Sean Rintel: Collaboration & AI Researcher

The richness of collaboration technology lies in its adaptability to purpose, not its fidelity to reality.

The same holds for AI. Models will keep getting more capable, but what will matter most for collaboration is AI that adapts to what people intend.

I study what people actually do when they work together through technology: the practices they build around the features, such as how they manage attention, deal with trouble and make sense of things together.

This question led me to Microsoft Research, where I studied hybrid meetings, telepresence robots, avatar communication, and the pandemic's impact on meetings.

More recently, my work has turned to a question about AI: How should Generative AI support human collaboration, without undermining the thinking that makes collaboration valuable?

The work has moved from communication research, through collaboration design, to protecting people's thinking from the unintended costs of AI. Along the way it has informed Microsoft Teams features, 12 US patents and the New Future of Work reports.

More Background

What I work on

My research develops knowledge, best practices and experiences that enable human agency, support creativity and collaboration, and ensure equitable participation. At Microsoft Research I was part of the People-Centric AI team in Cambridge UK. My most recent projects there included Tools for Thought, on how to help people think and collaborate better with AI, and the Intentional Meetings workstream, on evolving purposeful meeting systems, behaviours and cultures with generative AI.

I ran cross-org initiatives and shaped strategy. I put a lot of energy into making research legible to the people who can act on it, whether a product team, an executive audience or the public. The New Future of Work reports I helped create are read and cited in industry and policy discussions.

I am involved with the Australian Research Council Centre of Excellence for Quality Work in the Digital Age.

How I work

My background is in qualitative sociology. Since then I've used and managed a wide range of methods to understand how people's communicative choices interact with designers' technical ones. I'm also exploring how AI can augment research sensemaking, and how vibe-coding is changing research practice. I've mentored researchers, postdocs, and students at all levels, and worked closely with product, design, and policy leads.

Conversation Analysis Ethnography Ethnomethodology Design Research Field Studies Interviews Lab Studies Mixed Methods Surveys Technology Probes Telemetry

Research to Product

I have co-led research that transferred to Microsoft Teams (Companion Mode, Spatial Audio). I've delivered evidence-backed decision presentations and prototypes, and managed initiatives to present those of others, to shape the Teams meetings strategy. I've shipped AI-driven experiences both internally and as open source (such as Promptions to help users steer AI). As part of the Tools for Thought project, I've delivered educational presentations on AI usage to multiple organisational groups.

Employment

2014–2026: Senior Principal Research Manager, Microsoft Research, Cambridge UK.

2010–2014: Lecturer in Communication, The University of Queensland, Brisbane, Australia.

Education

PhD in Sociology, University at Albany, SUNY. Chaired by Professor Emerita Anita Pomerantz.

MA English, The University of Queensland.
BA (Hons) English, The University of Queensland.

Better MeetingsA meeting's purpose should shape everything about it, yet many meetings run without one stated. Good help with meeting goals moves people from wherever they start, with depth that fits the meeting.Better Thinking With AIGenerative AI moves people's thinking into stating what they want and judging what they get. The easiest way to use it can help least, so good tools ask people to think first.

Research Map

Communication

Close analysis of how people actually interact through technology, in conversation, expression and meaning-making. Good collaboration tools start from this understanding.

Collaboration

Designing systems that support collaboration as a social practice, in hybrid meetings, telepresence, spatial audio and avatars. How a meeting feels to the people in it matters as much as what the system transmits.

Tools for Thought

Making sure generative AI supports the thinking that makes collaboration valuable, instead of bypassing it. A tool for thought should stretch how people think.

All publications

Hover over a dot for details, and click to open the paper.

From IRC to Videocalling

People orient to the expressive possibilities of technology.

My research on how strangers manage interaction on Internet Relay Chat established an approach I still use: close analysis of how people actually interact through technology, rather than how they say they do.

This conversation-analytic method revealed that even in the simplest text-based chat, people construct sophisticated social practices: openings, turn-taking, and recovery from ambiguous silences. These practices are people's own achievements, made through the technology.

The same method revealed how long-distance couples opportunistically use video-calling glitches as relational resources, how domestic soundscapes can be instrumentalised for ambient awareness, and how membership categorisation works as embodied social action.

IRC conversation analysis extract — Close analysis of interaction, from IRC text to video-calling couples.

Video-calling conversation analysis extract — Close analysis of interaction, from IRC text to video-calling couples.

Crisis Memes paper figure — Crisis memes and fail pets: vernacular genres where people retool templates and brand mascots into expressive, communal resources.

Fail Pets paper figure — Crisis memes and fail pets: vernacular genres where people retool templates and brand mascots into expressive, communal resources.

Templates, Mascots, Memes

The same pattern holds beyond one-to-one interaction: people repurpose the expressive affordances of technology to make meaning together.

My work on crisis memes showed that the templatability of image macros is what makes them central to internet culture and freedom of expression: people take a shared visual frame and rework it for their own situation, turning private feeling into public, remixable commentary.

A related study of fail pets traced how error screens evolved from terse alerts into whimsical brand mascots, a small design move that turned moments of failure into chances for empathy, humour and brand identity.

In both cases, technology offers a frame and people work out together what it means.

For design, this means studying what people do with technology, which rarely matches what they say they do, and designing for the practice, with features in its service.

Remote & Hybrid Collaboration

Collaboration is built from social practices, and AI has to support them.

People talk collaboration into being. They negotiate who speaks, who's paying attention, and what the interaction means.

Across 30+ papers on meetings, hybrid work, and telepresence, the same patterns recur:

Embrace asymmetry in collaboration technologies. Hybridge gave each side a different interface. In the room, each remote person had their own display at the table. Remote people joined through a digital twin of the room, choosing their seat and where to look. In lab meetings with two people in the room and two remote, remote people's sense of agency no longer lagged significantly behind the room's. The gap in co-presence, the sense of being together, narrowed. Inclusion came from fitting each side's needs.
Design for dialectics. Meeting (the) Pandemic drew on diaries from employees working from home in 2020. We argue that the dialectic tension between effectiveness and sociality is inherent in working relationships and cannot be designed away. Videoconferencing fatigue arose partly because work practices and technologies assumed steady states and a taken-for-granted balance between the task and social dimensions of work. They also treated video meetings as static containers for any kind of communication. Design should move beyond meetings to new dynamic collaboration concepts, and let teams experiment with their own balance.
Design for conversational transitions. In Making Space for Social Time, we argue that traditional video meetings have no space or time for moving into a meeting at the start and out of it at the end, as we do in person. These conversational transitions are where people chat, huddle, and check on things. Our prototype opened a social space five minutes before each meeting began and kept it open five minutes after it ended, where people moved between groups as video bubbles. Early testers enjoyed mingling there.
Telepresence is as much a social problem as a technical one. A ten-paper program on mobile robotic telepresence at MSR found that workplace adoption fails when robot affordances mismatch workplace social practices. VROOM put an augmented reality avatar on a telepresence robot. In a study with ten pairs, it made pointing and gaze easier to read, yet seemed to leave less sense that the remote person belonged there.
Freelance platforms create management work that nobody accounts for. Two studies on corporate employees hiring via freelance platforms (Stuck in the Middle and Invisible Work) found that hirers must absorb significant transaction costs the platform was supposed to eliminate, while freelancers perform extensive coordination and relationship work that is structurally invisible to the platform and therefore unrewarded. The burden on both sides comes from the platform's design.

Hybridge: spatial inclusion for remote participants in hybrid meetings — Hybridge, Perspectives, CT-Space, and VROOM: prototypes that give remotes spatial presence around the table, restore the conversational transitions that meetings normally skip, and give telepresence robots a meeting-grade face.

Perspectives: inclusive and equitable hybrid meeting experiences — Hybridge, Perspectives, CT-Space, and VROOM: prototypes that give remotes spatial presence around the table, restore the conversational transitions that meetings normally skip, and give telepresence robots a meeting-grade face.

Meeting design must be meeting-centred, not user-centred.Best Paper

When technology failed remote colleagues, people in the room kept the meeting going by fixing the problem, letting it pass, or moving on without them. Design should support the whole meeting ecology of practices, technologies, and spaces.

(Re)Configuring Hybrid Meetings, CSCW

Spatial audio makes video meetings feel more like sharing a room.Hon. Mention

In a 75-person experiment, spatial audio made conversations feel more interactive and more like sharing the same space, by restoring the sound cues people use in person to tell who is speaking.

Hear We Are, CHIWORK

Multitasking follows how meetings are set up.Hon. Mention

In telemetry from a global technology company, about 30% of meeting attendances involved sending email during the meeting, more often in long, large, recurring and morning meetings. People multitasked to catch up or when a meeting seemed irrelevant to them, and some said it helped.

Large Scale Analysis of Multitasking, CHI

Parallel chat is powerful and perilous.

In diaries and polls from employees working from home in 2020, parallel chat had clear pros and cons. On the plus side, people said it let them share links and contribute without interrupting. On the minus side, side conversations split their attention and made presenting harder.

The Promise and Peril of Parallel Chat, CHI

Avatars & Extended Reality

Realism matters less than people expect.

Four studies on avatars and 3D video in meetings show that representation choices shape mood perception, evaluation, and inclusion in ways that pure fidelity does not predict.

In Mixed Reality, cartoon faces can outperform realistic ones over time. In a two-to-three-week HoloLens 2 field study, realistic avatars raised expectations their animation could not meet, producing more mood-perception errors than cartoon avatars. Words, tone, and movement mattered more than rendering style.
Avatar evaluation depends on ecological validity, meaning how closely a test matches real use. In two AR experiments with 56 people, viewers seemed to tolerate more facial animation noise in full conversations than in isolated clips. In conversation, their judgements also varied with their empathy and, possibly, the characters' gender. Claims about "acceptable" avatar quality need to control for context, culture, and character, which we call the Triple C.
Spatial 3D video drew attention but felt less comfortable. In An Equal Seat at the Table, we placed webcam-derived 3D video of participants in a shared virtual meeting room: people could tell who was looking at whom more easily than in the grid, but the grid still felt more comfortable and professional, and 26 of 48 preferred it. Familiarity is itself a design constraint.
Avatars that nod help people feel included. In an all-avatar videoconferencing study, people felt more comfortable and included when webcam-driven avatars showed their nods and smiles than with voice-driven or static avatars.

Avatars in mixed-reality meetings: realistic vs cartoon facial likeness — Realistic vs cartoon avatars in mixed reality, avatar facial-animation noise in AR, and 3D video in a shared virtual meeting room.

The Problem

The features that make GenAI productive can also erode the thinking that makes work valuable.

Generative AI is changing knowledge work quickly, and the important question is what it should do. My research finds a tension. The fluency and speed that make it productive can also erode the critical thinking and intentionality that make human work valuable.

AI tools should keep people thinking, by helping them monitor their own reasoning.

For design, this means AI interfaces that leave room to check, question and take responsibility for the work, as well as speed it up.

Tools for Thought Venn diagram: Working With Purpose × Thinking By Doing — The *Tools for Thought* framing: AI that helps us figure out the job as well as do it.

GenAI makes thinking feel easier.161K+ downloads

In the most downloaded CHI paper in the ACM Digital Library, we surveyed 319 knowledge workers and found that higher confidence in GenAI is associated with less reported critical thinking, while higher self-confidence is associated with more. Participants said AI cut the effort of finding information, solving problems, and drafting. It added effort in verifying answers, integrating output into their work, and steering the AI. We describe these as shifts from information gathering to information verification, from problem-solving to response integration, and from task execution to task stewardship. AI tools should build in moments to stop and question.

The Impact of Generative AI on Critical Thinking, CHI

Metacognition is both the demand and the opportunity.Best Paper

We argue that generative AI imposes metacognitive demands, requiring a high degree of monitoring and control of one's own thinking. People must state goals, decompose tasks, evaluate output, and decide when to use AI at all. Tools could integrate metacognitive support, such as prompts to plan and self-evaluate, and could reduce the demand through explainability and customisability.

The Metacognitive Demands and Opportunities of Generative AI, CHI

The "ironies" of GenAI parallel classic automation failures.

Drawing on Bainbridge's ironies of automation, our review of generative AI user studies argues that GenAI can reduce productivity. People shift from producing work to checking output they did not write, and AI suggestions interrupt them. The solution lies in better interaction design that keeps humans in the cognitive loop, rather than in more or less AI.

Ironies of Generative AI, IJHCI

The pervasive problems of ineffective meetings — Lack of intentionality is the persistent meeting problem motivating a research program that follows the meeting lifecycle.

Meeting Intentionality

How can Generative AI help people reflect on meeting goals?

Meetings often run without a stated purpose, and that makes them harder to run well. Our work studies meeting goals empirically across the whole meeting lifecycle: before, during, and between meetings.

GenAI is well-suited to this. It can surface goals that people haven't articulated, by asking questions organisers rarely think to ask themselves. It can prompt reflection at the moment when action is still possible. And it can carry intent across sessions, so goals accumulate rather than evaporate. The risk is that the same fluency that surfaces intent can also substitute for it. Our designs treat GenAI as a prompt for human thinking, not a replacement.

Stated goals may help twice. In our studies, people said putting goals into words helped them focus and plan. We argue that stated goals also give AI systems a record of intent to use across sessions, instead of guessing the purpose from scratch each time.

FoundationsMental Models of Meeting Goals

The formative study underpinning the program: how organisers and participants mentally model meeting goals, and what intentionality support would have to look like to fit those models.

Mental Models of Meeting Goals: Supporting Intentionality in Meeting Technology, CHI

Field experiment overview: 361 employees, 7196 meetings

Before meetingsNudging Attention to Workplace Meeting GoalsCHI 2026

A preregistered field experiment with 361 employees across 7,196 real meetings. Over two weeks, both the employees who rated each meeting afterwards and those who also set goals beforehand reported clearer meeting goals and better communication of them. Adding the prompt before each meeting brought no further measurable gain.

Nudging Attention to Workplace Meeting Goals, CHI

What Does Success Look Like? Prospective reflection prototype

Before meetingsWhat Does Success Look Like?

Employees who talked through upcoming meetings with an AI coach said putting goals into words helped them focus. Some then shared agendas or rethought recurring meetings.

Catalyzing Meeting Intentionality with AI-Assisted Prospective Reflection, CHIWORK

Before & during meetingsCoExplorer

A GenAI-powered adaptive interface technology probe spanning planning and running video meetings. Used as a generative design vehicle to surface what intentionality support could look like across the lifecycle.

The CoExplorer Technology Probe, DIS

Are We On Track? AI-assisted goal reflection

During meetingsAre We On Track?

Fifteen employees replayed their own recorded meetings with two AI prototypes. A quiet display of topics and goals helped them notice drift. Pop-up questions prompted action but could interrupt.

Are We On Track? AI-Assisted Active and Passive Goal Reflection During Meetings, CHI

Between meetingsTemporal Work Across Meetings

Interfaces that use Generative AI to scaffold the connective work people do between meetings: tying threads forward, surfacing unresolved goals, and making the gap between sessions productive rather than lost.

Designing Interfaces that Support Temporal Work Across Meetings with Generative AI, DIS

Innovation & Translation

Research that reaches products and patents.

12 US patents and 6 Microsoft Hackathon wins (1 Grand Prize, 4 Firsts, 1 Second).

Watch on YouTube

Product Feature

Microsoft Teams Companion Experiences: Meetings are easier when your computer and phone can work together

Teams meetings used to force a choice: join from your computer or your phone, not both. Companion Experiences removed that constraint. Join on your PC and use your phone to share live video, show your screen, remotely advance slides, or move your audio and camera between devices (without conflicting audio streams or a second meeting join). How to join on a second device

What it does and how it got there

Presenting live video from a phone in a Teams meeting

Sharing the phone screen in a Teams meeting

Remotely controlling PowerPoint from a phone

Present live video or photos. Remote participants get a view of your physical surroundings (a whiteboard, a product, a space) by joining your phone as a companion device and sharing live mobile video or photos displayed on your computer screen.

Present your phone screen. Share mobile apps or phone-only content directly into the meeting, or project your phone locally to the meeting room without remote participants seeing it.

Remote-control PowerPoint. Advance slides from your phone while presenting from your computer (useful when you need to move around the room). Multiple presenters can independently review the deck and request control, cutting the time gaps in handoffs during Q&A.

Switch devices without rejoining. Toggle your video, speaker, and microphone between devices in any combination as you move. No disconnect, no second audio stream.

The project began as Skype Unleashed, a 2015 Microsoft Hackathon entry (a vision video showing what calling could look like if a phone and PC worked in concert rather than competing). It won the Business category. The team spent two years building a working prototype called Project Wellington, which joined Brian MacDonald's early Microsoft Teams incubation. Features were announced at Build 2017 and Enterprise Connect 2018, where Teams took the top prize.

"I was showing the whiteboard through the phone to the remote participants. It genuinely helped the course of the meeting." — L., Microsoft UK Marketing and Operations

The multi-device meeting research behind this project contributed to three US patents: Companion Devices for Teleconference (granted 2022), Enhanced Joining Techniques for co-located device detection (granted 2021), and Proximate Resource Pooling for AV resource sharing across nearby devices (granted 2017).

Watch on YouTube

Open Source Feature

Promptions: If You’re Building UI for AI, Give Users the Power of Choice

Most AI interfaces put the burden of exploration on the user: write a prompt, get a response, guess at a better prompt. Promptions inverts this (it generates contextual UI elements: radio buttons, checkboxes, toggles) so users can explore variations of the AI's response without rewriting anything.

It is open source: clone the repo on GitHub or see more on AI Foundry Labs.

How it works and what it’s built on

Promptions and a standard AI interface, side by side

The standard AI interaction loop puts all the steering work on the user: you write a prompt, interpret the output, then try to refine your wording to get closer to what you wanted. A recurring frustration in user research: "I have to do so much work around my prompt to give it the context, point it to the tone, and a lot of times I'm repeating those same things."

Promptions breaks that loop. When you submit a prompt, the system generates a small set of contextual options alongside the response (controls specific to what you just asked, not generic settings). Adjust a toggle or pick a radio button and the response updates immediately. Users in early studies found it surfaced possibilities they hadn't considered: "It generated options that were things I would maybe assume the system couldn't handle."

Promptions system model: option module, rendering engine, chat module

Inside, an Option Module ingests the prompt and conversation history and outputs a structured set of options. A rendering engine turns those into inline UI controls. The Chat Module incorporates the selected options as grounding when generating or regenerating the response. Changing a control reruns the Chat Module immediately (no re-prompting required).

Promptions is grounded in CHIWORK 2025 research on dynamic prompt middleware. It ships as a TypeScript monorepo for React and Fluent UI applications with OpenAI integration. Invented by Ian Drosos (past Microsoft Research Resident) with Jack Williams, Advait Sarkar, Nicholas Wilson, Payod Panda, and Sean Rintel (a collaboration between the Tools for Thought and ENCODE projects at Microsoft Research Cambridge).

12 US Patents

These patents and prototypes came out of the research. Meeting-centred design led to spatial hybrid meeting systems, avatar research to personalised animation, multi-device research to companion devices, and robotic telepresence to multi-endpoint mixed reality.

Hybrid Meeting Systems

Collaborative System: spatial hybrid meetings with avatar-equipped displays and digital twinsgranted 2025
Established Perspective UI: spatially consistent seating in video meetingsgranted 2023, 2025
Simulated Choral Audio Chatter: crowd audio for virtual event atmospheregranted 2024

Extended Reality & Avatars

Multi-endpoint Mixed-Reality Meetingsgranted 2022, 2024
Headset Virtual Presence: headset-controlled avatars with emotional expressiongranted 2023
Personalized Avatar Animation: few-shot ML model for audio-driven avatar movementgranted 2025

Interaction & Input

Computing Device Headset Input: headphones as a general-purpose input devicegranted 2023
Semantic User Input: context-aware gesture-to-action mappinggranted 2023
Mixed Reality Workflow Generationgranted 2025

Multi-device Collaboration

Companion Devices for Teleconferencegranted 2022
Enhanced Joining Techniques: co-located device detectiongranted 2021
Proximate Resource Pooling: AV resource sharing across nearby devicesgranted 2017

6× Microsoft Hackathon Winner

2024

Promptly 2nd place

Dynamically-generated UI for AI: everyday, any purpose, anywhere. Now Promptions on AI Foundry Labs.

2021

Inclusive Avatars in MR Grand prize

Teach your avatar to move like you, using few-shot training.

LinkedIn Career Classroom in Microsoft Teams

2019

LinkedIn Career Classroom 1st place

Connecting children from anywhere to career role models.

2019

MILC 1st place

Mobile Intelligent Lenses for everyday Mixed Reality at work.

TACTILES hackathon: tangible cloud intelligence

2017

TACTILES 1st place

Tangible Active Cloud Intelligence: making the cloud a conversation.

2015

Skype Unleashed 1st place

Surface Hub cross-device patch-panel UI. Productized as Companion Mode in Microsoft Teams. Garage Wall of Fame.

What's Next

Next: AI that supports what people intend.

Generative AI is now in many collaboration tools, and much of it is built to make things faster. My research suggests asking a different question.

Did the AI help you think more clearly about why you're doing this?

Design Principles

Treat the medium as a participant in its own right. Communication technologies shape what counts as presence and contribution. Because the channel shapes what happens, study the practices that grow up around it.
In practice
- AI narration is a social act. When AI produces a summary or an action-item list, it creates an authoritative account that shapes who is held accountable: some contributions get foregrounded, others quietly dropped. Design for visible narration choices, and let participants co-author the AI's account of what happened.
- AI uptake reproduces who counts as a speaker. Speech recognition and language models trained mostly on majority speech can transcribe accented, overlapping, or quiet voices less accurately. When AI hears some people better than others, the medium decides who counts as a participant. Audit AI uptake with the rigour you already apply to interface accessibility.
Embrace asymmetry. People arrive at a meeting from different bodies, devices, situations, and positions of power. Equal participation comes from tailoring affordances to each person's situation.
In practice
- AI models the meeting from a single vantage point. Meeting AI usually produces one account, as if no one in particular authored it. But a meeting is lived differently by the host, by remote participants, and by those with less power, and one account privileges a single perspective while appearing objective. Design AI that can take a stated perspective, or lay several accounts side by side.
- Personalization is not individualization. AI that learns your preferences is tracking past behaviour, which reveals little about your structural position in the room. A remote participant differs from a collocated one by role and access rather than taste. Adapt AI to role and context, treating preference history as only a weak signal.
Put intent and process ahead of output. Help people think about why before optimising how; speed without purpose just produces faster noise. Goals should persist across sessions as named artifacts, legible to people and AI alike.
In practice
- Surface the intent layer before the generation layer. AI that jumps straight to output skips the most valuable step: making intent explicit. Ask users to say why they want something before it generates anything. That articulation is cognitive work worth supporting in its own right, and in a randomised experiment with 297 people, those prompted to set goals before AI drafted anything wrote better briefs, though they rated the AI as less helpful.
- Distinguish AI-assisted articulation from AI-substituted articulation. Helping you find words for what you already intend is a different thing from proposing new goals on your behalf. The first is assistance. The second quietly trains people to accept AI-generated intent. Make assistance the primary mode, and treat any AI-proposed goal as scaffolding to own and revise.
- Treat goals as persistent artifacts. AI that forgets a meeting's goals when the session ends treats intent as ephemeral, but goals are durable commitments. Keep them across sessions and visible to everyone, so AI can refer back to them over time.
Scaffold metacognition and calibrate confidence. Using GenAI well means monitoring your own thinking: knowing when to trust output and when to question it. That is hard, and a design opportunity. Build interfaces that help users calibrate trust and resist treating fluency as authority.
In practice
- Fluency gets mistaken for authority. GenAI produces polished output regardless of accuracy, and people can mistake well-written for correct. Counteract that: show where a claim comes from, and mark uncertainty in the interface itself rather than burying it in a disclaimer.
- Design for productive scepticism. Disclaimers and confidence scores inform consent, but they don't create active doubt. Build moments that make people think, such as asking them to predict an answer before AI shows one, or flagging when its output has changed since last time.
- A heavy verification burden is itself a design failure. When checking falls entirely on users, it disadvantages those with less expertise or time. Make it tractable: link claims to their sources, and flag which outputs are high-stakes enough to be worth checking at all.
Enable thinking by doing. Insight often comes through action. It emerges through making and revising. AI tools should treat externalisation as the work itself: lower the cost of producing rough, provisional things so people can think through artifacts, not only about them. Reversibility matters more than first-try fluency.
In practice
- Treat the first AI output as a provocation. "Generate, then edit" treats AI output as nearly done, but seeing a response anchors your thinking to its structure. Have AI open the possibility space first, with alternatives and counterproposals, before narrowing it. Its first job is discovery, and production only after that.
- Make AI outputs structurally reversible. If AI makes experimentation cheap, make undo and branching cheap too. Design AI interactions as explorations with clear return paths, where committing takes more deliberate effort than exploring.
- Design AI for dialogue about the task. AI is most useful when it helps you work out what to argue and what to leave out, before it writes anything. Support that conversation up front: ask what the group is trying to do, and what would count as success.
Design for meaningful shared attention. A firehose of notifications does not add up to collaboration support. As AI becomes a participant in shared work, interfaces should cultivate sustained attention instead of competing for it.
In practice
- Distinguish AI-generated urgency from human-generated urgency. AI notifications create synthetic urgency that can override priorities the group negotiated. A person raising a point exercises judgment. AI flagging drift is just executing code. Design AI that reads the room before intervening, weighing more than its own relevance.
- The unit of attention is the whole meeting. AI optimised for one person's attention fragments the group's shared focus. Design at the group level: ask what everyone needs to see together.
- Design for attentional recovery. Many AI features compete for attention. The harder and more useful problem is helping people return to shared focus after a drift or tangent.