In 2025, Sam Altman set a goal for OpenAI engineers to create 'automated research intern', a year later company publishes a long update

Roughly a year after CEO Sam Altman instructed OpenAI engineering teams to design an “automated research intern,” the artificial intelligence (AI) company has announced that it has successfully fulfilled that objective, while setting its sights on engineering a fully independent AI scientist by March 2028. In a long update released, the AI Lab detailed how autonomous programming tools are overhauling its day-to-day laboratory operations.According to the long note, rather than drafting routine scripts manually, staff members are handing off complex duties to automated assistants, leading to higher experimentation volume, faster software releases and a shift toward supervising AI-generated code. It also said that the technology has also taken over corporate troubleshooting. Daily help requests submitted to an internal IT support desk run by human technicians have plunged by more than 50% since January, as staff increasingly rely on autonomous bots to resolve software problems.Skyrocketing compute usage and token consumptionWhile various companies have limited the use of AI, adoption among OpenAI’s core research teams has surged over the past three months, growing at a far steeper rate than anywhere else inside the firm. OpenAI revealed that the median researcher consumed over $600 in computational tokens daily by mid-August, representing a sharp increase from the $162 daily baseline recorded in July.The top 10% of researchers are now utilising upwards of $7,000 in tokens every single day. These financial metrics represent estimated market rates calculated using external API pricing schedules rather than the underlying internal production costs incurred by OpenAI.Notably, in June, Altman disclosed that the startup's heaviest internal consumer processes 100 billion tokens per month. Demonstrating that scale, OpenClaw developer Peter Steinberger, who joined the lab in February, shared an internal dashboard in May showing an estimated $1.3 million monthly usage invoice, though the expense was completely absorbed by the employer.Corporate reckoning over 'Tokenmaxxing'The rapid evolution of programming agents is dividing the wider technology sector between productivity enthusiasts and corporations. While numerous software engineers say handing off baseline tasks to synthetic agents enhances their output, others remain skeptical about the workflow.Early in the year, enterprise firms actively nudged employees to use generative models, setting up internal scoreboards to highlight the most active users. The fueled a “tokenmaxxing,” where workers maximised AI interactions without prioritising practical outcomes.As mounting operational bills began to outstrip measurable efficiency gains, major enterprises pulled back. Tech firms like Amazon and Meta disbanded their internal usage leaderboards.Get the latest technology news and updates. Download the TOI App.