to explicitly map these intentions to autonomous system actions [cite: 29]. If the user's job is to consolidate multiple redundant architecture documents, a properly encapsulated workflow does not require a lengthy chat session. It requires an agentic system that can ingest the files, independently identify redundancies, propose a structural merge via an ephemeral review dashboard, and autonomously execute the final consolidation upon human approval [cite: 8, 30]. In this model, the artificial intelligence moves from being a conversational partner to a piece of dynamic infrastructure that the user commands to achieve outcomes.



### Subagents as Modular Jobs to Be Done

As the software industry progresses from isolated large language models to autonomous, multi-step agentic systems, the complexity of engineering reliable software increases exponentially. Monolithic agents tasked with handling broad, undefined roles frequently fail due to unclear state ownership, leaky abstractions, missing observability, and compounding probabilistic errors across multi-turn executions [cite: 20, 31]. Trying to automate an entire job category with a single agent is widely considered a recipe for systemic failure, as professional roles are composed of dozens of discrete tasks requiring varying levels of context and judgment [cite: 20].

To mitigate these distributed failures, enterprise artificial intelligence architectures, such as the Salesforce Agentforce framework, explicitly model "subagents" as discrete jobs to be done [cite: 32, 33]. Rather than building one universal agent to act as a general employee, developers define specialized subagents with strict classification descriptions, precise operational scopes, and bounded tool access [cite: 20, 32]. These subagents rely on predefined actions to perform their specific jobs, such as executing application code, calling specific APIs, running predictive models, or triggering automated flow sequences [cite: 32, 33].

When a user submits a complex query, a master routing agent identifies the specific job the user is trying to accomplish and delegates the execution to the appropriate subagent. This modular approach aligns fundamentally with JTBD theory. It ensures that the probabilistic nature of the language model is constrained within a highly specific functional context. By utilizing deterministic scaffolding—such as "Agent Script" frameworks that enforce immutable paths and mandatory authentication gates regardless of user input—architects can bridge the gap between probabilistic artificial intelligence and rigid business logic, significantly improving reliability and reducing the risk of autonomous hallucination [cite: 32, 33].

### Artificial Intelligence Agents as System Users

The shift toward autonomous agentic workflows necessitates a fundamental expansion in how human-computer interaction practitioners define system users. Traditionally, the term "user" has implicitly meant a human being interacting visually with a graphical interface [cite: 34]. However, artificial intelligence agents are increasingly interacting with these exact same digital interfaces to find information, fill out forms, execute transactions, and accomplish goals on behalf of human principals [cite: 34].

Designing for agents requires reevaluating accessibility principles to ensure machine readability. Elements that rely entirely on visual context—such as icon-only buttons or ambiguous hypertext—present significant hurdles for agents attempting to parse the structural markup of a site [cite: 34]. By prioritizing semantic HTML, predictable form patterns, and explicit state changes, organizations can reduce the likelihood of agent errors compounding across multi-step workflows, ensuring that the agent can successfully execute the job it was hired to perform [cite: 26, 34].

## Evaluating Non-Deterministic Artificial Intelligence Features

### The Insufficiency of Traditional Product Metrics

The most profound impact of non-deterministic artificial intelligence on product management is the necessity of adopting "Evaluation" as a structured, core discipline alongside traditional Discovery and Delivery phases [cite: 1]. In deterministic environments, product outcomes are reliably measured by raw metrics such as task completion times, active user counts, or defect escape rates [cite: 1, 31]. If a deterministic function receives a specific input, the output is structurally guaranteed, making success relatively easy to quantify.

Artificial intelligence fundamentally breaks these assumptions. Given the exact same input, an algorithmically-powered product will generate different outputs depending on sampling parameters, intermediate reasoning states, and subtle environmental contexts [cite: 2, 3, 35]. This variance is often referred to as the "non-deterministic multiplier." Research utilizing the $\tau$-bench framework demonstrated that agents achieving a 60% success rate on initial benchmark tests may exhibit only 25% consistency across multiple trials [cite: 36]. An agent that succeeds more than half the time on a single run might fail three out of four times in a reliable production environment [cite: 36].

Furthermore, the optimization of traditional metrics in machine learning systems is highly susceptible to Goodhart's law, which states that when a measure becomes a target, it ceases to be a good measure [cite: 37]. An algorithm might perfectly execute a prompt instruction, achieving high technical accuracy on a benchmark, while completely failing to resolve the user's underlying anxiety or workflow constraint, resulting in low JTBD fulfillment [cite: 35, 38]. Consequently, systems can exhibit "silent failures," where raw engagement metrics appear healthy, but the output is subtly inaccurate, slowly eroding user trust over time [cite: 1, 38].

### Infrastructure for Non-Deterministic Evaluation

To evaluate probabilistic systems through a JTBD lens, organizations must construct layered statistical evaluation infrastructures that explicitly separate exploratory capability testing from protective regression testing [cite: 36]. 

During the capability evaluation phase, teams map the limits of the artificial intelligence against the complex dimensions of a user's job. These evaluations are exploratory, utilizing live models and real data to discover emergent use cases, and they inherently accept low pass rates [cite: 36, 39]. Once a workflow is stabilized, regression evaluations act as protective, continuous tests. They ensure that as underlying models are swapped or prompts are modified, the system reliably completes the core functional jobs without degrading historical performance [cite: 36].

Evaluating these systems also requires distinct grading layers. While exact-match code-based graders can verify deterministic constraints like API payload structures, model-based graders ( utilizing LLM-as-a-judge patterns) are necessary to systematically assess probabilistic outputs against nuanced criteria such as conciseness, factual compliance, and tone [cite: 2, 36]. 

### Multi-Dimensional Evaluation Frameworks

Evaluating multi-agent systems requires assessing the entire trajectory of the agent's actions—including planning, tool use, reflection, and memory management—rather than solely examining the final text output [cite: 35, 40].

[image delta #2, 0 bytes]

 An agent that hallucinates a memory retrieval step early in a multi-step sequence will fail the overall job, even if the final generated response appears linguistically flawless [cite: 31, 35].

To capture this complexity, advanced evaluation frameworks utilize a multi-axis view to establish confidence and explainable behavior.

| Evaluation Dimension | Measurement Focus | Relevance to Jobs to Be Done Framework |
| :--- | :--- | :--- |
| **Capability and Efficiency** [cite: 35, 40] | Success rates, average response times, token usage, and computational resource consumption. | Assesses whether the system can technically execute the functional job within acceptable constraints (e.g., speed and cost). |
| **Trajectory and Planning** [cite: 31, 35] | Accuracy of tool selection, logic of the intermediate reasoning chain, and error recovery capability. | Evaluates if the agent understands the necessary steps to progress the user toward their ultimate outcome without fatal workflow deviations. |
| **Robustness and Adaptability** [cite: 35, 40] | System resilience in shifting environments, handling of edge cases, and performance under adversarial prompts. | Determines if the product can reliably complete the job across highly variable real-world contexts and non-standard user inputs. |
| **Safety and Compliance** [cite: 2, 35, 40] | Adherence to regulatory constraints (e.g., HIPAA), prevention of data leakage, and frequency of human escalation. | Ensures the emotional and social job of "trust and security" is maintained while executing the functional task. |



### Measuring Human-Artificial Intelligence Team Readiness

Evaluating artificial intelligence feature development through an outcome-driven lens ultimately requires measuring human behavioral changes immediately following the system's output [cite: 3]. Because the precise string of text or generated media cannot be deterministically predicted, the true success of the feature must be inferred from the user's subsequent actions in the workflow. If the system successfully fulfills the job, the user's behavior will demonstrate momentum, such as directly deploying the generated code or sending the drafted response. Conversely, if the system fails, the user will exhibit friction, characterized by repetitive follow-up prompts, manual corrections, or complete abandonment of the tool [cite: 3].

This paradigm moves evaluation away from pure model accuracy toward measuring "Human-AI Team Readiness" [cite: 38]. This measurement framework assesses whether human-AI teams are prepared to collaborate safely and effectively, operating on an Understand-Control-Improve (U-C-I) lifecycle [cite: 38]. A critical metric within this framework is reliance behavior. Evaluators must observe whether users confidently delegate appropriate tasks to the artificial intelligence, or whether they suffer from miscalibrated reliance—overusing the system when it is generating errors (automation bias) or underusing it when it could provide significant utility (algorithmic aversion) [cite: 38]. 

A compelling demonstration of evaluating workflow orchestration over isolated accuracy is found in clinical decision-making studies. Researchers at the Wharton School evaluated a multimodal large language model operating as an autonomous virtual provider within a partially observable Markov decision process (POMDP) simulation [cite: 41]. The study compared the model's policy against an expert emergency room physician benchmark across varying clinical cases. The findings indicated that the agent could successfully engage in a dynamic sequence of information-gathering, interpreting chest X-rays, listening to lung sounds, and initiating treatments under explicit time pressure [cite: 41]. Viewed through the JTBD framework, the study proved the agent could execute the overarching job of patient stabilization by orchestrating a continuous workflow of tasks, recognizing uncertainty, and updating its beliefs, rather than merely answering a static medical question [cite: 41].

## Limitations of the Jobs to Be Done Framework in Artificial Intelligence

While the JTBD framework provides an essential navigational compass for product development, it is not a panacea. The framework was originally conceived to evaluate consumer choices and market positioning, and it possesses inherent blind spots when applied to the unique risks of generative and autonomous systems.

### Epistemic Accuracy and Systemic Safety Constraints

A primary limitation is that the framework does not inherently address questions of epistemic accuracy or systemic safety [cite: 4]. A user may hire a generative system to perform a high-stakes job, such as rendering a medical diagnosis, calculating financial risk, or drafting a legally binding contract. From a pure outcome-driven perspective, if the system provides a confident, well-formatted response that satisfies the user's immediate need for an answer, the psychological "job" appears complete. 

However, foundational models are fundamentally prone to hallucinations and non-deterministic logic failures [cite: 31, 40]. Research demonstrates that while generative platforms might be hired for medical diagnosis, early iterations of major models incorrectly diagnosed highly specialized pediatric cases at alarming rates, despite producing linguistically convincing output [cite: 4]. In these scenarios, the algorithm fulfills the psychological job of providing a decisive answer while catastrophically failing the objective requirement of safety and clinical accuracy. Therefore, the JTBD framework must be coupled with rigorous, domain-specific guardrails, factuality metrics, and safety guarantees that sit securely outside the user's immediate perception of progress [cite: 4, 35, 40].

### Emotional Complexity and the Human Element

Furthermore, focusing exclusively on functional jobs to be done can occasionally over-generalize the emotional and social context of a user's goal, stripping away vital nuances required for ethical system design [cite: 42]. For example, in healthcare, seniors may hire artificial intelligence companions to alleviate loneliness, or families may utilize them to manage bereavement tasks [cite: 4]. While the functional job of providing conversational engagement is fulfilled by the machine, relying solely on an outcome-driven lens may obscure the profound ethical consequences of outsourcing genuine human connection, empathy, and vulnerability to algorithmic proxies [cite: 4]. 

Generative models are remarkably adept at fulfilling low-acuity jobs where fear, physical risk, and emotional vulnerability are absent. However, when complex psychological factors are present—such as in severe medical triage, empathetic counseling, or high-stakes financial distress—the JTBD framework does not naturally differentiate between tasks that *can* be automated by an algorithm and tasks that *should* remain human-delivered to preserve dignity, accountability, and trust [cite: 4]. Product engineering teams must exercise strict moral judgment to identify workflows where human friction and direct interaction are actually necessary features of empathy, rather than mere inefficiencies to be optimized away by an agentic system.

## Sources

1. [Indexnine: AI-Native Product Management](https://indexnine.com/insights/blogs/ai-native-product-management-how-to-evaluate-build-and-scale-trustworthy-ai-products)
2. [Reforge: Gen AI Product Strategy](https://www.reforge.com/blog/gen-ai-product-strategy)
3. [Christensen Institute: Why JTBD Theory is Helpful When Evaluating GenAI](https://www.christenseninstitute.org/blog/why-jobs-to-be-done-theory-is-helpful-when-evaluating-genais-use-in-health-care/)
4. [Salesforce: AI Jobs to Be Done Designers](https://www.salesforce.com/blog/ai-jobs-to-be-done-designers/)
5. [Wordware: The Evolution of AI Evaluation](https://blog.wordware.ai/the-evolution-of-ai-evaluation-building-reliable-ai-agents-in-a-non)
6. [Maxim AI: Evaluating AI Agents](https://www.getmaxim.ai/articles/evaluating-ai-agents-metrics-and-best-practices/)
7. [Jeff Gothelf: Finding Outcomes for Non-Deterministic Systems](https://jeffgothelf.com/blog/finding-outcomes-for-non-deterministic-systems/)
8. [QuantumBlack: Evaluations for the Agentic World](https://medium.com/quantumblack/evaluations-for-the-agentic-world-c3c150f0dd5a)
9. [The Context Lab: Non-Determinism Problem Evaluating Agents](https://www.thecontextlab.ai/blog/non-determinism-problem-evaluating-agents-reliably)
10. [HSS Journal: Generative AI Skill Architecture](https://journals.prz.edu.pl/hss/article/download/2228/1612/)
11. [The Decoder: Microsoft Study Shows Professions Affected by AI](https://the-decoder.com/microsoft-study-shows-these-professions-are-most-affected-by-generative-ai/)
12. [Visual Capitalist: Jobs Most Exposed to Generative AI](https://www.visualcapitalist.com/ranked-the-jobs-most-exposed-to-generative-ai-according-to-microsoft/)
13. [GeekWire: Microsoft Study Jobs Impacted by Generative AI](https://www.geekwire.com/2025/microsoft-study-shows-jobs-most-and-least-impacted-by-generative-ai/)
14. [ResearchGate: Generative AI and Transformation of Workforce](https://www.researchgate.net/publication/404427694_Generative-AI_and_the_transformation_of_workforce_A_job_postings-driven_analysis)
15. [Preprints: Interactive Installation Art & Affective Computing](https://www.preprints.org/manuscript/202408.2106)
16. [Medium: Ephemeral UI in AI-Generated Interfaces](https://isolutions.medium.com/ephemeral-ui-in-ai-generated-on-demand-interfaces-81dbc8cd4579)
17. [PMC: Active Inference and Learning in HCI](https://pmc.ncbi.nlm.nih.gov/articles/PMC11979188/)
18. [Dagstuhl: Future of Human-Centered Privacy](https://drops.dagstuhl.de/storage/04dagstuhl-reports/volume15/issue06/25261/DagRep.15.6.84/DagRep.15.6.84.pdf)
19. [Frontiers: Artificial Intelligence and HCI](https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2025.1451865/full)
20. [Remote Sparks: Product Strategy Frameworks](https://www.remotesparks.com/product-strategy-frameworks/)
21. [Overcode: How to Come Up with an App Idea](https://www.overcode.tech/blog/how-to-come-up-with-an-app-idea)
22. [E-Business School: Solving Wicked Problems](https://www.e-business-school.com/solving-wicked-problems-when-to-apply-design-thinking-instead-of-agile/)
23. [Fuselab Creative: UX Design Guide](https://fuselabcreative.com/ux-design-guide-2026-create-winning-strategy/)
24. [Edward Betts: Minimum Viable Product](https://edwardbetts.com/monograph/minimum_viable_product)
25. [Notion: Bob Moesta on Finding Work You Love](https://notion.notion.site/2edefdeead058052bda4ea3834e992f9)
26. [DTS Partnership: Best Practices for AI Automation](https://dtspartnership.org/wp-content/uploads/2024/09/DTSP_Best-Practices-for-AI-Automation-in-Trust-Safety.pdf)
27. [Liminary: State of Quantum Report](https://uploads-ssl.webflow.com/6523f13a748909d3e1bbb657/65e6d35be62d08db2e7b3b71_State-of-Quantum-2024-report.pdf)
28. [ResearchGate: Early Adoption of GenAI in Education](https://www.researchgate.net/publication/385921503_Early_Adoption_of_Generative_Artificial_Intelligence_in_Computing_Education_Emergent_Student_Use_Cases_and_Perspectives_in_2023)
29. [Scribd: AI in Supply Chain](https://www.scribd.com/document/929023414/in-supply)
30. [Liminary: Lenny's Podcast with Design Leaders](https://on.liminary.io/c/lennys-podcast)
31. [Concepts: Agent System Engineering](https://concepts.dsebastien.net/history/)
32. [Purdue: Understanding First-time User Experiences](https://hammer.purdue.edu/articles/thesis/Understanding_First-time_User_Experiences_in_an_Educational_Crowdsourcing_Platform/22639456/1/files/40209334.pdf)
33. [MLOps World: Testing AI Agents](https://mlopsworld.com/)
34. [MDPI: AI in Journalism](https://www.mdpi.com/2673-5172/7/1/39)
35. [Medium: Stop Talking to AI](https://skooloflife.medium.com/stop-talking-to-ai-9e12f46212d8)
36. [Tableau Blog: Rethinking Analytical Conversations](https://www.tableau.com/blog/tableau-research-rethinking-how-data-workers-revisit-analytical-conversations)
37. [Austlee: Chatting with AI Interface Design](https://www.austlee.com/glean)
38. [NNGroup: ELIZA Effect in AI](https://www.nngroup.com/articles/eliza-effect-ai/)
39. [PMC: Psychological Effects of Chatting with AI](https://pmc.ncbi.nlm.nih.gov/articles/PMC11217544/)
40. [Dokumen: Creativity in the Age of AI](https://dokumen.pub/creativity-in-the-age-of-ai-toolkits-for-the-modern-mind-1.html)
41. [ACM Interactions: Co-designing AI](https://interactions.acm.org/blog/view/why-codesigning-ai-is-differentand-difficult)
42. [ResearchGate: The Human-GenAI Value Loop](https://www.researchgate.net/publication/392186320_The_Human-GenAI_Value_Loop_in_Human-Centered_Innovation_Beyond_the_Magical_Narrative)
43. [ICSE 2024: Conference Program](https://conf.researchr.org/program/icse-2024/program-icse-2024/)
44. [Politecnico di Milano: AI in Participatory Processes](https://www.politesi.polimi.it/retrieve/93b892ea-b10d-40b4-b49a-675116a3b511/2025_10_Biferale_Tesi_01.pdf)
45. [Medium: AI in UX Design Guide](https://medium.com/commencis/the-future-of-ai-in-ux-design-a-comprehensive-guide-to-integrating-artificial-intelligence-into-2339f168dd56)
46. [NNGroup: Prepare for AI in Design](https://www.nngroup.com/articles/prepare-for-ai/)
47. [NNGroup: AI for Design Workflows Course](https://www.nngroup.com/courses/ai-for-design/)
48. [NNGroup: AI Agents as Users](https://www.nngroup.com/articles/ai-agents-as-users/)
49. [Reforge: Create Your Data Plan](https://www.reforge.com/guides/create-your-data-plan)
50. [Scribd: Market Sizing Playbook](https://www.scribd.com/document/895445968/Market-Sizing-Playbook-1)
51. [Notion: Rewired Group JTBD](https://notion.notion.site/2edefdeead058052bda4ea3834e992f9)
52. [HCI International: AI-HCI Program](https://2024.hci.international/AI-HCI-program.html)
53. [Stanford HCI: Evaluating Computing Systems](https://hci.stanford.edu/publications/paper.php?id=498)
54. [ArXiv: Human-Centered AI Shift](https://arxiv.org/html/2501.18948v1)
55. [Medium: CHI'24 Preprint Collection](https://medium.com/human-centered-ai/chi24-preprint-collection-hci-ai-0caac4b0b798)
56. [ResearchGate: The Impact of AI on Jobs - HCI is Missing](https://www.researchgate.net/publication/388633933_The_Impact_of_AI_on_Jobs_HCI_is_Missing)
57. [Visual Capitalist: Jobs Exposed to GenAI Analysis](https://www.visualcapitalist.com/ranked-the-jobs-most-exposed-to-generative-ai-according-to-microsoft/)
58. [Moveworks: Transformative GenAI Use Cases](https://www.moveworks.com/us/en/resources/blog/transformative-generative-ai-use-cases-examples)
59. [Medium: Map of Generative AI Capabilities](https://medium.com/@minparkk/what-generative-ai-really-do-a-map-of-its-core-capabilities-and-interaction-design-patterns-31f2361efc41)
60. [Tribe AI: Generative AI Applications](https://www.tribe.ai/applied-ai/generative-ai-use-cases)
61. [RTB House: Examples of Generative AI](https://www.rtbhouse.com/blog/16-examples-of-generative-ai-applications)
62. [Francesca Tabor: UX Design Without Designers](https://www.francescatabor.com/articles/2025/9/6/ux-design-without-designers-how-llms-are-rewriting-ui-in-real-time)
63. [Medium: Rise of Autonomous AI Agents](https://medium.com/aura-vc/investment-thesis-debundling-the-market-landscape-the-rise-of-autonomous-ai-agents-ae618e5ff07e)
64. [Wharton: Evaluating LLMs for Clinical Decision Making](https://mackinstitute.wharton.upenn.edu/wp-content/uploads/2026/02/Evaluating-LLMs-for-Dynamic-Multimodal-Clinical-Decision-Making.pdf)
65. [GitLab: Jobs to Be Done Framework](https://handbook.gitlab.com/handbook/product/ux/jobs-to-be-done/)
66. [Sequoia Cap: Training Data Podcast](https://sequoiacap.com/podcast/training-data-noam-brown/)
67. [Salesforce: Levels of Determinism in Agentforce](https://www.salesforce.com/agentforce/levels-of-determinism/)
68. [Kanaka Sabesan: Product Management Constraints](https://kanakasabesan.com/category/product-management/)
69. [Salesforce: Agentforce Legacy Guide](https://www.salesforce.com/agentforce/legacy-guide/)
70. [UberConf: Microservices for Agentic AI Integration](https://uberconf.com/speakers/rohit_bhardwaj)
71. [Salesforce: AI Lessons Building Enterprise Agents](https://www.salesforce.com/news/stories/ai-lessons-building-enterprise-agents/)
72. [Parallel HQ: What are Jobs to Be Done](https://www.parallelhq.com/blog/what-are-jobs-to-be-done)
73. [User Interviews: JTBD Framework Guide](https://www.userinterviews.com/ux-research-field-guide-chapter/jobs-to-be-done-jtbd-framework)
74. [Perspective AI: Async Customer Interview Tools](https://getperspective.ai/blog/best-async-customer-interview-tools-2026-8-platforms-compared)
75. [User Growth: Jobs to Be Done SEO](https://usergrowth.io/blog/jobs-to-be-done-seo/)
76. [NNGroup: Personas vs Archetypes](https://www.nngroup.com/articles/personas-archetypes/)
77. [RapidOps: Top LLM Use Cases](https://www.rapidops.com/blog/top-groundbreaking-llm-use-cases/)
78. [Lumen Alta: Top Enterprise LLM Use Cases](https://lumenalta.com/labs/lessons-learned-the-top-10-enterprise-use-cases-for-llms)
79. [Forbes: Real-World Use Cases for LLMs](https://www.forbes.com/councils/forbestechcouncil/2024/03/07/successful-real-world-use-cases-for-llms-and-lessons-they-teach/)
80. [Evidently AI: LLM Applications](https://www.evidentlyai.com/blog/llm-applications)
81. [Content Gecko: Businesses Succeeding with LLMs](https://contentgecko.io/blog/case-studies-on-businesses-succeeding-with-llmo/)
82. [Reforge: AI Product Management Artifacts](https://www.reforge.com/artifacts/c/ai/ai-product-management)
83. [Reforge: Category Atlas Product Strategy](https://www.reforge.com/artifacts/c/ai)
84. [Reforge: AI Strategy Guidelines](https://www.reforge.com/artifacts/c/ai/ai-strategy)
85. [Reforge: Jobs to Be Done Journey Map](https://www.reforge.com/artifacts/c/strategy/jobs-to-be-done-framework)
86. [Kevin Holland: Philosophical Principles of Data](https://www.kevintholland.com/tag/product-management/)
87. [ArXiv: Evaluation of LLMs and Non-Determinism](https://arxiv.org/abs/2407.10457)
88. [ResearchGate: Metrics for Human-AI Decision Making](https://www.researchgate.net/publication/402859688_From_Accuracy_to_Readiness_Metrics_and_Benchmarks_for_Human-AI_Decision-Making)
89. [JMIR: Evaluation Frameworks for AI in Healthcare](https://www.jmir.org/2024/1/e46407/)
90. [PMC: Metrics Gone Wrong in AI](https://pmc.ncbi.nlm.nih.gov/articles/PMC9122957/)
91. [Jobs To Be Done: Outcome-Driven Innovation](https://jobs-to-be-done.com/outcome-driven-innovation-odi-is-jobs-to-be-done-theory-in-practice-2944c6ebc40e)
92. [CMU: Designing AI Products Challenges](http://reports-archive.adm.cs.cmu.edu/anon/hcii/CMU-HCII-24-104.pdf)
93. [Tech Lead Journal: ML Engineering Best Practices](https://techleadjournal.dev/episodes/203/)
94. [Purdue Thesis: Crowdsourcing Platform Onboarding](https://hammer.purdue.edu/articles/thesis/Understanding_First-time_User_Experiences_in_an_Educational_Crowdsourcing_Platform/22639456/1/files/40209334.pdf)
95. [USENIX: Privacy Engineering Workflows](https://www.usenix.org/conference/pepr26/program)
96. [The Open Group: DPBoK Community Edition](https://theopengroup.gitlab.io/dpbok-community-edition/html/DPBoK-CE.html)
97. [Cambridge: Conceptual Design of ML-Enhanced Products](https://www.cambridge.org/core/journals/ai-edam/article/towards-the-conceptual-design-of-mlenhanced-products-the-ux-value-framework-and-the-comlux-design-process/23D4B3A9C29036F66C89B1C5A7F2D5DF)
98. [Computer.org: Enterprise AI Agenda](https://www.computer.org/csdl/magazine/it/2025/01/10893856/24sGpV6Fl1S)
99. [ArXiv: AI Interface Evaluation Framework](https://arxiv.org/pdf/2602.22343)
100. [Medium: Workflow Encapsulation HACD](https://medium.com/design-bootcamp/workflow-encapsulation-in-human-agent-centered-design-from-doing-work-to-governing-outcomes-de416b27226d)
101. [Emerald: Permissibility vs Feasibility AI in Service](https://www.emerald.com/jsm/article/38/1/87/1214738/Permissibility-vs-Feasibility-AI-in-service-from-a)
102. [ResearchGate: GenAI Value Loop in Innovation](https://www.researchgate.net/publication/392186320_The_Human-GenAI_Value_Loop_in_Human-Centered_Innovation_Beyond_the_Magical_Narrative)
103. [SBM ITB: AI-Integrated Chatbot Design](https://journal.sbm.itb.ac.id/index.php/mantek/article/download/5883/2196)
104. [Wharton: Diagnostic Pathways and LLMs](https://mackinstitute.wharton.upenn.edu/wp-content/uploads/2026/02/Evaluating-LLMs-for-Dynamic-Multimodal-Clinical-Decision-Making.pdf)
105. [Politecnico di Milano: AI in Participatory Processes Thesis](https://www.politesi.polimi.it/retrieve/93b892ea-b10d-40b4-b49a-675116a3b511/2025_10_Biferale_Tesi_01.pdf)

**Sources:**
1. [Link](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGn6Vc-WNfh9avdrD8h9QEpAsTibApKWleSgPov9vt7jBWh19Fw2zReuWQxFuD3WqDXKBYNEP8saaBDsQYg_whYg3fdoUDogjsLEl9UIUvz_9Wu-k7SIsPCJaeMJbCIZfm4zs5aGBDCsjAuS6aV1p4mEQ2Ow1dZNqcEfp7ZEucJBmN-ou-1WWJ-APg7BAEmkRcvsQbrJjvQAA9fMvMP1EUmGdNMjqpNTbQhau4jnKI9)
2. [wordware.ai](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQH0wDXXmffHd2XODsto1ojLKrRKeJ3mBzUX4nhVeBsmw-0DfaiG3nDpyzl0zQ1dgoC-LAIGj-bk6nDWm4B3x27gCzZnJPLYcIH3p4BufmVhCxliKAVnY5OSINTbuS6HOcy9IgjCWmTfnpptei03da1E7ymQOqyFpvifuKvhSPmPUk1RY-gXxw4vK1hS8-Go2b5ugg==)
3. [jeffgothelf.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHo4NRqhYRbS33lne_si_avadvgdxLpMzSHxiXuVjuFTG0nyzMvFNfYxlI4h1DJzUkSJ0nnL5aiQmnxKjCjUgdSJHMmJqoOpmUgrIybaRunk7T9NWpuGTtnRtdTX2IVUm4p1N0wQHo9HJy7m2ZMavU0tmhj4_5ilzQkG-0fgE9qdl23)
4. [Link](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQF4eH83v3sf3sJOwQ0nu5LPyJBOBVVb3R5Q3HNxQx2_fctJfsHzy9wnSzgFOkEYFmN_QXXxyoABepM5nONPFNvLN_xd1jXWHZBEJYVv4bPZ4diCua05HiC_nQJ5p5HcsoaaumTOg12adZQNXqz_doCLxW2J99knEEAykxcJlx6fzuTa_T9uzSE3uML7fICnj8ajzB0mYKTy0Ar4_VKoxpHxHYTKnhwoXJrbqddjGdl59Q==)
5. [remotesparks.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHV8369X_7Z-b1q6ZbewlKdRX2qV7-95FgXNr_IleyUN0PjBy35Ty_Xz30RbkM0ldMA8t6WV3EPsN4BqUX2iAhykZw_8aMBfOn11xm0lXS30BbLx26Cddvof61VMXAD_81UkXVAyUlUyPyZszb9D0s=)
6. [overcode.tech](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFngPRlVObFtPEAAqXXR-yJjwGtCJJo5qV6hcVKv4yl7blEKPrBHw1Qm9cABMyjsCSzYs-0ufIRPT57WtEqG_N5iJF5yYAhb_WnSR4yqj5n-FXF5hbokNVBXKC0UmzVPuP7xdkuMP4YOzawaP_fLjJxSrLrzw==)
7. [e-business-school.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFWob8SYIgTtsHJOgJjM6hk6tSf6J3_rvfJ5zyKc6FijXdtx8eRuEtdBcG0G7_mTYDMbo5WhqL1fB40vfs7l3wqDqv_ZkJ3KMR2Vm05djCgT-7C11FljppE74wNVMPSEPhJDpCPuEuMUsr3XmzxfF79BiRJizofiXyGKrha6h-mZ-T3mWPVmi21SzU2zfuh1JuXB3-R5oIQO_gJtTPRuC8=)
8. [medium.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQG9t32gzO-FHS8s4EZLJBWZn21U0YuhoaoxCH_FJmoEUrkYUt2GnEsK2P86s7heTbwtapsF3fCUuKYdaFSPQxTIAUFiv2BGYCIn6VF3AIyQr-ZR4R8sZrf6iIUGbglr7mNjd6LXCjsWBvyhnWNhTp_FW9GBQg==)
9. [usergrowth.io](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGkw0hB05cvpEa9ShaMQgNt94KOGAXJJdi1MXFE6mmbgE8rr_llwBiAMfwkuhtRruDLnjVs3A5UnSYy7T_jvVkEkz2mJbZfgKYETp0dEJF2zSe1oRCpgVnHjM5yGY3nij08aGTG3Q==)
10. [parallelhq.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQExKTQoNTQCnAxOOv_9VhmWf3DuUxRTPVe_CAygipIC_T_so-VfjnsyaLIkQDvehWpGgE4A_F4eibyaVMxhLuCHtCa4lOH3-HbF0wB4M7AGU60ZQUFZCm_cwb-F7_BJgT9xDWBfIy8c1QwS0Xha0Q==)
11. [Link](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGeWc4c8XnsJcWmx0XS0Z-f_jQ1rVuhojD00DGkXLsgj79qWMGlwXkTX9YkquwQK_M0zigrHpcuSGIKx6FL5rIjDqPyyh_mc5Icnd2cQWcijQ6EUNtDLA9S8M6hBK4XEXgUi_v-Um6nn5Qb)
12. [jobs-to-be-done.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEl5kbdOPBj7VTmwMLp7h3LTBGJq-Ez__XuPvWUYKH-B3jAbWELOlrMExByJmDi7kYalYFflZ0YRFyu8Xz-qPnwt-GEZYHqVpTtXLndEFG-AH1CrZNdrjgTwAu804yHYPAKgxCUlyYIrQn01C7zrSMSDxU_niCo10YVEW4GSRmBaoEVPj91QpqBKLKcEwJxZYQdpBp8zGJezric0qMLoJQnEDQ=)
13. [cmu.edu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGFjxP4TwvCXNOE171dP796cwRg-J_p9rq2kZtKq_N8HclcovQbOdybZFKgH3cwA8Id9cKCGYpNBmaj5IilH-gA0ahxuppGv-bKFugW-rzMOVhBVO8Z7yTMIDkcIowDJXVgKrzzlvizeS0nMw_0-EiDo_zKGF97UU-Z)
14. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGH9695BS0cKPHLd9N97gOaTqIORRwK75cDLdAkbY3_yuncJeppHr0loKMwclJ9Sv0FOOQdTR-4IVc1ZGKX8OwnHX8iQI4nEBn7SZ2umYhsjvU6Sz1yfAUIUw==)
15. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHZq1LPbET7zbkLFHjPzfOMzctvPpGgFsnl9Qfn4HOZxTcql1ffm3qI0YcG6fmkn9pf1BvSryVEHyn9hvlSfH7nxEWhznyFpqLPSlck-6okfO05uLd2EOMnFnEpOBJikIbnfeg9TmIFS3iQB_5bqm0PRQCe53ly5rsI9ENkxmgUvh0q1Opk29FfdkCR1vobtOI=)
16. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGq-H5vXFWZ6GEdj6qOZoVV9aT9SKYIFc_Bj7-wmCv8qNs4FIKDaLnG-Ns3RPn7SbhgvIhwZpAQnGmXjz2gcmPHKpEL4NTT53DoEjXgnQokV8VeGOkNzvxlEEYtlZePdao8HSVIkYIwieb_ZM5q0TPkTvVAZuuHZDxXJO1h1ikmdJxcbuDUzAZAZTJp-lw4oF3TCHxxgX36va3LkXpVX2y-THfpQ-3h_7npgQ4NJ2EGrtk9T-_d4mg2cpzPkh8p)
17. [the-decoder.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHfUxuxb6eM_uvAojdLcct7YE9yJC2V4wCxmr28zxfZaQlF9YHiRa13245htuUWbLKzPTADO4I6sNSb2LNpiYk68Cva1aR-9JKs-I5Bsp2mJIzCJ3TJyIJ7osk8Q3K9X2RrjUM0W92TjVV-7aSZfB-DJaX4isfgtgFw8cAQDe6Uq7nGjDmeDcwU4Ds_YPvJuIB2g58K0aQKsuo=)
18. [geekwire.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEkBvgvdhDUspwQKMrUcgHKd8g06dAhIqR992Hu15KJirVFfZGY5eEi7zVKW_CsjuR_5Aw573pMTMTnSUaGCH_5xb0B_N66EIfiekO_xoIBQZ36gEvIsQGrMKwUz4eaMsLmgzKZU_Pwdt8nMnE-J89a_F1Gth49jdlAKMLcwsbeRonheKLtFGxRALEPT49Ivgesmza0gSUDgA==)
19. [visualcapitalist.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEVPmd33ZSMQdMM3bGDPugJNQ2HA0RtOhV11hKYWS8u-ZjFFNABl0wXVwJ4NDZrZnEaYq-QSiNELf9Gzse_16zbqyIv37ezPm0HMEhFfrMR-OdgLo0CkWDET4hR7ZCbGBlh_PPrQ0IG4_yKX5IdAZXH82jYG2xaXUZoEqXRPNpUG50x2lnH3mRvVG73VMuFQsp3zIpfqFAOJsX31C8=)
20. [salesforce.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQE5cSU4zU-oT_h98pcs-4mMQzk1zwXOgzmoW3f0UKE-DDzmfSW4NSgKEZFm4FXKH-CfVfosvQrk7LH5qJXLU8VN2BpcGrNdDxbLbRbsOs-qB8wp7_UhwvESodnWJaXvIFEnaelGf2upwoBVrtDWdEQ15eeGw1QAo_0HenFvvtGQyCbp7v4=)
21. [emerald.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGctrFtGk3rJTRr6Yxpa1NStqhivc-PMBplOhpQDaNH0Jn5KfG-5T3c4j0gkDGlewW616usMAEYGCJgZEZLYHiJj-QmWQtQ9dX-HnwIetG05btyGHmEUjJRyMm8mHJehPfNFWNGUKf5V1SC0KA_YfvVuFuoWMFy1leTZPOUJyDO50gnD1dBTj2z9poq1b689u3Fxz-URJONVd1F5dc=)
22. [medium.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGhMWiaGdjt-laXfwdo1izUSpFA99OwzwUifQM34fIj9XvUDYUo1z9CDW3SPUM1KwIk1l6TEXmYsGAg-ScYFDRSma6bCMUjrEuiEkj2XyN5G004XBcO4swbde86lQp9iwAKbHQlq5xIPNBnOEJ4d6c4svCdcBrLyJyXEW7aeMGjAL3JHPQ4FOxjgXh2wYwn2cTwRrOKyyD6_lei-GQH40KapIg7SA5Kr7AzSgu7IYUI8TyrDnm5BbEUcqBCIA==)
23. [notion.site](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQF2Updj6xEbFZlChjhQJT0zCEaEECah3qUNSacrovYRG7Ok-6tpxzmDlJFo_VGTv2YN0VIGJ0qzsSmxXZToYhNo6NHsXxkvmAZENrSuSwghSPfxAnn4XNuw2xYP6XALyD0yT1Lzg8UAXlGALfJfzakI4Q==)
24. [lumenalta.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQE1Fe6OYbukzsBx0h-vnBFNM9qbOeuuHn4CdEXyJFVjVE_8nIQFXi3WzGsXkEGjTT0Rwh024ZsjBcVg-EO3vL2-YfZeu9mOX1vHPRrZJJrpe9CELNpkDwWhjjnQ7VNHt_P5McqqYHizx2FvSdCzSSwZ5WT5C7j2jrZsSUG3rK2JaPP0p7KN9jBr1w==)
25. [forbes.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGZyFTkgK4pFCeP08rYT5ns00rgPBA8Bsn8aFajzRjXTtYzdfj3osGnG2yr4xE1HTa3zBeJTXA8QcEZiyTI4rDDXGiHHWNYTwFB5jgQ7lUIIcWcaFWiMH9Y1VtRJrFWfZT0pM73gXZfA1znUWp1Xki-mOu3niAN8ZfMUieTo9M2PSH9k78w1TCo9lpVdJDTp8IjqFa_GR1pXgAgBfTQJx008p0zrB0GSNrVYTK1col9q7f52A==)
26. [contentgecko.io](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGeBMufMcJVizmROVdj1ouLcw3iUSUrJBvmMXsdDtTwTk2oP5LvvLGQQH8ROkGHsbJxyqWs-rj2rQ3iTqmDiTSfnNaPq2YgGg4xHGgqezfNp_IqmIzyjOv03ZRzfpFY4_1yHiie4czPPO0kkpaUvzc_iSNQdZ15WKJulzSMNVdye6mI7w==)
27. [tableau.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGuyjv4lsJFcU9zI6HNML5Roy48t8i5j55MGpGf4j5DXEQxj5aC5dZYYFnMH6xKuHtDrxw1Lom_9-M4sm4D4L1vhDpdEUN6dnuSFAketjClzeGwLP1Wi7NBLOI7FjgKW_aZnxEnsULPKtUaOBmjn9x_ODGrA_nF__7DbhWXCT9WsMl4n2CfiH2Zy6IapLZ6noC-dTU8EGkeJdgptHThD_PC)
28. [medium.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEOdZbc3xZVo-3zB3bQUwffvZi6qvDCVGBtvkafCH3fOszCW28_pj8NyQnsGwxl1wdOz_m2pIFuKNyPw6mFzGZv2LuRQQNc4SmDgWVNuj2UYpUGKOSROyuRcgSsZ0lyvX5LtOnUCrZz0wxu-9fcNtHofm6LTyqueJpY_RbjnjbGRIHF9Og9deCHAVAhyxL7EAdMdQ==)
29. [medium.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHIWlDpFHNHC86DjHUfshWykgHLr-4fLTF9szw09wTHGzbN1CN5UGDPXTM4tf-jY_RhtDUn84_Ox-JVwD0MqTcoWzRPqw3Er-EKd2dHxq0BGCZKkaJUDQrZ_yLMkZ0poV-9XX-d_LPQ60uFycvYgFVtswGp14aeNnu4odI1UPWkEpCq3Ha7gDzgiby1jwpiGsLQlP37uacnt5MiLClEH6HwaZBHPQvRJwnfX3I0FZ0U4jyjn3mwps2iSV1dI7Caj58r1Q==)
30. [francescatabor.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFkPFnO2DdEkSBHe3MtUp95ouH-pjmDXH9xK3AI59N5dlF-8tRGJIJeVf7-vKZhavRqfc0aM3D0mfWa0Q0LG2J1xi3ZbzJtFjGGqpE_zQIgemnPyhRy0h5iraboea3kblhHLmxJ3xXmzXzGQLbhEg84AvZ9QvLlUSyQbQOnwN3j6YTK2zbGeSOjPg2Pi-rWGVo_4DltktNGSP5f472HYPBrTaLNdqZNEI_e)
31. [kanakasabesan.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFzywa1CpPClJ3sCM7E82SiQmg6nDCTqmMnQMvBPWgHb2MlQep-Hzmv1UkAyG4uHllZx7Y3jOeND3_PTgWv4faxtQ05Fh4rrxfx_ERvcR3gTsKKd4GA3JTjQROoMlZAS2aaW_0tOHZJ4khLjvw=)
32. [salesforce.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEhaEpuk_WE8aSOR-dows-za79ZyeHVdbi02loUllzmHf-6Q-Zk-uUA3TxeldsgxSd-PIOT1WCta1-2GY1JYdWriLrD4YDBghx0koBqxg64Y07ss0o5-44cwJxoXcQtXlqtOgSELeRlJjGFsuopft6ksgE=)
33. [salesforce.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEMPPwKtNonvmwTFtcZjdYiojCmxTEWTWwO8vDW_Dbw6HyKLpXP9AD7W0s0iKZoY8RkSH8rfXMh4JsHdRHnMLTtjxMkUxVMCSWmGDFFt_Gn3HKise4iRwN1vRG1M7Q7ZnwEqg0X7GUnotg=)
34. [nngroup.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGOhJB14GKAsEyODeoQlAXAUx9_kqxK_gZDlHg1XAHJ8oT-qoBTYKthNGCW2ux6nOPlWU6utsCiMBqktIl0aJZ_FIMI51Qi4Z82UH3pxVPCjm4ppYgwnXLZ2anwRhjEkLX6Reyx5F_uUrXz)
35. [getmaxim.ai](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQH2uzaBt_ExR7NFkSWfoj-DMeMLvZJzrey251T4En7sy11qPhEXv6tA7Xzf-9Bmb8zL_BGqgx6a2ZSkUTBIaQOVTxuuJMgRzykRQk-l0_GVcoFcvPSCfwlT_obxelR8cJkM7WGiaArTqo-mpcAR_pQPzu9fztWP7X_56DS3kAnyyoBpdqDl64c=)
36. [thecontextlab.ai](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGliqPT4D6vmibuMeE_Eq2lf26L1bBXZebDCkUN8s-RzIDYAdAV7vAm-4V5A6OdnWlmBabWqE4y-KXQumIINKQjm6EdorBEYw2-da1O56rVNyu3KDgou-lLFwLnJYYrDnff-8RW2j2xHtv6cRO4s8cPkDdFvtef104b4BUjh07dFoJSmxvXZc9CQ2A=)
37. [nih.gov](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQG77JUkPOWX8bI3MbvaGyl5H9pvyBQjVeMSu2D6K8F936v-9iBhuA65sFdOASS36h_ZjnbfyFVh1hzzqnj-WFWpUpk3mwLppS0FZP6KPGNcXxU7D7k3igMbnLvwrdd02KCc8VB3MYpm)
38. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQH-WwlNDBnKYRG602Yvx3mxrikKCxLgQiWj2kwgxrVw-rF6Bgnrog6GRDszTkF0RZvyO0gfNpdr0iFVxYogs7Gn-idnXgoZzHOM3PmjKTLeSfdmQ-ZHTG_4Ytu1dHIZR98Trm_LK48W4LelJV0YX9b61tSwQjFDCmkwzMh5M1o1QTTpoNUWsWh0QdZT0AlPkeV96aLzo-_ng9UrzUkb8Ud6VkXWhiebsURAn1tBwIPGTRdlGdYLgc9I)
39. [liminary.io](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEVZFBdxrzoPOdgOAgE6A-MPaRk1pijcd_Gypn673DkC3z82ujx94_l5w-1Ssmj-ps9kC38I_1I_EMTl3Bv6eEQoSrLCNESjNfR8qziCvDqCZs5hkcX2FmGpxOneWY=)
40. [medium.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFQ9Ed9HYTLPi_2uJa8GOAWbrA_lroCsCJT7ty8SkgUlF4EmWLE3R-b-GBKfcUe6eg5ywOjxMcwCJi45xrVOIWXQlzuijnwV_df-0dLsYo_G7j9ixHBO2_76R-kNixv1BeHKFpFWrv-nsFchiz6HcykKUEBhL_1-Zc7rJhL5HiOjE6cPgM=)
41. [upenn.edu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGNji3pybYeD4GQ7iKMP7Dkeq0Tqv1mEq3jdxl2q81g-L4osC4cuYVTIrUe7hhz20LKGxiQ2X0Pc9fYuRRvhZZyp6vj2kIlq-5YS55ZK6b2k6Wyp2v2XtI5o3-AiWms7pMx6AP2EIzGNa_D391Vj_XfsWS4_kxUQTvBRJEcpFyv6Qi-9siTLpra141dVi9yp3u57jjL9zmlnse3_wmRRxLifD8gb5WhMNFoY4ecWZcyzcmn0jt57tISt--aQq4=)
42. [userinterviews.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHTB2XZzj7i6752BUCMXBdWj_JmRoizVop_3u3wKAV8JTRVl97CoSFzjQ9VGKtyZn2WjrvXH5yIPKIIPIEZHOs9BxQpintjCd139s3elAPRwQ3oH3SvhZskCbrntEmYQWSKLG2WBi2nYI6JNZ_nU13I5FISi9oXy5fX5-1HQKxx_ItDrV4UzCv111rhS6N66m4yxXo=)