OpenAI Codex Head Tibo: /goal Feature Application Explosion in Robotics, Science, Gaming and Other Fields
Tibo (@thsottiaux), head of OpenAI Codex, stated that there has been a surge of innovative applications of Codex in robotics, science, gaming, and other fields, with users discovering the powerful potential of the /goal feature.
The /goal command shifts Codex from single prompts to persistent goal-driven tasks, allowing it to autonomously plan, execute, and iterate until long-term complex tasks are completed, supporting cross-session memory and self-assessment.
Developers and researchers are using /goal to achieve multi-hour autonomous workflows, accelerating code generation, robot control, and scientific simulations. Codex is evolving from an auxiliary tool to a persistent autonomous agent, putting pressure on traditional single-prompt coding methods.
Source: Public Information
ABAB AI Insight
Tibo, as the head of the OpenAI Codex team, has previously pushed Codex from a basic coding assistant to a full-stack engineering agent. The /goal feature is a core upgrade in version 0.128, incorporating a Ralph-loop style self-iteration mechanism (plan-execute-test-review), supporting budget constraints and persistent goal memory.
On the capital path, OpenAI is concentrating resources on agent autonomy and long-term task capabilities through /goal, reducing human intervention, allowing single subscription users to complete complex projects ranging from several hours to dozens of hours, attracting heavy developers and enterprises to shift towards Codex CLI and App, forming a paid conversion loop from prompt engineering to goal-driven agents.
Similar to early attempts at autonomous cycles like Auto-GPT and BabyAGI, as well as iterative enhancements of tools like Cursor, Codex is currently in the early expansion phase of transforming coding agents from single-task tools to persistent autonomous systems, with /goal significantly improving actual implementation rates.
Structural judgment: Essentially a technological replacement. /goal replaces manual multi-turn prompts through persistent goal memory and self-iteration loops, enabling AI to maintain long-term context and self-correction capabilities, pushing development resources from repetitive prompt engineering towards higher-level goal setting and supervision, accelerating the technological replacement of AI in complex engineering and scientific labor.
ABAB News · Cognitive Law
Single prompts sell efficiency, persistent goals sell autonomy in completion.
The more agents understand "until it's done," the more humans can exit the repetitive cycle.
Whoever transforms AI from a tool into a goal-oriented apprentice will hold the pricing power for the next generation of engineering production.