TOPIC

Design Frameworks

A subject the collection covers, read through 68 synthesis notes.


View as

How should agents interact with humans across their relationship lifecycle?

Explores what design principles govern human-agent interaction quality across four stages: initial setup, moment-to-moment interaction, long-term adaptation, and failure recovery. Matters because agents increasingly affect human work and autonomy.

Explore related Read →

How does GenAI's unpredictability break traditional lab study methods?

How does GenAI's non-deterministic output destabilize the controlled conditions that HCI lab studies depend on? Understanding this gap matters for designing evaluation methods that actually capture how people interact with unpredictable AI systems.

Explore related Read →

What dimensions structure the design of AI writing assistants?

Researchers asked what common features and choices appear across AI writing assistant designs. Understanding this design space helps researchers and builders talk about assistants systematically and spot unexplored combinations.

Explore related Read →

Do formal language prototypes improve reasoning across different domains?

Can training language models on abstract reasoning patterns in Prolog and PDDL help them generalize to new reasoning tasks? This tests whether shared logical structures underlie seemingly different problem domains.

Explore related Read →

Why do AI agents miss most of what users actually want?

UserBench explores why current models align with user intent only 20% of the time, even when users reveal preferences across multiple turns. The question examines whether agents can learn to actively clarify ambiguous or evolving goals.

Explore related Read →

Does AI exposure hit low-wage workers harder in some fields?

This research asks whether AI's impact spreads evenly across skill levels or concentrates among workers least able to adapt. The distinction matters because vulnerable workers with fewer resources face greater risk from automation.

Explore related Read →

Is AI creating common skills across jobs or deepening divisions?

Whether AI diffusion produces a uniform set of competencies across occupations or widens occupational divisions. This matters for understanding how labor markets will adapt to AI exposure.

Explore related Read →

Can autonomy controls stay distinct rather than collapse into one toggle?

Does breaking agent autonomy into separate UI controls—budgets, approval gates, per-action modes—help users maintain meaningful oversight, or do these patterns still suffer from habituation and oversight erosion in practice?

Explore related Read →

Do UI patterns for verifiability actually build user trust?

The AI UX Playground names specific interface patterns—citations, chain-of-thought reveal, granular consent—as trust-building mechanisms. But do these patterns actually change how users decide whether to rely on AI systems?

Explore related Read →

Should agent trust signals stay separate or merge together?

Design practice splits agent transparency into three patterns—observability, confidence, and citations—but it's unclear whether keeping them separate actually helps users calibrate their trust correctly.

Explore related Read →

Should agent control be one setting or several separate controls?

Design practice distinguishes permissions, autonomy level, and stop buttons as separate controls. The question is whether fragmenting these into three patterns helps users understand and manage agent behavior, or merely multiplies configuration burden.

Explore related Read →

How should AI design connect individual work to team knowledge?

Explores whether AI interaction design should extend beyond individual productivity to support team-level knowledge sharing, context awareness, and collaborative workflows while maintaining privacy and seamless user experience.

Explore related Read →

How should generative AI design embed responsible AI principles?

Explores how AWS Cloudscape translates five responsible-AI dimensions into concrete UI obligations for teams building generative AI products, and whether these practices address the core interaction problems generative systems create.

Explore related Read →

What UX principles do workplace users want in AI agents?

A multi-method study asked business users what design principles matter most for human-AI collaboration at work. The findings prioritize oversight and reliability over convenience.

Explore related Read →

How should chatbot design vary by relationship duration?

Do chatbots serving one-time users need different design than those supporting long-term relationships? This matters because applying the same design to all temporal profiles creates usability mismatches.

Explore related Read →

Do spatially-anchored previews stabilize VR geometry editing tasks?

This study explores whether showing users visual previews of candidate interpretations anchored in the 3D workspace helps them reach task goals more smoothly and with fewer conversation turns when working with an LLM-assisted geometry editor.

Explore related Read →

Can AI become a design material for adaptive interfaces?

Can designers treat AI as a core material—like clay or steel—to build interfaces that reshape themselves at runtime based on user requests, rather than scripting predetermined responses?

Explore related Read →

Should AI systems stay collaborative rather than fully autonomous?

Explores whether keeping humans in the loop with AI agents is more reliable than pursuing full autonomy. Investigates whether collaboration solves problems that autonomous systems structurally cannot.

Explore related Read →

Does presence in a group chat equal consent to message processing?

When someone agrees to an AI agent joining a shared chat, do they also consent to that agent logging and storing everyone's messages and photos? This matters because group members may have no say in how their data flows through the system.

Explore related Read →

Can designers shape LLM behavior without deep technical knowledge?

Explores whether LLMs can be treated as adaptable design materials that designers can tinker with directly, rather than fixed components handed over by engineers. Matters because it determines whether user-centered judgment reaches model adaptation early.

Explore related Read →

How should agent capabilities be organized into functional layers?

Explores whether agent systems can be understood as stacked layers—from basic chatbots through system prompts and RAG to hierarchical supervisors—and what design trade-offs this architecture creates.

Explore related Read →

Is design for agents becoming the next major design paradigm?

Does the emergence of LLMs signal a fundamental shift in design practice, where agent design subsumes interaction design the way interaction design once absorbed graphic design? This explores whether we're witnessing a historical paradigm change comparable to mass manufacturing's rise.

Explore related Read →

When should practitioners stop decomposing LLM evaluations?

Explores whether there's a natural stopping point for breaking down LLM outputs into fragments before trusting aggregate scores again. Matters because endless decomposition wastes effort, but premature aggregation hides unresolved disagreements.

Explore related Read →

Do evaluation criteria shift when we grade outputs?

When practitioners build LLM evaluation systems, do they define criteria upfront and stick to them, or does observing actual outputs reshape what criteria should be? Understanding this matters for designing reliable evaluation tools.

Explore related Read →

Do firms substitute labor for AI at different rates?

Explores whether companies exposed to AI shocks replace contracted workers with AI tools uniformly or at varying rates, and what firm-level differences reveal about the economics of AI adoption.

Explore related Read →

Should AI features integrate into existing workflows or stand alone?

Friedman explores whether AI products work better by enhancing existing user workflows (AI-second) or by building dedicated AI-first interfaces. The question matters because product strategy around AI placement may determine whether users adopt or abandon new features.

Explore related Read →

Can task-oriented UI patterns reduce the friction of AI chat interfaces?

Friedman investigates whether adding design patterns beyond chat—like visual intent capture, output lenses, and canvas-based refinement—can overcome the usability burden of typed prompts in AI interactions.

Explore related Read →

Can breaking outputs into fragments reveal more evaluation problems?

This explores whether decomposing LLM outputs into fragment-level functions per criterion helps practitioners spot evaluation misalignments that holistic scores miss, and whether this leads to better calibration of trust in automated judges.

Explore related Read →

Can previewing AI output before submission reduce prompt iteration?

Do generative AI interfaces need to show users what the system will produce before they submit a prompt, rather than only giving feedback after? This explores whether feedforward previews can reduce the cognitive load of repeated refinement cycles.

Explore related Read →

Does generative AI prototyping focus on model architecture or content curation?

When teams prototype generative AI applications, what actually drives their design decisions? This study explores whether practitioners shape systems through technical model control or through curating example content.

Explore related Read →

Do generated analysis UIs really work better than chat?

TaskArtisan investigates whether putting a GUI into LLM-assisted analysis workflows improves usability and clarity, and what trade-offs emerge when analysts need to modify or reuse generated interfaces.

Explore related Read →

Do generated interfaces outperform text-based chat for most tasks?

Explores whether LLMs should create interactive UIs instead of text responses, and under what conditions users prefer dynamic interfaces to traditional conversational chat.

Explore related Read →

How should users control systems with unpredictable outputs?

When generative AI produces different outputs from identical inputs, how do interaction design principles help users maintain control and develop effective mental models for stochastic systems?

Explore related Read →

How should AI systems respond when confidence is low?

When an AI feature lacks confidence in its output, what should the interface do? This explores whether low confidence should trigger fallback paths and how to frame uncertainty so users see it as fixable rather than terminal.

Explore related Read →

How should AI feature risk guide the level of human oversight?

GitLab's Pajamas design system proposes scoring AI risk as error probability times impact cost. The question explores whether this formula can predict how much autonomy an AI feature should be allowed before human review is required.

Explore related Read →

Should agentic AI interfaces require users to specify risk tolerance upfront?

Does designing intent frameworks that capture forbidden actions and acceptable risk before execution actually prevent harm in autonomous AI systems? And can users reliably articulate these constraints in advance?

Explore related Read →

Does perceived control matter more than actual control for AI trust?

Explores whether users feel safe with agentic AI because they can intervene when needed, rather than because they constantly exercise oversight. Tests whether the psychological sense of available control, not its frequency of use, drives trust.

Explore related Read →

How do communication modalities shape human-agent collaboration patterns?

Does varying how humans and agents exchange information—text, voice, or structured channels—produce measurably different negotiation, trust, and awareness outcomes in collaborative tasks?

Explore related Read →

When should human-agent systems ask for human help?

Explores the timing problem in collaborative AI systems: since there's no objective metric for optimal interruption, how can we design deferral mechanisms that know when to involve humans without constant disruption or silent failures?

Explore related Read →

Why do people share more openly with machines than humans?

Does the absence of social goals in human-machine communication explain why people disclose sensitive information more readily to chatbots? Understanding this mechanism could reshape how we design conversational AI.

Explore related Read →

Do humans apply human-human scripts to AI interactions?

Does CASA theory correctly explain how people interact with media agents, or have decades of technology use created separate interaction scripts? Understanding which scripts drive behavior matters for AI design.

Explore related Read →

Does consistent AI labeling actually reduce automation bias in practice?

IBM's Carbon Design System uses light-inspired styling and a popover component to mark AI presence and provide explanations. The question explores whether this layered disclosure mechanism actually changes how users engage with AI-generated content or trust it more critically.

Explore related Read →

Does visible design structure keep HCI expertise necessary as AI advances?

Lam and colleagues propose that just-in-time objectives architecture makes design decisions visible rather than hidden, potentially preserving the need for HCI expertise even as AI systems become more capable. This note explores whether structural transparency can resist the automation of design work.

Explore related Read →

Does comparing machines to humans actually reduce social responses?

When survey items explicitly contrast machines with humans, do they undermine the mindlessness explanation for why people respond socially to AI? This matters because it questions how machine heuristic is being measured in HMC research.

Explore related Read →

Can mindlessness in CASA effects be measured independently?

Does the standard explanation for social responses to computers rest on circular reasoning, where mindlessness is inferred from the very behaviors it supposedly causes rather than measured separately?

Explore related Read →

Can declaring experiments in code make them reproducible?

Prose methods sections traditionally specify research protocols, but live human-AI experiments involve complex technical coordination. Can a single configuration file serve as the actual method specification, enabling other labs to audit and rerun the same protocol?

Explore related Read →

Can language models discover what users actually want from activity logs?

Users pursue month-long interest journeys that transcend individual item clicks. Can LLMs extract these persistent goals from behavioral patterns, and does this change how we should think about personalization?

Explore related Read →

Why do LLMs excel at feasible design but struggle with novelty?

When LLMs generate conceptual product designs, they produce more implementable and useful solutions than humans but fewer novel ones. This explores why domain constraints flip the novelty advantage seen in research ideation.

Explore related Read →

How should AI agents balance helpfulness with staying out of the way?

Microsoft Design explores whether AI agents should proactively intervene or remain invisible until needed, and how to make them discoverable without being intrusive or replacing human judgment.

Explore related Read →

Should agent customization be split into foundational and extended tiers?

Microsoft's design guidelines propose dividing agent customization into identity choices (icon, name) and interaction choices (dialogs, responses). This note explores whether this two-tier framework actually improves design outcomes or represents prescriptive vendor doctrine without validation.

Explore related Read →

Do people interact mindlessly or mindfully with in-car AI?

Do social responses to conversational AI like Nomi emerge automatically without reflection, or do high-quality social cues prompt conscious engagement? This challenges decades of media psychology assumptions.

Explore related Read →

How should UI widgets coexist with conversational interfaces?

OpenAI's Apps SDK proposes keeping the ChatGPT composer visible across three display modes—inline, fullscreen, and picture-in-picture—to preserve conversation flow. The question is whether this architectural choice actually prevents users from experiencing interruption.

Explore related Read →

Do people sympathize with ostracized AI agents without blaming humans?

Explores whether bystanders who see an AI agent excluded in a game feel sympathy for the agent while still viewing the human excluder as acceptable. This tests whether people apply human moral standards to AI consistently.

Explore related Read →

Do people treat excluded AI agents like excluded humans?

This research asks whether people automatically apply social inclusion norms to AI agents the same way they do to humans, and whether this automatic response varies by age or experience.

Explore related Read →

Can systems infer what users want from watching their screen?

Exploring whether passive observation of user behavior can reveal immediate, actionable goals that LLMs can use to produce more targeted outputs than generic responses.

Explore related Read →

Do productivity agents actually protect users better than emotional ones?

A heuristic evaluation compared nine deployed AI systems against 14 interaction design principles. The question asks whether differences in adherence reflect genuine safety tradeoffs or uneven product choices across agent categories.

Explore related Read →

How do teams decompose prompts when collaborating on generative AI?

When UX designers, developers, and product managers prototype with generative AI, do they naturally break prompts into consistent sub-components? Understanding this could reveal where cross-role collaboration breaks down.

Explore related Read →

Do AI agent design choices override model capability in shaping social perception?

Whether an AI feels like a tool or an entity depends on design variables—naming, identity, social positioning—rather than how smart it is. This matters because it explains why the same model can seem fundamentally different across interfaces.

Explore related Read →

Do more social cues always make AI feel more present?

Explores whether quantity of social cues matters as much as their quality in triggering social responses to AI. Tests whether multiple weak cues can substitute for one strong one.

Explore related Read →

How can designers prototype real-time multimodal AI agents?

Existing text-based evaluation tools fail to capture the complexity of real-time, multimodal GenAI agents. What methods can designers use instead to test proactive agents that must integrate speech, video, and screen input while responding within seconds?

Explore related Read →

Does CASA theory still shape how people treat computers?

The original CASA study showed people apply politeness norms to computers. But a 2022 replication found no such effect on desktop computers. Has the theory aged out, or do people now use different interaction scripts?

Explore related Read →

How do interaction signals reshape what generative AI actually receives?

Prior taxonomies describe GenAI interface patterns but struggle to compare designs. Can modeling structured interaction data as an explicit entity between user actions and model input reveal design differences that otherwise collapse together?

Explore related Read →

Why do users respond so differently to AI chatbots?

What explains the wide variation in how people relate to chatbots—from trust and attachment to suspicion and fear? Existing theories capture isolated factors but lack a unified framework for understanding this variation.

Explore related Read →

Can user embeddings personalize language models more efficiently than prompts?

Does distilling user interaction history into learned embeddings outperform stuffing that history directly into prompts for personalizing large language models? This matters because interaction data is long and expensive to process as tokens.

Explore related Read →

Can AI systems preserve moral value conflicts instead of averaging them?

Current AI systems wash out value tensions through majority aggregation. Can we instead model how values like honesty and friendship genuinely conflict in moral reasoning?

Explore related Read →

Do people respond to machines mindlessly or reflectively?

Human-machine communication research uses two competing explanations for social responses to machines: automatic cue-triggered reactions versus conscious negotiation about what machines ontologically are. Which mechanism actually drives interaction?

Explore related Read →

Should agentic AI autonomy expand based on earned trust?

Does a staged rollout gated on demonstrated user trust offer a better path to agentic autonomy than fixed timelines or capability thresholds? And what evidence should count as proof of trust?

Explore related Read →

Do six UX patterns improve agentic AI user trust?

Victor Yocco proposes six interaction patterns spanning pre-action, in-action, and post-action phases to help users maintain control and confidence with AI agents. The question is whether these patterns actually deliver the psychological safety and trust they claim to provide.

Explore related Read →