Virtual Collaborative Intelligence

Teams will increasingly include both humans and AI agents. We need to learn how best to configure them. A new Stanford University paper "ChatCollab: Exploring Collaboration Between Humans and AI Agents in Software Teams" reveals a range of useful insights. A few highlights: 💡 Human-AI Role Differentiation Fosters Collaboration. Assigning distinct roles to AI agents and humans in teams, such as CEO, Product Manager, and Developer, mirrors traditional team dynamics. This structure helps define responsibilities, ensures alignment with workflows, and allows humans to seamlessly integrate by adopting any role. This fosters a peer-like collaboration environment where humans can both guide and learn from AI agents. 🎯 Prompts Shape Team Interaction Styles. The configuration of AI agent prompts significantly influences collaboration dynamics. For example, emphasizing "asking for opinions" in prompts increased such interactions by 600%. This demonstrates that thoughtfully designed role-specific and behavioral prompts can fine-tune team dynamics, enabling targeted improvements in communication and decision-making efficiency. 🔄 Iterative Feedback Mechanisms Improve Team Performance. Human team members in roles such as clients or supervisors can provide real-time feedback to AI agents. This iterative process ensures agents refine their output, ask pertinent questions, and follow expected workflows. Such interaction not only improves project outcomes but also builds trust and adaptability in mixed teams. 🌟 Autonomy Balances Initiative and Dependence. ChatCollab’s AI agents exhibit autonomy by independently deciding when to act or wait based on their roles. For example, developers wait for PRDs before coding, avoiding redundant work. Ensuring that agents understand role-specific dependencies and workflows optimizes productivity while maintaining alignment with human expectations. 📊 Tailored Role Assignments Enhance Human Learning. Humans in teams can act as coaches, mentors, or peers to AI agents. This dynamic enables human participants to refine leadership and communication skills, while AI agents serve as practice partners or mentees. Configuring teams to simulate these dynamics provides dual benefits: skill development for humans and improved agent outputs through feedback. 🔍 Measurable Dynamics Enable Continuous Improvement. Collaboration analysis using frameworks like Bales’ Interaction Process reveals actionable patterns in human-AI interactions. For example, tracking increases in opinion-sharing and other key metrics allows iterative configuration and optimization of combined teams. 💬 Transparent Communication Channels Empower Humans. Using shared platforms like Slack for all human and AI interactions ensures transparency and inclusivity. Humans can easily observe agent reasoning and intervene when necessary, while agents remain responsive to human queries. Link to paper in comments.

A new paper from USC introduces "Mixture of Thoughts" - and it might fundamentally change how we build AI systems. Here's the problem they're solving: We have dozens of specialized AI models now - some excel at math, others at coding, others at language. But when we try to combine them, we typically just pick one expert per task or awkwardly average their outputs. It's like having a team meeting where everyone works in isolation and you just vote on the final answer. Not exactly collaborative intelligence. Mixture of Thoughts comes with a different approach: Instead of waiting until the end to combine outputs, these models share their intermediate "thoughts" - their hidden representations - as they process information. A lightweight router picks the best experts for each query, but the primary expert can peek at what all the other experts are "thinking" at multiple points during processing, not just at the end. It's genuine collaboration at the level of reasoning, not just output aggregation. The system outperforms individual expert models by up to 10% while maintaining the efficiency of single-pass inference. More importantly, it works with completely different model architectures - you don't need matching models to collaborate. And when one expert fails? The system gracefully degrades, maintaining performance through the remaining collaborators. Instead of racing toward ever-larger monolithic models, we might build ecosystems of specialized, collaborative agents. Imagine your organization's custom models seamlessly collaborating with open-source specialists and commercial APIs, all thinking together rather than just voting on outputs. We're moving from AI as a tool to AI as a team. What would be your first AI "colleagues" today? ↓ 𝐖𝐚𝐧𝐭 𝐭𝐨 𝐤𝐞𝐞𝐩 𝐮𝐩? Join my newsletter with 50k+ readers and be the first to learn about the latest AI research: llmwatch.com 💡

An earlier version of this study appeared in pre-agentic #AI time. The implications are worth rethinking now that we are developing AI agents. The Why and What: ideation and deliberation across large groups of people, for instance, large organizations or decentralized political systems - our current models, from asynchronous threaded conversations to synchronous "assemblies," don't scale well. AI can help her, acting as a conduit between the human groups, in a swarm model. The How: Conversational Swarm Intelligence (CSI), is an AI-facilitated method for enabling real-time conversational deliberations and prioritizations among large human groups. CSI is inspired by biological Swarm Intelligence, like the decision-making dynamics of fish schools. It addresses the problem that traditional large group conversations quickly lose effectiveness by dividing the large group into a network of small subgroups (ideally 4-7 members). Within these subgroups, AI agents called "Conversational Surrogates," powered by Large Language Models, observe discussions, distill key points, and pass insights and ideas between subgroups. This process weaves the subgroups into a single, larger conversation where ideas can emerge and propagate efficiently. CSI allows large groups (from 50 to potentially thousands) to brainstorm, debate, prioritize, and converge on solutions in real-time. Prior studies showed CSI can increase participation, foster more balanced dialog, and amplify collective intelligence, even achieving "gifted" status in an IQ test setting. The results from 147 participants showed a significant majority preferred brainstorming with the CSI structure over traditional chat. Participants reported that the CSI method felt more productive, more collaborative, and better at surfacing quality answers. They also felt more heard, had more ownership, and more buy-in in the final answers using CSI.

🚀 Excited to share my latest Fortune column on truly groundbreaking academic work from my co-authors Professor Karim Lakhani and Fabrizio Dell'Acqua at Digital Data Design Institute at Harvard (D^3), where I serve as an executive fellow. This remarkable field experiment with 776 Procter & Gamble professionals fundamentally challenges what we thought we knew about teamwork. The research reveals the emergence of the "cybernetic teammate"—AI that doesn't just assist but actively participates in collaboration. Three breakthrough findings: 1. AI Can Replicate Team Benefits Individuals working with AI achieved nearly 40% performance gains—matching traditional two-person teams. AI is providing the same collaborative benefits we've long attributed to human teamwork. 2. Cross-Functional AI Teams Generate Breakthrough Innovation AI-augmented cross-functional teams were 3x more likely to produce top 10% solutions. This isn't marginal improvement—it's a multiplicative effect that neither human-only teams nor AI-enabled individuals could achieve alone. 3. AI Breaks Down Silos (For Real This Time) R&D specialists with AI proposed commercially viable solutions. Commercial professionals developed technically sound approaches. AI acted as a bridge, enabling each team member to think holistically across functions—achieving the "silo breaking" that leaders have struggled to accomplish through org chart reshuffles. Bonus finding: AI collaboration increased positive emotions by 64% in teams. This isn't cold, mechanical work—it's energizing and engaging. At Seven2, we're translating this research into practice with our portfolio companies, building these AI-augmented cross-functional teams to drive innovation and competitive advantage. This is the future of collaborative work—not AI replacing humans, but human-AI ensembles that combine the best of both worlds. Read the full analysis: https://lnkd.in/ef3f3pED #AI #Innovation #HBS #D3Institute #FutureOfWork #PrivateEquity #TeamDynamics

AI isn't just a tool; it's becoming a teammate. A major field experiment with 776 professionals at Procter & Gamble, led by researchers from Harvard, Wharton, and Warwick, revealed something remarkable: Generative AI can replicate and even outperform human teamwork. Read the recently published paper here: In a real-world new product development challenge, professionals were assigned to one of four conditions: 1. Control Individuals without AI 2. Human Team R&D + Commercial without AI (+0.24 SD) 3. Individual + AI Working alone with GPT-4 (+0.37 SD) 4. AI-Augmented Team Human team + GPT-4 (+0.39 SD) Key findings: ⭐ Individuals with AI matched the output quality of traditional teams, with 16% less time spent. ⭐ AI helped non-experts perform like seasoned product developers. ⭐ It flattened functional silos: R&D and Commercial employees produced more balanced, cross-functional solutions. ⭐ It made work feel better: AI users reported higher excitement and energy and lower anxiety, even more so than many working in human-only teams. What does this mean for organizations? 💡 Rethink team structures. One AI-empowered individual can do the work of two and do it faster. 💡 Democratize expertise. AI is a boundary-spanning engine that reduces reliance on deep specialization. 💡 Invest in AI fluency. Prompting and AI collaboration skills are the new competitive edge. 💡 Double down on innovation. AI + team = highest chance of top-tier breakthrough ideas. This is not just productivity software. This is a redefinition of how work happens. AI is no longer the intern or the assistant. It’s showing up as a cybernetic teammate, enhancing performance, dissolving silos, and lifting morale. The future of work isn’t human vs. AI. The next step is human + AI + new ways of collaborating. Are you ready?

What kills collaboration faster than conflict? Silence. How AI can fix it.     We've all been there: a meeting ends, everyone nods, no one asks questions... and yet, the project still goes sideways. The truth? Silence doesn’t mean clarity. Silence in teams can feel like alignment, but it's often confusion in disguise. It usually means someone didn’t feel safe or empowered to ask for it.   Even the best teams hit roadblocks:   Misunderstandings from assumptions Hesitation to ask questions Miscommunication that leads to rework   These challenges aren't new, but the way we tackle them can be.   This is where AI can quietly transform how your team collaborates. By acting as a neutral, judgment-free assistant, AI makes it easier for people to understand questions, clarify tasks, and stay aligned without fear of “looking dumb.”    Here's how:   ✅ Clarify complexity – AI can quickly summarize dense threads, documents, or meeting notes. ✅ Encourage curiosity – With the right prompts, AI makes it safe and easy to ask “obvious” questions. ✅ Keep teams in sync – AI can reinforce shared goals and priorities without sounding repetitive. It’s like adding a smart, impartial facilitator to every meeting, every teams thread, every project doc.   💡 Try this prompt to get started: "You are a helpful team assistant. Whenever I ask a question, respond with a reasonable amount of detail to help the team work together effectively." Simple but powerful to make missing information to all team members visible.     Ready to bring this into your team culture? Start with these steps:   1. Pick one team ritual (e.g., weekly meeting, retros, or docs) and layer in AI support. Let AI summarize, generate follow-up questions, or identify unclear points. 2. Encourage “clarifying questions” as a norm, not a nuisance. Use AI to increase curiosity and good inquiry. 3. Train with prompts. Craft a few go-to prompts your team can use in AI tools like Co-Pilot or whatever tool you use.   Collaboration doesn’t break down because people don’t care. It breaks down when people don’t feel clear and get frustrated.

🤔 Weekend reading - 🔄 AI and the Evolution of Collective Intelligence In 2018, I wrote a piece in "AI and Society" exploring the relationship between artificial intelligence and collective intelligence, arguing for what I called augmented collective intelligence; using AI not to replace groups, but to help overcome the transaction costs, coordination failures, and cognitive limits that often constrain collective problem-solving. 💻 Read: https://lnkd.in/ezMRya9 A lot has changed since then. A recent article by Jacob Taylor and Scott E. Page, published by The Brookings Institution, now argues that AI is changing the very physics of collective intelligence: how ideas move, how groups learn, and how decisions emerge. What resonates is the shift from asking whether AI helps groups think better to asking under what conditions it does. Some reflections, seven years on: ✅ AI can radically lower the cost of assembling diverse perspectives, but diversity without design still fails. ✅ Generative and agentic AI can help translate between narratives, models, and data, but only if embedded as infrastructure, not bolt-on tools. ✅ The real risk is not that AI weakens collective intelligence, but that it accelerates poorly designed processes, shallow consensus, or institutional monocultures. ✅ The opportunity is to intentionally design AI-enabled “room + model” systems that connect deliberation, data, and decision-making across scales. ✅ The next frontier isn’t smarter models alone; it’s new norms, governance, and collaboration architectures that ensure AI strengthens shared understanding, agency, and action. 👉 How do we design collective intelligence systems that are faster, more inclusive, and more adaptive without hollowing out trust, judgment, and responsibility? 💻 Read: https://lnkd.in/eJ7vs8ig #CollectiveIntelligence #AIandSociety #DataGovernance #PublicInterestAI #SystemsThinking #AugmentedIntelligence

𝗧𝗵𝗶𝗻𝗸 𝗹𝗶𝗸𝗲 𝗮 𝘁𝗲𝗮𝗺—𝗲𝘃𝗲𝗻 𝘄𝗵𝗲𝗻 𝘆𝗼𝘂'𝗿𝗲 𝘀𝗼𝗹𝗼. Turn ChatGPT into your 𝘃𝗶𝗿𝘁𝘂𝗮𝗹 𝘀𝘄𝗮𝗿𝗺 𝘁𝗲𝗮𝗺: Tackle tough problems by simulating a room full of experts—CEO, CFO, Innovator, Customer, and more. Think like a team. Decide like a strategist. Solve like a pro. 𝗧𝗵𝗲 𝗥𝗼𝗹𝗲 𝗟𝗲𝗻𝘀 𝗜𝗻𝘀𝗶𝗴𝗵𝘁𝘀 𝗙𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸 Role Lens Insights is a powerful way to swarm problems, expose blind spots, stress-test ideas, and generate better solutions. 𝗪𝗵𝘆 𝗜𝘁’𝘀 𝗣𝗼𝘄𝗲𝗿𝗳𝘂𝗹 • Turns solo thinking into 𝗺𝘂𝗹𝘁𝗶-𝗱𝗶𝗺𝗲𝗻𝘀𝗶𝗼𝗻𝗮𝗹 𝗶𝗻𝘀𝗶𝗴𝗵𝘁 • Builds 𝗲𝗺𝗽𝗮𝘁𝗵𝘆 for different stakeholders • Surfaces 𝗰𝗿𝗲𝗮𝘁𝗶𝘃𝗲 𝘁𝗲𝗻𝘀𝗶𝗼𝗻𝘀 𝗮𝗻𝗱 𝗯𝗹𝗶𝗻𝗱 𝘀𝗽𝗼𝘁𝘀 • Helps you 𝘀𝘁𝗿𝗲𝘀𝘀-𝘁𝗲𝘀𝘁 and 𝗿𝗲𝗳𝗶𝗻𝗲 decisions fast • Amplifies your 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗮𝗻𝗱 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝗶𝗰 𝘁𝗵𝗶𝗻𝗸𝗶𝗻𝗴 𝗛𝗼𝘄 𝘁𝗼 𝗨𝘀𝗲 𝗜𝘁 1. 𝗗𝗲𝗳𝗶𝗻𝗲 𝘁𝗵𝗲 𝗖𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲 Clearly state the problem, decision, or idea you want to explore. 2. 𝗖𝗵𝗼𝗼𝘀𝗲 𝗥𝗼𝗹𝗲𝘀 Select 3–5 expert lenses relevant to your challenge (e.g., CEO, CFO, Innovation Expert, Customer, etc.). 3. 𝗣𝗿𝗼𝗺𝗽𝘁 𝗘𝗮𝗰𝗵 𝗥𝗼𝗹𝗲 Ask ChatGPT to respond from each role's perspective (e.g., “As the CFO, what risks do you see?”). 4. 𝗙𝗮𝗰𝗶𝗹𝗶𝘁𝗮𝘁𝗲 𝗗𝗶𝗮𝗹𝗼𝗴𝘂𝗲 Have the roles "discuss" the idea as if in a team meeting. This dialogue reveals tensions, assumptions, and synergies. 5. 𝗘𝘅𝘁𝗿𝗮𝗰𝘁 𝗜𝗻𝘀𝗶𝗴𝗵𝘁𝘀 Identify key themes, trade-offs, blind spots, and opportunities across perspectives. 6. 𝗦𝘆𝗻𝘁𝗵𝗲𝘀𝗶𝘇𝗲 & 𝗔𝗰𝘁 Integrate the learnings into a better, more rounded solution. You can also apply thinking tools like 𝗦𝗶𝘅 𝗧𝗵𝗶𝗻𝗸𝗶𝗻𝗴 𝗛𝗮𝘁𝘀 or the 𝗕𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗠𝗼𝗱𝗲𝗹 𝗖𝗮𝗻𝘃𝗮𝘀 to guide deeper analysis. 7. 𝗜𝘁𝗲𝗿𝗮𝘁𝗲 𝗮𝘀 𝗡𝗲𝗲𝗱𝗲𝗱 Adjust roles, reframe the problem, or simulate new strategies to explore further. 𝗤𝘂𝗶𝗰𝗸 𝗘𝘅𝗮𝗺𝗽𝗹𝗲: 𝗘𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗻𝗴 𝗮𝗻 𝗔𝗜 𝗦𝘁𝗮𝗿𝘁𝘂𝗽 𝗜𝗱𝗲𝗮 You prompt ChatGPT to form a virtual team with 5 roles: • 𝗖𝗘𝗢: Focuses on vision and market opportunity. • 𝗖𝗙𝗢: Analyzes financial risk, ROI, and funding needs. • 𝗜𝗻𝗻𝗼𝘃𝗮𝘁𝗶𝗼𝗻 𝗘𝘅𝗽𝗲𝗿𝘁: Evaluates uniqueness and feasibility. • 𝗠𝗮𝗿𝗸𝗲𝘁𝗶𝗻𝗴 𝗟𝗲𝗮𝗱: Assesses customer fit and positioning. • 𝗔𝗜 𝗦𝗽𝗲𝗰𝗶𝗮𝗹𝗶𝘀𝘁: Explains the technical approach and scalability. Together, they discuss an AI-driven platform that predicts customer needs in real-time. Through their dialogue, you surface: • 𝗢𝗽𝗽𝗼𝗿𝘁𝘂𝗻𝗶𝘁𝗶𝗲𝘀: Personalized, proactive CX is a differentiator. • 𝗥𝗶𝘀𝗸𝘀: Cost of real-time data processing, competitive landscape. • 𝗡𝗲𝘅𝘁 𝘀𝘁𝗲𝗽𝘀: Build a lean MVP, target e-commerce, and validate with early adopters. You then 𝗮𝗽𝗽𝗹𝘆 𝗦𝗶𝘅 𝗧𝗵𝗶𝗻𝗸𝗶𝗻𝗴 𝗛𝗮𝘁𝘀 to explore the idea emotionally, logically, creatively, and cautiously—sharpening the strategy even further. What challenge will you swarm today?

Unlocking AI “Teammates” in K12 Sunanna Tara Chand asked me a great question on AI in K12 that has had me thinking deeply over the last couple of weeks. "What are we not talking about enough when it comes to AI in education?" My short answer to this is the potential to reimagine our vision for how we "staff" schools. Most of the conversation recently has been focused on productivity. While this is a big deal, and I see tools saving teachers hours a week on tasks, the larger opportunity is how we can rethink staffing models with the concept of near-unlimited virtual FTEs.  Last week Harvard Business School published a paper on this exact topic - "The Cybernetic Teammate." Ethan Mollick 📌 Summary & Key Insights from Harvard’s Research ◼️ Study focused on how AI can impact team performance ◼️ Experiment ran with 776 professionals at Procter & Gamble. ◼️ 2x2 experimental design: (1) an individual working without GenAI, (2) a team of two humans without GenAI, (3) individuals with GenAI, and (4) a team of two humans plus GenAI. ◼️ Cross-functional teams with commercial and R&D expertise Problem Statements had an innovative/generative focus and represented real challenges in business units. i.e.- "How to motivate consumers who have never tried product form X to try it as part of their regimen" ◼️ Teams with AI were provided training that included strong prompt templates 📈 Outcomes 📈 💡 On a number of evaluation metrics including quality of output, time savings, innovative solution, etc. Teams with AI > Individuals with AI > Teams without AI > Individuals without AI. 💡 The study also highlighted a leveling of the playing field of non-technical staff creating technical solutions on par with the more technical R&D individuals 🔍 Why This Matters for K12 🔍 ◼️ Resource Constraints: Schools often can’t staff large project teams; teachers juggle multiple tasks across various domains. By default, many workflows are inherently solo activities. ◼️ AI as “Virtual Teammates”: This study further supports my conviction that leveraging AI tools can help support our teachers and administrators with a virtual "team" of experts who can assist them in their work ◼️ Unleashing Innovation: Access to AI and strong training can help K12 staff deliver on critical work with the help of a virtual team. These virtual teammates can increase collaboration across departments and ultimately help our teachers feel supported. This mental model of virtual or "The Cybernetic Teammate" can break down the resource/talent availability barriers, allowing us to imagine what a world could look like with a team of experts supporting every educator and administrator in a school building. In the not-too-distant future, with a team of 20+ support agents per FTE. Oliver Sicat Yusuf Ahmad Kevin Shaw Link to study in comment #edtech #AI #k12 #futureoflearning

AI shouldn't just do your job. It should help you do it better. Too many organizations still ask: How can we automate this? But the real question for forward-thinking leaders is: How can we collaborate with AI to unlock our team’s full potential? In Issue #06 of the Meaningful AI newsletter, I dive deep into Collaborative AI, a new frontier where humans and machines work together, not in competition. → Teams working with AI aren’t just faster, they’re smarter, more motivated, and more creative. → But it only works if we redesign workflows, preserve autonomy, and upskill people, not just plug in tools. → This edition gives you a clear framework to rethink collaboration from mapping capabilities to measuring psychological impact. If you’re serious about building augmented teams, not just automated processes, this is your blueprint. Read it to find out what it really takes to build seamless human-AI collaboration.