subagents

Overview

Subagents are a workflow pattern in which a primary AI system delegates parts of a larger task to specialized helper agents that can work in parallel or handle narrower scopes. Instead of forcing one model instance to carry every detail of a complex job, the main agent can break work into smaller units, assign them to subagents, and then combine the results. This pattern is especially useful when tasks are too large, too long-running, or too context-heavy for a single agent to manage efficiently.

For AI Product Managers, subagents matter because they offer a practical way to improve speed, context management, and reliability in agentic products. They can help teams design systems that scale beyond single-threaded prompting, reduce the impact of LLM context limits, and support workflows like code migrations, research synthesis, and multi-step execution. As more platforms add native subagent support, this pattern is becoming an important product design primitive rather than just an implementation detail.

Key Developments

• 2026-01-23: Cursor introduced subagents for parallel task execution, highlighting faster execution, better context usage, and support for long-running tasks.
• 2026-02-22: Boris Cherny described using subagents with Git worktrees to parallelize large codebase migrations by assigning each agent a subset of folders while a main agent handled merge conflicts.
• 2026-03-17: OpenAI Codex announced general availability of subagents and support for custom agents, enabling patterns similar to Claude Code's explorer, worker, and default agent setup, including TOML-defined custom agents.
• 2026-03-18: Simon Willison explained subagents as a way to manage LLM context limits by splitting larger tasks into smaller agentic components, preserving top-level context and improving outcomes on tasks that exceed model memory constraints.

Relevance to AI PMs

• Designing around context limits: PMs can use subagent patterns to structure products that keep the top-level agent focused on orchestration while delegating detailed subtasks to helpers. This is a practical way to reduce context-window pressure and improve performance on complex workflows.
• Improving product speed and throughput: Parallel subagents can shorten time-to-result for tasks like code review, research, migrations, or document analysis. PMs should evaluate where parallelism creates real user value and where orchestration overhead may outweigh gains.
• Creating more configurable agent systems: As platforms add custom-agent support, PMs can define reusable worker roles, permissions, and task boundaries. This opens product opportunities around templates, specialized agent roles, and enterprise workflow customization.

Related

• claude: Claude has been associated with early popularization of subagent-style workflows, especially in coding contexts.
• claude-code: Claude Code includes agent patterns such as explorer, worker, and default roles that illustrate how subagents can be structured in practice.
• openai-codex: Codex added general availability for subagents and custom agents, signaling broader platform adoption of the pattern.
• custom-agents: Custom agents are closely related because subagent systems often rely on preconfigured specialist roles for delegation.
• llm-context-limits: One of the main reasons subagents matter is that they help work around model memory and context-window constraints.
• cursor: Cursor introduced subagents for parallel task execution, making the concept more visible in developer tooling.
• git-worktrees: Git worktrees pair well with subagents in software workflows by letting multiple agents operate in parallel on isolated code branches.
• toml: TOML is relevant because some platforms use it to define custom agent behavior and configuration.
• boris-cherny: Boris Cherny provided a concrete example of subagents in production-style migration workflows.