OpenAI's chief scientist trusts AI with experiments but says it's not at the level to design complex systems

2026-03-21 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Intermediate, quick

Summary

OpenAI Chief Scientist Jakub Pachocki, who previously coded manually using Vim, now utilizes AI for experimental coding tasks that once consumed a full week, completing them over a single weekend. While he trusts AI with these specific experiments, he maintains that current AI capabilities are not advanced enough for complex system design. Pachocki's current focus is on developing an "autonomous AI research intern" by September, capable of handling multi-day tasks. By March 2028, OpenAI aims to deploy a "full AI Researcher" – a multi-agent system designed to independently solve complex problems across various scientific and economic domains, potentially leading to unprecedented concentrations of power as data centers could perform the work of entire corporations.

Key takeaway

For AI Scientists and Research Scientists evaluating AI's role in their workflows, recognize that while AI can significantly accelerate experimental coding, its current limitations necessitate human involvement for intricate design and strategic goal-setting. You should integrate AI for task automation where appropriate, but maintain direct oversight and leadership in defining research objectives and architecting complex systems to mitigate risks associated with concentrated AI power.

Key insights

AI excels at automating experimental coding tasks but lacks the sophistication for complex system design.

Principles

• AI utility varies by user and task.
• Human oversight remains crucial for goal setting.

Method

OpenAI is developing an autonomous AI research intern for multi-day tasks, evolving into a multi-agent "AI Researcher" by March 2028 to tackle complex problems across diverse fields.

In practice

• Delegate routine coding experiments to AI.
• Retain human control for complex design.

Topics

• AI Automation
• Autonomous AI Research
• OpenAI
• AI Development Tools
• AI Capabilities

Best for: AI Scientist, Research Scientist, AI Engineer, AI Researcher, CTO

Related on AIssential

• OpenAI is throwing everything into building a fully automated researcher — OpenAI is prioritizing the development of autonomous AI agents to conduct complex research, aiming for a "research lab in a data center."
• Stanford, Berkeley, MIT, UNC: AI Scientist AutoResearchClaw — Autonomous AI research systems require strategic human collaboration and robust verification to achieve meaningful scientific discovery.
• Andrej Karpathy on Code Agents, AutoResearch, and the Loopy Era of AI — AI agents are fundamentally transforming software development and research by enabling extensive task delegation and automation.
• Scientists Trapped 1000 AIs in Minecraft. And they Created A Civilization without being told to do so. — AI agents, given minimal prompts, can autonomously develop complex social structures and emergent behaviors.
• How much is AI really going to change the near future (5-20years)? — Elite scientists acknowledge AI's immediate supremacy in coding and analytical tasks, driving rapid adoption despite ethical and skill atrophy concerns.
• MirrorCode: Evidence that AI can already do some weeks-long coding tasks — AI systems are demonstrating the ability to complete complex, multi-week coding projects autonomously.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.