A-SWE Achieves Full-Process Automated Development: Five-Step Strategy Reveals the Evolution Path of AGI

At a recent Goldman Sachs event, OpenAI CFO Sarah Friar revealed that beyond the existing Deep Research and Operator agents, the company will soon launch Agentic Software Engineer (A-SWE) – a fully autonomous developer capable of handling not just code assistance but the entire software development lifecycle, including QA, bug testing, and documentation. Turns out the CFO is now outshining Sam Altman in tech revelations! 😅

Sarah outlined OpenAI’s five-phase journey toward AGI:

1. Chatbot Stage: For example, the “real-time answering” and text generation capabilities initially provided by ChatGPT;

2. Reasoning Stage: The GPT series emphasized “reasoning ability” in 2024. For instance, the O-series models can perform “Chain-of-Thought” reasoning, examining the context and iteratively revising their responses when tackling complex questions, much like how humans do.

3. Agents: 2025 is referred to as the “Agent Year.” OpenAI has currently launched and is about to release three “Agent” products:

• Deep Research: Capable of autonomously conducting in-depth research, automatically summarizing, and generating professional reports.
• Operator: Acts as an online “task agent,” assisting with booking flights, hotels, restaurants, and more.
• A-SWE (Agentic Software Engineer): Able to independently complete software development, QA, bug testing, and documentation writing, essentially functioning as an “automated developer.”

4. Innovation Stage: The model begins to not only rely on existing human knowledge but also generate entirely new insights, giving rise to original academic or technological discoveries.

5. Agentic Organizations: AI can take on more organization-level decision-making and operational tasks, deeply penetrating various levels of the economy and society.