Google Research this week introduced the /teamwork feature of Antigravity, which lets multiple AI agents divide and perform work in the same development environment while jointly advancing scientific and engineering projects. Our view is that the next barrier for foundation-model products is not merely answer quality, but whether they can decompose complex work into multiple, coordinated roles.
What this is
/teamwork can be understood as enabling a group of AI agents to complete a complex task together: they handle planning, coding, information retrieval, or result verification separately while sharing working context. Compared with a conventional chat assistant, it is closer to a digital work team that can advance multiple tasks in parallel.
Google initially focused the feature on scientific and engineering scenarios because they involve many interdependent steps and require both code and design proposals as well as research and logical validation. If agents can divide the work, they can reduce the cost of manually transferring information between stages.
Industry view
What is encouraging is that competition is shifting from “how good the code a model can write in one pass” to “whether a system can complete tasks reliably.” When individual models have similar capabilities, the system that can coordinate specialized agents, resolve conflicts, and leave an inspectable trail is more likely to enter real R&D workflows.
But the risks are clear: multi-agent systems do not automatically equal higher efficiency. If tasks are poorly divided or context is passed incorrectly, the system may generate more rework; when agents cite one another's faulty conclusions, they can amplify errors. Google has demonstrated a collaboration mechanism, not universal validation of scientific accuracy or engineering productivity. Permission controls, process tracing, and decisions about when humans should intervene remain critical to large-scale deployment.
Impact on regular people
For enterprise IT: Businesses will gradually assess whether AI can extend from supporting writing into R&D, testing, and information organization. But greater multi-agent permissions also mean that auditing, data isolation, and accountability must be addressed first.
For individual workers: Knowledge workers may increasingly adopt “AI teams” for research, proposals, and prototyping. In the short term, the change looks more like a redistribution of tasks than the immediate elimination of an entire job.
For the consumer market: Consumer AI products will provide more automatic task division and background collaboration. Users will see more complete results, but the process will also be harder to evaluate, requiring clearer explanations of results and robust rollback mechanisms.