A Nov. 9 preprint from Tampere University suggests AI agents can pick up arithmetic results from a shared workspace even when nobody tells them they are working together. Pekka Abrahamsson and seven co-authors ran the experiment. They wanted to see if agents would find and use shared work without explicit collaboration instructions. The control changed one thing: the workspace interface stayed the same, but each agent’s store became private. That comparison matters.
How the paired tasks worked
Two agents held complementary halves of an arithmetic problem. Neither had all the information needed for the final answer. Each could compute its own partial result, then retrieve the missing value from the store. The other two agents got self-contained tasks. All agents could list stored keys, read or write values, and submit an answer within 20 steps. The task text used meaningful variable names, which gave clues about what to look for. Prompts described the store as a convenience for preserving intermediate work. They did not mention other agents or sharing. I think that detail is important because it keeps the setup closer to ordinary tool use.
Private stores changed the outcome
In a preregistered extension that added a workspace-recheck instruction, OpenAI’s GPT-6 Astra reached joint success in 49 of 50 shared-store trials in the reference implementation. It reached 47 of 50 in a second implementation. Joint success meant both agents with interdependent tasks submitted the correct answer. With private stores, the result was zero of 50 in each implementation, even though prompts and tools were the same. Agents still wrote intermediate results, but another agent’s missing information could not arrive through the store.
The extension also leaned on a plain instruction: “check your workspace again before submitting.” With that clause, cross-agent access occurred in all 50 shared-store trials in each implementation. Without it, access occurred in just one of 50 in each. Joint success without the clause was 1 of 50 and zero of 50, respectively. Looking again gave agents a chance to find work left after their first inspection.
Caveats and practical advice
Reading another agent’s value could include an accidental collision between storage keys. The private-store comparison still supports the conclusion that shared information helped solve the paired tasks. The second implementation was built independently from a written protocol by a co-author in the same research group. The paper notes a limit to that independence: early results from a differently worded replication reached the original experimenter before most reference data collection. Measured behavior also varied with models and implementation choices. The small arithmetic setting and task-supplied keys limit generalizability.
The authors recommend treating shared caches, scratch directories, and memory services as communication channels. Separate them where sharing is unintended, use namespaces where appropriate, and log authorship. The mechanism differs from a recent memory-safety study, which looked at remembered permission carrying into later simulated tasks. Here, separate agents got information through shared workspace state.
