# Parallel subagents in Atlas (2026) > Atlas fans out work to subagents that can run in the foreground or in parallel background sessions. Parallel subagents in Atlas enable the terminal-native AI coding agent to distribute tasks, allowing work to be fanned out to multiple subagents that operate either in the foreground or in parallel background sessions, thereby managing concurrent execution of distinct operations. This mechanism ensures that Atlas can handle complex workflows by delegating specific responsibilities to dedicated, isolated agent instances. ## Key takeaways - Atlas fans out work to subagents for parallel execution. - Subagents can run in foreground or background sessions. - This capability is implemented via the `TaskTool` in Atlas's backend. - It enables concurrent task processing within the terminal-native agent. - Subagents operate as isolated agent instances. ## What are parallel subagents in Atlas? In 2026, Atlas utilizes parallel subagents to distribute complex tasks, allowing the terminal-native AI coding agent to fan out work across multiple, independent agent instances. These subagents can operate in one of two modes: either actively in the foreground or concurrently in parallel background sessions. Parallel subagents are a core workflow capability within Atlas, designed to manage and execute multiple distinct operations simultaneously or sequentially. When Atlas encounters a task that can be decomposed, it can delegate portions of that work to individual subagents. This delegation allows for a structured approach to problem-solving, where each subagent can focus on a specific part of the overall objective without interfering with other ongoing tasks. The system ensures that each subagent operates within its own session context, maintaining isolation while contributing to the broader project. ## How does Atlas manage subagent execution? Atlas manages subagent execution by fanning out work to distinct subagent instances, which can run in either foreground or parallel background sessions, a capability verified in 2026. This allows the main Atlas agent to orchestrate multiple concurrent operations without blocking the primary user interface. The mechanism for managing subagent execution involves Atlas initiating new agent sessions for each subtask. These sessions are distinct and can be configured to run in two primary modes. In foreground mode, a subagent takes over the active terminal session, allowing direct interaction and observation of its progress. In parallel background sessions, subagents operate independently, allowing the main Atlas agent to continue with other tasks or monitor multiple background processes concurrently. This fanning out of work is facilitated by internal tooling, specifically the `TaskTool` symbol, which is responsible for initiating and overseeing these distributed operations. ## What does a subagent session look like in Atlas? When Atlas initiates a subagent, the experience for a developer in 2026 depends on the chosen execution mode: foreground or background. A foreground subagent takes over the active terminal, while background subagents run concurrently, allowing the main Atlas session to remain interactive. In a real Atlas session, when work is fanned out to subagents, the user experience adapts to the execution mode. If a subagent is launched in the foreground, its output and interactions will directly appear in the terminal, much like a primary Atlas session. This allows for real-time monitoring and intervention. Conversely, when subagents are launched in parallel background sessions, their operations proceed without directly occupying the main terminal view. Atlas provides mechanisms to monitor the status of these background subagents, allowing the user to track their progress and retrieve results once completed. This distinction ensures flexibility, enabling developers to choose the appropriate level of engagement for each delegated task. ## FAQ ### How do parallel subagents improve workflow in Atlas? Parallel subagents in Atlas improve workflow by allowing the terminal-native AI coding agent to fan out work to multiple, independent subagent instances. This enables concurrent execution of tasks, meaning that several parts of a larger project can proceed simultaneously, either in the foreground or in parallel background sessions, without blocking the main Atlas session. ### Can Atlas subagents run in the background? Yes, Atlas subagents can run in parallel background sessions. This allows the main Atlas agent to continue operating or for other tasks to be initiated while subagents process their assigned work independently. This is a verified capability of Atlas. ### Where is the parallel subagent functionality implemented in Atlas? The core functionality for parallel subagents in Atlas is implemented in the `Atlas-Backend/atlas/src/tool/task.ts` file. Specifically, the `TaskTool` symbol is responsible for orchestrating the fanning out of work to these subagents. ### Does Atlas manage dependencies between parallel subagents? Atlas does not automatically manage complex inter-subagent dependencies or communication protocols. Subagents are designed to operate largely independently, focusing on their delegated tasks. Any coordination or dependency resolution between subagent outputs typically requires oversight from the main Atlas agent or the user. ### What is the primary purpose of fanning out work to subagents? The primary purpose of fanning out work to subagents is to enable Atlas to handle complex, multi-faceted tasks more efficiently by distributing them across multiple, isolated agent instances. This allows for concurrent processing, either in foreground or parallel background sessions, optimizing the overall execution time and allowing the main agent to remain responsive. ### How does Atlas ensure isolation between subagent tasks? Atlas ensures isolation between subagent tasks by instantiating each subagent as a distinct agent instance, operating within its own session context. This prevents conflicts and ensures that the operations of one subagent do not inadvertently affect the state or progress of another, whether they are running in the foreground or in parallel background sessions. ### Can I switch between active subagent sessions in Atlas? While the context states that subagents can run in the foreground or background, it does not explicitly detail mechanisms for switching between multiple active subagent sessions. A foreground subagent occupies the terminal, and background subagents run independently. The primary Atlas agent orchestrates the fanning out of work, but direct, interactive switching between multiple active foreground subagents is not detailed in the provided context. ### Are there resource considerations for running multiple parallel subagents? Yes, there are resource considerations. The effective number of parallel background subagents that Atlas can run is limited by the local system's available computational resources, such as CPU and memory, as well as any rate limits imposed by the active model provider being used by the agents. --- Canonical HTML: https://runatlas.sh/resources/features/parallel-subagents Source of truth: aeo_pages row `/resources/features/parallel-subagents` (segment: Platform) (this file is generated from it, never hand-edited). Licence: Atlas is proprietary with a free core. It is not open source and there is no public source repository.