OpenAI Codex Can Now Fork Agents

OpenAI's Codex has evolved significantly from its 2021 debut as the model behind the first version of GitHub Copilot. Today, it operates as an autonomous agent in a secure cloud sandbox, capable of not just writing code, but also running tests, fixing bugs, and proposing complete pull requests for human review. The new forking capability is a step towards multi-agent systems, a paradigm where complex software problems are solved by a team of collaborating AIs. In these systems, different agents often take on specialized roles like "planner" to decompose tasks, "executor" to write code, and "reviewer" to validate the output. This update allows a single Codex agent to spawn parallel sub-agents, mirroring how a human team lead distributes work. For example, a developer could assign a high-level feature, and the primary agent could fork processes to have one sub-agent build the API while another simultaneously writes the integration tests for it. For startups, this approach can act as a significant force multiplier, enabling small engineering teams to accelerate development by tackling different parts of the codebase in parallel. This allows the human engineers to focus more on system design and product strategy, rather than getting bogged down in implementation details for every feature. This evolution in tooling also reflects a key career crossroads for engineers: the choice between the Individual Contributor (IC) and management track. Using an AI agent to complete a task is an IC function, while orchestrating multiple agents to achieve a larger goal is fundamentally a management-like activity, focusing on decomposition and oversight. The shift from single-prompt code completion to multi-agent orchestration signals a broader trend where AI assistants are becoming more like active collaborators. OpenAI underlined this direction with the February 2026 release of a dedicated macOS app, described as a "command center for agents" designed to manage these complex, parallel workflows.