A Labor Day special: agents have spent 2026 transforming individual work, but the half of knowledge work that happens between people remains untouched. NLW makes the case that the frontier is shifting from single player to multiplayer AI — shared context, shared sessions, shared agents — and launches a free four-week sprint to help your team get there first.
The next frontier of agents is multiplayer: from individual leverage to team capability.
Agents have only been able to touch the roughly half of work we do alone — but the majority of knowledge work runs through team context. Every, Anthropic's Claude Tag, the OpenClaw 2.0 rebuild, and now Y Combinator's fall request for startups all point the same direction: agents are moving out of individual silos and into shared spaces, with team-owned context, visible work, and live participation. To help teams get ahead of the shift, the show is launching the Multiplayer AI Sprint — a free, four-week, self-directed program to inventory your team's AI usage, build shared context, map overlapping work, and ship your first shared agent.
2026 is undisputedly the year of agents — but so far they've changed how we work only on an individual level. Most of us split work between solo and team contexts, and the best, most dynamic AI-using teams are about to shift from single player AI to multiplayer AI.
The AIDB New Year's Resolution, ClawCamp, Agent OS, and the AI Summer Adventure were all free, self-directed, project-based learning experiences. But like nearly all agentic work so far, they followed the same pattern: people building and leveraging individual agents for their individual work.
The pedagogy behind every AIDB program is simple: to learn how to use AI, you just have to use AI. The programs exist to provide a framework for actually going and doing the work.
A survey of roughly 16,500 office workers found 39% of the workday is spent working alone versus 42% working with others. Another found 57% of time goes to communicating versus 43% creating individually, and a third put about 60% of time on 'work about work' — communication, search, coordination, and process.
Think about the early experiments you've heard about: researcher agents, writer agents, coding agents, the personal chief of staff. Even advanced users experimenting with agent teams are building teams that serve only the individual.
Individual agents aren't going away — everyone now gets to be a manager of an extensive team of agents that can spawn sub-agents. That's just part and parcel of being an effective knowledge worker. But it only covers the work we do in our own silos.
That means team-owned context — one single repository shared across the team, not duplicated documents in everyone's folders. It means shared sessions, observable work, and live steering and handoffs, with multiple people giving input at the same time in the same session.
From private outputs to visible work everyone can see. From feedback prompts after the fact to live participation while work is happening. From personal memory to shared context owned by the team, channel, or project. And ultimately from individual leverage to team capability — agents as reusable organizational infrastructure.
The team at Every started agentic experimentation with everyone owning an individual agent that mirrored its owner. Before long, they realized that wasn't how work actually got done and shifted to a model with agents living in shared spaces, doing work that intersected the team.
Unlike the original Claude-in-Slack integration, where you tagged in your personal Claude, Claude Tag instances are shared across an entire team via specific channels — each with its own context, tool access, and data access. When you tag Claude in your coding channel, it's the shared team agent, not yours.
Because there's one Claude per channel, anyone can see what it's working on and pick up where the last person left off — Anthropic describes it as 'much more like interacting collaboratively with a teammate.' It follows the channel over time, builds context, and can even be set to take ambient initiative on relevant work.
Per Anthropic's own announcement, nearly two-thirds of the product team's code is now created by their internal version of Claude Tag — not individual developers running their own Claude agents to contribute PRs, but a shared space with a shared agent working between them.
After going quiet for about seven weeks to coordinate the 2.0 rebuild, the OpenClaw maintainers found that individual agents collaborating in Discord wasn't collaborative enough. They built a new multiplayer web UI so both developers could open the same session, see the same context, and jump in when the agent paused — no screenshots, no copied transcripts.
— Colin, OpenClaw maintainer, on the 2.0 rebuild. The feature that made the difference wasn't seeing another avatar online — it was sharing a session while work was happening, with either developer able to add information directly into the same context. What sounds like a small interface improvement changes the way you collaborate with an agent.
YC's quarterly RFS is a window into where a very advanced group of investors thinks the world is headed — and one of the fall 2026 themes is exactly this: turning AI from a chat box only you can see into shared live agent sessions anyone on a team can watch, redirect, and hand off.
— Aaron Epstein, Y Combinator partner, in the fall 2026 request for startups. Google Docs replaced Word; Figma beat Photoshop — solo tools turned into places where teams do their best work together. AI agents are the most powerful new tool a team has, Epstein argues, yet the one thing people still use by themselves: a prompt, an answer in a box, and at best a read-only transcript link.
Agents are starting to run tasks that take hours, days, even weeks — work at that scale was never meant to be done alone. YC sees a version of multiplayer agents for every kind of work: engineers coding in real time, sales teams working a deal, support teams resolving a ticket, lawyers drafting a contract, analysts building a model, marketers shipping a campaign.
The show's fifth free self-directed program of the year is the first built for teams rather than individuals: a four-session sprint to get your team ready for multiplayer agents and then actually try one in practice. Each session pairs individual work with a team meeting to share it, and an invite code puts your whole team in the same shared space.
Is everyone still just prompting ChatGPT or Claude? Has anyone built an agent that does recurring work, or used context files and skills? Almost every team has a wide range — and figuring out where everyone is is the essential first step before making the leap to multiplayer usage.
If your team is still nascent on AI, the inventory week still works two ways: audit what's blocked adoption so the group can address it together, or find advanced users elsewhere in your company and have them present how they use AI and agents within the same guardrails and governance your team faces.
Moving from single player to multiplayer AI means moving from individual memory to shared context — which requires the team to figure out what that shared repository needs to know. Each person extracts their own context (with the platform's AI, or a downloadable worksheet), then the team assembles it into one shared repository.
Session three maps where your work overlaps, then scores each candidate one-to-five on four dimensions: shared need (how many people need the same context), staleness cost (how much it hurts when versions drift), permission sensitivity (how much restricted data it touches), and checkability (can you quickly tell if the agent got it right). Squint at the highest scores for where to start.
Use tools you already have (Claude Tag makes it easy), load your team context, and have at least two people use one shared agent on real work. Some overlapping use cases just won't be the right shape for an agent yet — that's fine; move to the next candidate. The final session is designed to be run over and over.
You'll still run entire fleets of individual agents — but that only represents one part of the work we do. It's only natural that agents come for the other part: the work that lives between us. Teams that work through this shift now will be positioned to lead as multiplayer AI tools come online.