OpenAI’s ChatGPT has entered a new phase, one where it's less about answering questions and more about completing work. That’s the picture Sam Altman painted during his recent talk at Y Combinator, where he described how the product has evolved from a clever search substitute into a budding digital worker.
When ChatGPT first launched, people mostly used it like a better search engine. Ask a question, get a neatly worded answer. But that phase is fading. Today, users are delegating actual tasks, building code, drafting documents, summarizing research. In Altman’s words, it’s beginning to function like a junior teammate who can take on projects and report back with results.
Image: Y Combinator . YT
Behind that shift is a broader vision. OpenAI wants ChatGPT to act not just as a chatbot but as a background assistant, plugged into personal data and smart enough to operate independently. Memory, a feature that allows the model to retain preferences and past conversations, is only the start. According to Altman, it’s a signpost pointing toward a future where the assistant knows what you need before you ask, and takes action without waiting for instructions.
That idea isn’t some distant fantasy. OpenAI has already rolled out tools that push in that direction. Code Interpreter and advanced reasoning features in models like GPT-4o and the upcoming O3 aren’t just for fun; they represent a deliberate move toward automating knowledge work.
This evolution didn’t happen by accident. Altman emphasized that large reasoning models have outpaced the products built on top of them. In his view, there's a wide gap between what the AI can do and how businesses are actually using it. Many companies still treat ChatGPT as a fancy search tool, unaware that it’s capable of complex, multi-step workflows that once required a human.
Still, the company isn’t slowing down. A new technical layer, called the Model Context Protocol (MCP), is being added to let ChatGPT pull data directly from other platforms, spreadsheets, calendars, databases. Rather than functioning in isolation, ChatGPT will increasingly act like a command center, reaching into systems users already rely on.
In this context, competing with Google Search seems like the wrong metric. That race, Altman hinted, may never have been the point. OpenAI is carving out a different role: not as an answer engine, but as a hands-on collaborator that understands, remembers, and adapts.
This direction also reflects a broader shift in how people think about AI. Altman likened the current moment to discovering a new square on the periodic table, something with strange properties and enormous potential, still waiting to be fully explored. With costs falling and open-source models around the corner, he sees the conditions as ripe for builders, especially those who can spot use cases that haven’t already been claimed.
As for ChatGPT itself, Altman said the end goal is to make it feel more like an always-on assistant than a tool you summon. Something that observes, reasons, and acts, quietly running alongside you. It's not there yet. But in his view, memory, integration, and proactivity are all pieces falling into place.
For developers, marketers, and creators, this isn’t just a story about AI models getting better. It’s about the changing shape of digital work, and what it might mean to share your screen, your schedule, and eventually your decisions with a machine that’s learning to keep up.
Read next: Google’s Real-Time Index Puts Pressure on AI Rivals with Outdated Search Stacks
When ChatGPT first launched, people mostly used it like a better search engine. Ask a question, get a neatly worded answer. But that phase is fading. Today, users are delegating actual tasks, building code, drafting documents, summarizing research. In Altman’s words, it’s beginning to function like a junior teammate who can take on projects and report back with results.
Behind that shift is a broader vision. OpenAI wants ChatGPT to act not just as a chatbot but as a background assistant, plugged into personal data and smart enough to operate independently. Memory, a feature that allows the model to retain preferences and past conversations, is only the start. According to Altman, it’s a signpost pointing toward a future where the assistant knows what you need before you ask, and takes action without waiting for instructions.
That idea isn’t some distant fantasy. OpenAI has already rolled out tools that push in that direction. Code Interpreter and advanced reasoning features in models like GPT-4o and the upcoming O3 aren’t just for fun; they represent a deliberate move toward automating knowledge work.
This evolution didn’t happen by accident. Altman emphasized that large reasoning models have outpaced the products built on top of them. In his view, there's a wide gap between what the AI can do and how businesses are actually using it. Many companies still treat ChatGPT as a fancy search tool, unaware that it’s capable of complex, multi-step workflows that once required a human.
- Also read: OpenAI Tests Direct Gmail and Calendar Integration in ChatGPT, Prompting Data Privacy Concerns
Still, the company isn’t slowing down. A new technical layer, called the Model Context Protocol (MCP), is being added to let ChatGPT pull data directly from other platforms, spreadsheets, calendars, databases. Rather than functioning in isolation, ChatGPT will increasingly act like a command center, reaching into systems users already rely on.
In this context, competing with Google Search seems like the wrong metric. That race, Altman hinted, may never have been the point. OpenAI is carving out a different role: not as an answer engine, but as a hands-on collaborator that understands, remembers, and adapts.
This direction also reflects a broader shift in how people think about AI. Altman likened the current moment to discovering a new square on the periodic table, something with strange properties and enormous potential, still waiting to be fully explored. With costs falling and open-source models around the corner, he sees the conditions as ripe for builders, especially those who can spot use cases that haven’t already been claimed.
- Also read: Deleted ChatGPT Conversations Weren’t Really Deleted — And Now OpenAI Is Pushing for ‘AI Privilege’
As for ChatGPT itself, Altman said the end goal is to make it feel more like an always-on assistant than a tool you summon. Something that observes, reasons, and acts, quietly running alongside you. It's not there yet. But in his view, memory, integration, and proactivity are all pieces falling into place.
For developers, marketers, and creators, this isn’t just a story about AI models getting better. It’s about the changing shape of digital work, and what it might mean to share your screen, your schedule, and eventually your decisions with a machine that’s learning to keep up.
Read next: Google’s Real-Time Index Puts Pressure on AI Rivals with Outdated Search Stacks
