Justy: Okay so this post from Aatish Nayak is going around saying AI works great for individuals and kind of stinks for organizations, and the answer is treating agents like social participants in multiplayer stuff. I want to argue with whoever wrote this.
Cody: No I actually want to argue with you about it, because I think you're going to defend it and I'm going to be the grump. Right, let's do it.
Cody: So the claim is the gap is social, not technical. And my read is… that's a vibes argument. "Social participants" — what does that even mean operationally? Agents don't have social lives, Justy. They have context windows and tool lists.
Justy: Mm-hm.
Cody: When a team of humans fails to get something done, it's almost never because they're bad at being social. It's because they don't share context, they don't have shared state, the handoffs are garbage. The fix there isn't teaching the agents manners, it's plumbing.
Justy: Okay but here's where I think you're being a little bit of a grump on purpose — the post is gesturing at something real. When I use an agent solo it's basically a superpower. When I try to use it with my team, the second someone else touches the thread or the doc or the tool it's suddenly in a different universe.
Justy: And nobody's really built the layer that makes that feel continuous yet. So when he says "social participant," I read that as "there needs to be a shared substrate where multiple people and multiple agents are operating on the same picture at once." That's a real product gap.
Cody: Sure, but then call it shared context. Call it team memory. Don't call it social participants, because now every founder with a slack bot is going to slap "collaborative intelligence" on their landing page and we're back in the agent circus.
Cody: The thing is, individual AI tool adoption inside companies is exploding. People are getting value. The org-level disappointment is real but it's not because the agents aren't being polite to each other — it's because the agents literally cannot see what the other agents did, or what the humans decided five minutes ago.
Justy: Right, right.
Cody: And that's a hard engineering problem. Shared memory across sessions, across owners, across permissions. You don't solve that with vibes about intelligence being collaborative.
Justy: Okay fair. But here's the thing — when I talk to PMs building internal tools right now, the ones who are winning aren't bolting more capability onto a single agent. They're building the layer where a bunch of agents and a bunch of humans can work on the same artifact without stepping on each other.
Justy: That's literally the social participant thing, even if the post dresses it up in fancier language. The shift is from "agent assists me" to "agent operates in a room with other agents and people." That reframe changes what you build.
Cody: Mm-hm. And I'll give you that the reframe is useful for product people. I just think the post overclaims what the bottleneck is. Most teams I'd talk to don't need fancier social dynamics, they need one place where the context isn't getting re-explained every single session.
Cody: Like the actual moonshot isn't "agents that play well with others," it's "agents that remember what happened yesterday." That's way more boring and way more true.
Justy: Yeah, you're not wrong. But I'd push back on one thing — boring memory isn't the only gap. Coordination is. I have watched three different agents in a Slack thread try to answer the same customer, because none of them knew the other two had already replied. That's not a memory problem, that's a who's-doing-what problem.
Cody: Okay, that's a fair example. And honestly that one is closer to what the post is actually pointing at, even if the language is sloppy.
Cody: But here's my worry — if "collaborative intelligence" becomes the new buzzword the way "agentic" did six months ago, every enterprise vendor is going to claim they're doing it and the actual engineering underneath is just… shared state and permissions. Which, fine, that IS the work, but let's call the work what it is.
Justy: Right.
Justy: Okay so where do we land. I think the post names a real failure mode that people feel but can't describe yet, and "social participant" is a rough first attempt at language for it. You think it's a vibe dressed up as an argument.
Cody: Basically. The instinct is right, the specifics are missing. If you're building for teams right now, invest in shared context and coordination primitives before you invest in making your agent charmingly collaborative.
Justy: See, this is why I keep you around. You take my exciting reframe and you give it a to-do list.
Cody: Somebody has to. You'd be shipping vibes-as-a-service without me.
Justy: All right. Honestly though — if you're a PM listening to yourself complain that your agent works for you but not for your team, the diagnosis is probably not "we need smarter agents," it's "we need one brain for the team, not nine brains in silos." That's the actual unlock.
Cody: Yep. I'll take it. Episode five hundred and whatever, still arguing about buzzwords.
Justy: We'll see you next time.