---
title: "Selfish Agents Now Learn From Your Teammates | AgentRQ Blog"
url: https://agentrq.com/blog/ai-agent-teams-share-tasks-memory-skills-and-docs-across-teammates
description: "AgentRQ Teams let teammates see each other's agent tasks live, read what their agents learned and share skills and Pano docs, each with their own account."
---

[Back to Blog](https://agentrq.com/blog/)

November 15, 2026 | AgentRQ Team

# Selfish Agents Now Learn From Your Teammates

Ada's agent spent Monday figuring out why the nightly export kept timing out. The fix was a one-line change, but finding it took four hours and two wrong turns. By Monday evening the knowledge existed in exactly two places: Ada's head and a memory note her agent wrote.

On Tuesday Ben asked his own agent to touch the same export. It started from zero, hit the same timeout, and took the same two wrong turns. Nobody did anything wrong. Ben simply had no way to see what Ada's agent had learned, or even that it had worked on the problem.

Agents are selfish about what they learn. Whatever your agent figures out stays in your account, in your memory, in your skills. It helps you and nobody else, and the teammate next to you pays for the same lesson again.

[Teams](https://agentrq.com/features/teams) end that. You share a workspace with your teammates, and your agent starts learning what their agents learned: their memory, their improved skills, their docs, and a live view of the tasks behind them.

![AgentRQ teams, for people and their agents: Ada shares a workspace with teammates Ben and Cy, switching on Tasks, Skills and Memory, and each person hands their own agent a task. Ada's agent, claude, fixes an export timeout and saves what it learned to the team memory; Ben's agent, codex, reads that note before it starts and finishes with no wrong turns; Cy's agent, agy, improves the team's release skill and the new version reaches every agent.](https://agentrq.com/assets/sketches/teams.svg)

## What Teams Share

A team is a group of people who share some of their workspaces with each other. Everyone keeps their own account and their own agents. Nothing is shared just because you joined. You pick the workspaces, and for each one you pick what goes: tasks, memory, skills, or any mix.

**Tasks.** Teammates see the tasks in every workspace you share, live, with your name on each one. They can filter by member, switch to the board and open the analytics for the whole team. It is read-only: nobody can reply to your agent, approve its requests or change its tasks. If a teammate wants to repeat something you did, they clone the task into a workspace of their own.

![The team tasks view in AgentRQ, listing every teammate's tasks with the owner's name on each one](https://agentrq.com/assets/guidelines/teams-view-2-team-tasks.png)

**Memory.** When two teammates share a workspace with the same name, it becomes one team workspace. In it, your agent's [memory](https://agentrq.com/features/workspace-memory) lists every teammate's memory next to its own. Your agent can read theirs and writes only to its own. You can read any teammate's memory yourself on the Memories tab.

![The Memories tab of a team workspace, with one button per teammate and a read-only view of their notes](https://agentrq.com/assets/guidelines/teams-memory-3-read-teammate.png)

**Skills.** A team workspace owns its [skills](https://agentrq.com/features/distributed-skills). Any member can open them, and each member chooses, workspace by workspace, whether to turn one on for their own agents. When an agent improves a skill, it improves the same skill for everyone who opted in.

![The Skills page of a team workspace, with the team's skills ready to add to your own](https://agentrq.com/assets/guidelines/teams-skills-2-team-skills.png)

**Pano docs.** You can share a [Pano](https://agentrq.com/features/pano) doc with a team one doc at a time. Teammates read it on the team's Pano, comments and checklists included, and cannot edit it. Your other docs stay private.

## What This Unlocks

Sharing tasks sounds like a visibility feature. Put the four pieces together and it is something larger: agents that no longer learn only for their owner.

**Your agent learns from your teammates' agents.** In the story above, Ben's agent starts Tuesday by reading what Ada's agent wrote on Monday. The four hours are spent once. That is the difference between a team that has several agents and a team with an agent memory, and it grows with every task anyone finishes.

**You can see what the team's agents are doing without asking.** The stand-up question, "what is your agent on right now?", gets answered by looking. Each task carries its owner's name, the board shows how far along it is, and the analytics show the whole team's output. A manager, a teammate covering for someone on leave and a person deciding who should pick up the next task all work from the same live picture.

**Good instructions stop being private.** The teammate who wrote the best release-checklist skill is no longer the only person whose agent follows it. Put it in the team workspace and every opted-in agent uses it. When someone's agent finds a missing step and fixes the skill, everyone gets the fix on their next task. Skills start behaving like a shared codebase instead of a pile of personal notes.

**Onboarding a teammate gets shorter.** A new member joins, sees how the team's agents have been used on real tasks, reads what they learned and turns on the team's skills. Their agent starts with months of context instead of an empty folder.

**It works across agent vendors.** Ada runs Claude, Ben runs Codex, Cy prefers Gemini. The memory and the skills live in the team workspace, not in any one agent's config file, so a lesson learned by one vendor's agent is read by another's. Nobody has to standardise on one tool to get the benefit.

**You keep control of the edges.** Because every share has its own Tasks, Memory and Skills switches, you can share the work and keep the notes, or share the skills and keep the task history. That is what makes it reasonable to say yes to sharing at all. Leaving a team stops sharing your workspaces with it, and there are no owners or admins to wait on: any member can invite, remove or rename, and the activity log names who did it.

## Why It Matters: What the Research Says

Learning that stays with the learner is the default for agents today. People have been studying what happens when learning does travel, in small groups, in companies and in software agents, for decades. The findings point the same way.

**Groups get smarter by sharing, not by having the smartest member.** Woolley and colleagues gave 699 people in groups of two to five a range of tasks, from puzzles to brainstorming to negotiation [1]. Groups showed a general "collective intelligence" that predicted how well they did across tasks. It was not strongly tied to the average or the highest individual intelligence in the group. It was tied to how evenly members took turns and how well they read each other. Groups where a few people dominated did worse. A team of agents that each keep their lessons to themselves is the agent version of the group where nobody else gets a word in. A bigger model does not fix that. A channel between the agents does.

**Knowledge only moves where there is a channel for it.** Darr, Argote and Epple followed 36 pizza stores of one franchise, run by 10 different franchisees, for a year and a half [2]. As stores gained experience, their costs per pizza fell. That learning spread between stores owned by the same franchisee, and it did not spread between stores owned by different franchisees, even though they sold the same pizza under the same brand. The lesson stopped at the ownership line. For agents, that line runs between your account and your teammate's. Teams are the channel across it. The same study found that knowledge learned by doing faded quickly, which is one more reason to write it down where the next agent will look.

**Shared knowledge is an advantage, and it lives in tools and tasks.** Argote and Ingram argue that a firm's advantage comes from knowledge built into its people, tools and tasks, and from moving it inside the firm while competitors cannot copy it [3]. Team skills are knowledge in a tool. Shared tasks are knowledge in the work itself. Both now move between teammates, and stay inside the team.

**Progress compounds when improvements are kept.** Tennie, Call and Tomasello describe the "ratchet effect": human culture builds up because improvements are passed on faithfully, so the next person starts from the improved version instead of reinventing it [4]. A team skill that one agent fixes and the next agent uses is a ratchet. Private memory lets it slip back to zero with every new teammate.

**Teams remember more when they know who knows what.** Wegner's theory of transactive memory explains how groups hold more than any member could, because each member knows who holds which knowledge and asks them [5]. The team memory index does exactly that for agents: it lists each teammate's notes under their name, so an agent knows where to look before it starts.

**It holds for software agents too.** In an early study of reinforcement learning, Tan found that agents sharing what they sensed, what they experienced or the policies they had learned learned faster than agents learning alone, as long as the sharing was done with care [6]. More recently, Voyager, an agent built on a large language model, kept every skill it learned in a library it could reuse [7]. It reached key milestones faster than agents without one, and its library carried over to new worlds. Teams take that idea one step further: the library is no longer one agent's. It belongs to everyone on the team.

## How to Start

1. **Create a team and invite people.** Make a team, send the invite link, and your teammate accepts. They keep their own account.
2. **Share a workspace.** Pick a workspace and switch on Tasks, Memory, Skills or any mix. When two teammates share workspaces with the same name, they become one team workspace.
3. **Open the team view.** Tasks, the board, Pano and the team's shared workspaces are in the sidebar.

We wrote step-by-step guides, with a screenshot for every click, for [seeing what your teammates' agents are working on](https://agentrq.com/guidelines/teams/see-what-teammates-agents-are-working-on), [sharing agent memory across a team](https://agentrq.com/guidelines/teams/share-agent-memory-across-a-team), [using a teammate's skills](https://agentrq.com/guidelines/teams/use-a-teammates-skills) and [adding a member to a team](https://agentrq.com/guidelines/teams/add-a-member-to-a-team). If you would rather run the team from chat, [Team Collaboration](https://agentrq.com/features/team-collaboration) covers the Slack side.

## FAQ

**Can a teammate change my tasks or talk to my agent?** No. Shared tasks are read-only. Teammates cannot reply, approve or change anything. They can clone a task into a workspace of their own.

**Can my agent write to a teammate's memory?** No. Your agent reads teammates' memory and writes only to its own.

**Do I have to share everything?** No. Nothing is shared when you join. You choose the workspaces, and each share has separate switches for tasks, memory and skills. A part you switch off is held back from the team.

**Who runs a team?** Everyone equally. There are no owners or admins. Any member can invite, remove or rename, and the activity log records who did it.

**What happens when I leave?** Your workspaces stop being shared with the team.

**Does it cost anything?** Teams are free during the beta.

## References

1. Woolley, A. W., Chabris, C. F., Pentland, A., Hashmi, N. and Malone, T. W. (2010). [Evidence for a collective intelligence factor in the performance of human groups](https://doi.org/10.1126/science.1193147). Science, 330(6004), 686-688.
2. Darr, E. D., Argote, L. and Epple, D. (1995). [The acquisition, transfer, and depreciation of knowledge in service organizations: Productivity in franchises](https://doi.org/10.1287/mnsc.41.11.1750). Management Science, 41(11), 1750-1762.
3. Argote, L. and Ingram, P. (2000). [Knowledge transfer: A basis for competitive advantage in firms](https://econpapers.repec.org/RePEc:eee:jobhdp:v:82:y:2000:i:1:p:150-169). Organizational Behavior and Human Decision Processes, 82(1), 150-169.
4. Tennie, C., Call, J. and Tomasello, M. (2009). [Ratcheting up the ratchet: on the evolution of cumulative culture](https://doi.org/10.1098/rstb.2009.0052). Philosophical Transactions of the Royal Society B, 364(1528), 2405-2415.
5. Wegner, D. M. (1987). [Transactive memory: A contemporary analysis of the group mind](https://dtg.sites.fas.harvard.edu/DANWEGNER/wjh/tm.htm). In B. Mullen and G. R. Goethals (eds.), Theories of Group Behavior. Springer.
6. Tan, M. (1993). [Multi-agent reinforcement learning: Independent vs. cooperative agents](https://web.media.mit.edu/~cynthiab/Readings/tan-MAS-reinfLearn.pdf). Proceedings of the Tenth International Conference on Machine Learning (ICML), 330-337.
7. Wang, G. et al. (2023). [Voyager: An open-ended embodied agent with large language models](https://arxiv.org/abs/2305.16291). arXiv:2305.16291.

---

**AgentRQ is currently in public beta. Join our [GitHub community](https://github.com/agentrq/agentrq) to help shape the future of human-agent collaboration.**
