AI as a 'junior colleague' goal
OpenAI’s chief scientist says the company is on track to build AI systems that function like research interns by 2026 and is targeting fully autonomous research capabilities later, framing the next benchmark as an AI ‘junior colleague’ rather than a smarter tool. The remark accompanies calls to rethink work patterns—OpenAI has encouraged trials like four‑day weeks as workplaces adapt to heavier AI use. That shifts hiring expectations: routine vetting and supervision may become more valuable than raw model‑building skills. (startupfortune.com) (africa.businessinsider.com)
OpenAI is no longer describing its next big step as a better chatbot. Its chief scientist, Jakub Pachocki, says the target is an artificial intelligence system that works like a human research intern by 2026, with a fully autonomous researcher later. (startupfortune.com) That is a different promise from “answers questions faster.” A research intern is expected to read papers, run tests, write code, follow up on dead ends, and come back with something useful after hours or days of work. (technologyreview.com) Pachocki said recent gains in coding and mathematical reasoning are the sign he was waiting for. In the new interview, he said he “definitely” sees OpenAI as “on track” toward that intern-level system. (businessinsider.com) (startupfortune.com) OpenAI has been talking about this timeline for months, not just this week. Reports from an October 2025 livestream said the company was aiming for an intern-level research assistant by September 2026 and a more complete automated researcher by 2028. (techcrunch.com) (officechai.com) The benchmark OpenAI keeps returning to is autonomy time. Pachocki has described progress in terms of how long a model can keep working on a task without a human stepping in, which is closer to judging a junior colleague than judging a search box. (technologyreview.com) At the same time, OpenAI is telling governments and employers to prepare for work changing around these systems. In a policy paper published on April 6, 2026, the company backed pilots of a 32-hour workweek with no loss in pay while keeping output and service levels unchanged. (openai.com) (cdn.openai.com) The company’s paper treats shorter weeks as one way to turn artificial intelligence productivity into time instead of just more tasks. It also pairs that with ideas like portable benefits and wider access to artificial intelligence tools, which suggests OpenAI expects disruption to hit job structures, not just software budgets. (cdn.openai.com) (openai.com) If the “junior colleague” frame sticks, hiring changes before full automation arrives. Teams would need more people who can assign work, check outputs, catch subtle errors, and decide when the model is confidently wrong, which is closer to managing a fast intern than replacing a whole department. (technologyreview.com) (openai.com) That also helps explain why OpenAI still hires early-career researchers while trying to build their software equivalent. Its residency program is built around mentorship, hands-on research, and judgment, and those are exactly the qualities companies will still need if artificial intelligence can do more of the drafting and less of the deciding. (openai.com 1) (openai.com 2) The short version is that OpenAI is trying to move the conversation from tool to teammate. Once a model can take a multi-day assignment, use code and papers as working memory, and return with a result that needs review instead of constant prompting, the office chart changes even if the headcount number does not. (technologyreview.com) (startupfortune.com)
