OpenAI co-founder Greg Brockman is describing a future in which ChatGPT is less like another app to manage and more like a background layer for getting digital work done. The vision is simple to state: people should not have to learn software, and the interface itself should fade away.
That ambition also comes with a clear warning from OpenAI's own recent product history. The company has already tried to connect ChatGPT to web search and third-party apps, and Brockman now says the models were not ready for that earlier approach.
Why the old Plugins push fell short
OpenAI launched Plugins in 2023 to expand what ChatGPT could do. The feature was meant to bring web search and third-party apps like Gmail into the ChatGPT experience.
According to Brockman, the problem was not just product packaging. It was model capability. He described the result plainly:
"That didn't work. It didn't work at all because the models weren't ready,"
That statement matters because it reframes the failure. The issue was not simply that users needed a cleaner menu, a better button, or more app connections. The deeper challenge was that the AI system behind the product could not yet carry the workload reliably enough.
For anyone watching AI product launches, that lesson is important. A feature can sound powerful when it is introduced, especially when it promises to connect everyday software tasks. But if the model cannot understand the context, choose the right action, and execute with enough consistency, the interface around it cannot solve the core problem.
The new goal is less product, not more
Brockman's next direction is notably different from adding more visible features. He argues that people should not need to learn software at all. Instead, ChatGPT should become an invisible layer that accepts digital tasks and handles them on the user's behalf.
His description points away from the standard software model, where users learn dashboards, settings, workflows, and specialized tools. Brockman puts the desired endpoint in stark terms:
"You want almost no interface, you want no product,"
In that framing, the best AI interface may be one that users barely notice. The user would not move through a growing stack of product surfaces. They would hand off work to a persistent, context-aware agent that can act on its own.
This is a major shift in how software is imagined. Traditional apps make their value visible through screens, controls, and feature sets. Brockman's version suggests that the most valuable AI system could be one that removes those surfaces rather than multiplying them.
Current tools still expose the gap
The source makes clear that OpenAI's current products do not yet match this vision. A tool like Codex is described as being far from an invisible interface. That matters because Codex shows how much structure and interaction may still be required when AI is used for demanding digital work.
The larger problem is reliability. The source states that AI models are still not reliable enough. As long as that remains true, users and businesses cannot simply expect an agent to understand everything, remember the right context, and act correctly without support.
That gap creates practical work around the model. The source points to heavy prompt work and custom integrations as part of what is still needed. In other words, the dream is a low-friction, almost invisible assistant, but the current reality often involves careful setup and specialized implementation.
This is why the interface question is also an infrastructure question. Making software disappear is not just a design decision. It depends on whether the model can handle tasks with enough consistency, whether the right context is available, and whether integrations are built to fit real workflows.
Why businesses still need hands-on AI integration
The source also notes that Anthropic, OpenAI, and Microsoft have all created separate companies focused on helping businesses integrate AI. These teams go on-site to support that work.
That detail underlines the distance between the vision and the present. If AI could already function as a seamless, context-aware layer across software, companies would need much less hands-on help. The need for on-site integration suggests that many organizations still require tailored work before AI systems fit into daily operations.
Several practical needs follow logically from the source's account:
- AI systems must connect to the software people already use.
- Models must become reliable enough for delegated digital tasks.
- Prompt work and custom integrations remain part of the current path.
- The interface can shrink only when the underlying system can carry more responsibility.
That does not make Brockman's vision irrelevant. It makes it a benchmark. The closer ChatGPT gets to acting as a persistent, context-aware agent, the less users may need to think about separate tools, menus, and workflows.
The future depends on trust, not just simplicity
The idea of an almost no interface future is appealing because software has become something many people must constantly learn. Brockman's argument is that AI should reduce that burden. Instead of asking users to adapt to software, the AI layer would adapt to the task.
But the source's own evidence shows why the transition is difficult. Plugins launched with a broad promise, including web search and third-party apps like Gmail, but Brockman now says the models were not ready. Current tools still require visible interaction, prompt work, and integration support.
So the key question is not whether ChatGPT can have fewer buttons. It is whether AI models can become reliable enough for people to hand off meaningful digital work. Until then, the no-interface future remains a direction of travel rather than a finished product.