Sam Altman aimed for 'automated research intern' in 2025, now OpenAI shares update

Nearly a year after OpenAI CEO Sam Altman directed his teams to create an "automated research intern," the artificial intelligence company has unveiled that it has achieved its goal, with aspirations to develop a fully autonomous AI scientist by March 2028. In an extensive update, the AI laboratory explained how autonomous programming tools are transforming their day-to-day research operations.

Instead of manually drafting standard scripts, staff members are delegating complex responsibilities to automated assistants. This shift has resulted in an increased experimentation volume, swifter software releases, and a focus on supervising AI-generated code. Furthermore, the technology has taken over corporate troubleshooting as well. Internal IT support requests have decreased by more than 50% since January, with staff relying on autonomous bots to address software issues.

OpenAI's core research teams have shown a significant increase in AI adoption over the past three months, with the median researcher spending over $600 in computational tokens daily by mid-August, a sharp rise from the $162 daily benchmark recorded in July. The top 10% of researchers are now utilizing up to $7,000 in tokens daily. These financial figures represent estimated market rates based on external API pricing schedules, rather than OpenAI's internal production costs.

In June, Altman disclosed that OpenAI's internal processes consume 100 billion tokens monthly. A developer, Peter Steinberger, who joined the lab in February, shared an internal dashboard in May displaying an estimated $1.3 million monthly usage invoice, which was entirely covered by the employer.

The rapid advancement of programming agents is causing a divide within the broader technology sector, with some productivity enthusiasts supporting the use of synthetic agents, while others remain skeptical about the workflow. Some enterprise firms previously promoted the use of generative models, establishing internal scoreboards to recognize the most active users.

This fueled a "tokenmaxxing," where employees maximized AI interactions without considering practical outcomes. However, as mounting operational costs overtook measurable efficiency gains, major enterprises, including Amazon and Meta, abandoned their internal usage leaderboards.