Learning Library

← Back to Library

AI as Distributed Team Cognition

Key Points

  • The NASA space‑shuttle story illustrates that critical expertise often resides in the collective interactions of a team, not in any single individual’s knowledge or documentation.
  • Current discussions about AI focus heavily on individual productivity hacks, overlooking how AI fundamentally reshapes team dynamics and collective cognition.
  • High‑performing product teams that thrive with AI treat it as a distributed teammate, establishing shared rituals, prompt libraries, evaluation norms, and letting AI handle coordination tasks that once required meetings.
  • Most organizations, however, simply drop AI into existing workflows without redesigning team practices, risking the loss of the very collaborative knowledge that enables breakthroughs.

Full Transcript

# AI as Distributed Team Cognition **Source:** [https://www.youtube.com/watch?v=dohW_oogbis](https://www.youtube.com/watch?v=dohW_oogbis) **Duration:** 00:07:37 ## Summary - The NASA space‑shuttle story illustrates that critical expertise often resides in the collective interactions of a team, not in any single individual’s knowledge or documentation. - Current discussions about AI focus heavily on individual productivity hacks, overlooking how AI fundamentally reshapes team dynamics and collective cognition. - High‑performing product teams that thrive with AI treat it as a distributed teammate, establishing shared rituals, prompt libraries, evaluation norms, and letting AI handle coordination tasks that once required meetings. - Most organizations, however, simply drop AI into existing workflows without redesigning team practices, risking the loss of the very collaborative knowledge that enables breakthroughs. ## Sections - [00:00:00](https://www.youtube.com/watch?v=dohW_oogbis&t=0s) **Losing Collective Knowledge to AI** - The speaker warns that AI hype’s focus on individual productivity risks eroding the vital, tacit team-based expertise that made feats like the space shuttle possible. - [00:03:26](https://www.youtube.com/watch?v=dohW_oogbis&t=206s) **Team AI Needs Shared Context** - The speaker emphasizes that effective AI adoption depends on embedding it into collaborative workflows by deliberately curating and feeding shared context, rather than relying on isolated individual prompts. - [00:06:48](https://www.youtube.com/watch?v=dohW_oogbis&t=408s) **AI as Collective Team Intelligence** - The speaker urges teams to reflect on whether they use AI simply to speed up existing habits or to cultivate true collective intelligence, and invites discussion while referencing a detailed Substack on performance signals. ## Full Transcript
0:00Hey guys, I want to talk about something 0:02that is keeping me up at night that 0:04almost nobody is talking about. It's 0:06about teams and AI and a pattern that's 0:10unfolding in the way we talk about AI 0:12that is prejuditial, that is 0:16anti-team. But first, I'm going to tell 0:18you about the space shuttle. So, to cut 0:21a long story short, NASA forgot how to 0:24build space shuttles. 0:26not in like a vague way where they could 0:28go and get it out of the library. They 0:30actually lost the knowledge. The teams 0:32that built the space shuttle disbanded, 0:34documentation got scattered, expertise 0:37walked out the door. And what's 0:39compelling is that expertise didn't live 0:42in any individual's head. You could get 0:45the blueprints, you could get the 0:46documents, you could get the 0:48specifications, but you still can't 0:50build one today. 0:52The real knowledge about how to build a 0:55space shuttle existed in between in the 0:57collective knowledge of the team. And 1:00that's how all of our greatest 1:02achievements have worked have worked. 1:04It's not in any one person's head. It 1:06lives in the connections between 1:08hundreds of teams in thousands of small 1:11decisions. I've worked at big companies 1:13and I've seen amazing things get built 1:16when teams can work like this. 1:18And what matters right now is that we're 1:20at risk of missing something equally 1:23profound with the way AI is impacting 1:25our 1:26teams. Everyone's obsessed over AI 1:29productivity, prompts, workflows, 10x 1:32your output. Look, I've written some of 1:34that stuff and I am a big believer in 1:37individual productivity gains for AI. 1:39It's measurable, but I think there is a 1:41reason why it is difficult to talk about 1:44team productivity and AI right now. And 1:47I think we're missing something really 1:48big. Fundamentally, we are letting AI 1:53into existing team structures and just 1:55expecting it to work. That's not how it 1:58works. I've been watching teams work 2:00with AI for a long time now, and I'm 2:02seeing a really fascinating divide 2:06emerge. The high-erforming product teams 2:09that I have 2:11seen are doing something completely 2:13different. They're not just using AI 2:15individually. They're fundamentally 2:18distributing their 2:19cognition from the heads that are on the 2:23team now that are human to the AI as 2:26well. That means they've developed new 2:28team practices. They have specific 2:30rituals around AI generated content. 2:33They have common and collective 2:35understanding of how AI works and how 2:38prompts work on their team and what 2:40prompts are working well and not. They 2:42understand eval as a team. They 2:44understand how the workflows that they 2:46did 2:47before in some ways don't need to be 2:49done anymore. They allow AI to pick up 2:51coordination load that would have 2:53previously required a meeting. They are 2:56rethinking from the ground up how their 2:59team works with the assumption that 3:01instead of the team knowledge, you know 3:03how I talked about the space shuttle, 3:05instead of the team knowledge living 3:06just between human heads, now AI is in 3:09the mix too. and you have to find new 3:11ways of working to allow AI to be a part 3:14of that distributed 3:16cognition. On the other hand, most teams 3:19don't work that way right now. Most 3:21teams get a chat GPT subscription. 3:24They're happy to generate ideas 3:26individually. They often take ideas 3:28uncritically. They may inappropriately 3:31substitute a chat GPT chat for product 3:34requirements. I've seen that done. The 3:37difference isn't in the tools. It's not 3:39in whether or not you got Chad GPT or 3:41Grock or Gemini. No, it's in how the 3:45tools reinforce team culture. Look, 3:49we're a very individualized culture on 3:50the internet. I get why we get excited 3:52about individualized productivity gains 3:54with AI. It is a big deal. But teams 3:57build amazing things. And I think that 4:00we are missing a revolution in how teams 4:02function. 4:04For the first time in human history, our 4:07intelligence isn't just living in human 4:09heads anymore. Parts of our thinking are 4:12starting to live outside our minds, our 4:14decision-making, our problem solving, 4:15our creativity. They're living in the 4:18back and forth interactions that we have 4:20with AI. That's what I'm calling 4:22distributed cognition. And teams need to 4:25figure out how to do this 4:27collectively. Teams need to figure out 4:29how to manage shared context explicitly. 4:34A team needs to not just have a prompt 4:37bible or documentation. Teams need to 4:40think about context as something they 4:42are actively feeding to AI. Key 4:45decisions, refined 4:47outputs, a wide range of inputs that 4:49they deliberately 4:51curate. It needs to feel like a part of 4:53the natural workflow. And I know that it 4:55won't at first, but it's really really 4:58critical to enabling AI to be a 5:00successful partner on the team and to 5:02actually harvesting not just individual 5:04productivity gains, but team level 5:06benefits. Because if the individual 5:08comes and says, "Hey, I wrote that 5:09product requirements." And they feel 5:11happy and they give it to you and it's 5:13not good, but they got it done 10 times 5:15faster. The team as a whole did not 5:17benefit. It was just the individual 5:19feeling like they' done a better job. 5:20And I'm seeing that happen too often. 5:23I'm seeing teams forget that they need 5:27to rethink how they make decisions 5:29because now AI is 10xing or 100xing our 5:32optionality. If you can get 10 5:34iterations of a marketing message, you 5:36should probably rethink how you do your 5:37process. But we're not doing 5:40that. And so my sense is the gap between 5:45these approaches is getting wider all 5:47the time. AI models are continuing to 5:50grow. They're continuing to become more 5:52powerful. They're supercharging the few 5:54teams that know how to use them well. 5:56They're offering real benefits to 5:58individuals, but by and large, teams are 6:00being left behind. And I think teams are 6:02a fundamental unit of work. I opened 6:05with the story of the space shuttle for 6:06a reason. Making things is hard. Humans 6:10have figured out that making things 6:12together allows us to do more than we 6:15could do individually. And we have to 6:17figure out how to reformat the team so 6:20that AI can be a fully functioning team 6:23member. Otherwise, we're going to 6:25consistently underestimate the real 6:27potential of AI. Not because we don't 6:29understand the intelligence AI brings to 6:31the table, but because we don't 6:32understand how to make that intelligence 6:35social. The lesson here isn't just like 6:38document everything in markdown. It's 6:40about understanding that intelligence 6:43operates in the spaces between people. 6:46And now we have to think about 6:48intelligence operating in the spaces 6:50between people and also AI. Ultimately, 6:54that's what's going to help teams be 6:55more productive. And so my challenge to 6:57you is basically how is your team 7:00thinking about AI collectively? Are you 7:01just using AI to speed up old patterns? 7:04Are you actually engaging with it as a 7:06form of collective intelligence? I'm 7:08really curious to hear about this. So, 7:10drop a comment if you want to read more. 7:12I actually wrote up um a long substack 7:15on this that talks through like eight 7:18different signals of high performing 7:19teams, signals of low performing teams, 7:21how you break through. Lots of good 7:23stuff in there. And if you don't want to 7:25read it, I don't care. That's fine. Um 7:27yeah, tell me how you are getting to 7:30collective intelligence with your AI 7:33usage on your teams. Cheers.