Sam Altman's Latest Interview: AI Agents Will Conduct Independent Research in 2026, ChatGPT Becomes an 'Operating System'
Honestly, when I saw this interview with Altman, I was eating takeout and scrolling through my phone. When he said ‘AI will be capable of independent scientific research by 2026,’ I almost dropped my chopsticks.
This isn’t science fiction—it’s OpenAI’s CEO speaking in a 36Kr interview.
From ‘Talking’ to ‘Doing’
Altman shared an interesting observation: early users treated ChatGPT as a search tool, younger users see it as a life assistant, and college students are already using it like an operating system.
From my experience, this observation rings true. The heaviest AI users I know are students. They use AI to write code, revise papers, analyze data, even generate presentations—not asking ‘how to do it’ but directly asking AI to ‘do it for me.’
This generational difference in usage habits essentially reflects the leap in AI capabilities.
Altman revealed the roadmap:
Phase 1 (now): AI can chat and generate content, but each step requires human guidance
Phase 2 (2026): AI gains autonomous planning and closed-loop execution, capable of independent research
Phase 3 (future): AI operates like an OS, understanding your complete life context and collaborating across apps
It’s quite interesting. If AI can truly conduct independent research, what becomes of scientists’ roles?
The ‘Operating System’ Ambition
Altman described a scenario: future ChatGPT remembers all your conversations, emails, schedules, then proactively completes tasks based on this information.
For example, you tell AI ‘help me arrange next week’s business trip.’ It doesn’t just give advice—it checks flights, books hotels, emails colleagues, syncs calendars—all without step-by-step guidance.
This reminds me of something my mom always says: ‘The best service is when you don’t feel the service exists.’
But honestly, I have mixed feelings about this vision. On one hand, it’s convenient. On the other, letting one system hold so much personal information carries risks.
Altman himself admits current algorithms have 10-100x room for improvement. In other words, today’s AI is still far from being an ‘operating system.’
Strategy for Ordinary People
One detail from the interview stuck with me: Altman said younger generations already view AI as an operating system, not just a search tool.
This is actually a signal for ordinary people—how you use AI determines how much value you get from it.
My suggestions:
First, don’t use AI only as a search engine. Asking ‘what is quantum computing’ versus ‘explain quantum computing in simple language with three real-life examples’ yields completely different value.
Second, try asking AI to ‘complete’ rather than ‘teach you how.’ Instead of asking ‘how to write a Python script for Excel,’ just give AI the Excel file and let it generate runnable code.
Third, pay attention to AI Agent development. 2026 may be the inflection point for large-scale agent adoption—those who understand this early will have first-mover advantage.
Final Thoughts
After watching this interview, my biggest takeaway is: the AI industry is shifting from ‘technology demonstrations’ to ‘product deployment.’
Previously, companies competed on model parameters and benchmark scores. Now they’re competing on who can build genuinely useful products.
Altman predicts agents will go mainstream this year, which I think is optimistic. But looking at changing user behaviors, AI is indeed evolving toward ‘autonomous execution.’
For us ordinary people, rather than worrying whether AI will replace jobs, we should first learn to use AI effectively. After all, people who know how to use tools will always be more valuable than the tools themselves.