Greg Brockman: OpenAI's 'almost no interface' vision

On July 4, 2026 OpenAI co-founder Greg Brockman said the company’s 2023 Plugins initiative failed because the models were not ready, and he laid out a different goal: make ChatGPT an invisible, persistent agent so people no longer need to learn software. Brockman said, "That didn't work. It didn't work at all because the models weren't ready," and argued, "You want almost no interface, you want no product."

What went wrong with Plugins?

Plugins launched in 2023 to add web search and third-party apps like Gmail to ChatGPT, but Brockman says the models could not support that approach. That is the bottom-line: the feature set existed, but model capability did not. He told users the failure was not a marketing problem but a capability gap, saying bluntly, "That didn't work. It didn't work at all because the models weren't ready." The piece also notes OpenAI continued to market the technology even as it struggled to deliver reliable results.

What does Brockman mean by an "almost no interface" future?

Brockman wants ChatGPT to act as a persistent, context-aware agent that executes digital tasks on its own rather than live inside ever-more complex apps. The idea is to remove the product layer: not another feature-packed app, but an invisible service that accepts tasks and completes them with low friction. The article contrasts that goal with existing tools, noting a tool like Codex "is light-years from an invisible interface" and that current models remain unreliable without heavy prompt engineering and custom integrations.

How are companies handling integration work now?

Brockman and the article point out that because models are not yet reliable, companies are doing hands-on integration work. Anthropic, OpenAI, and Microsoft have each spun up separate companies that send teams on-site to help businesses integrate AI, reflecting the practical engineering and operational lift still required to deploy these systems in enterprise settings. The article frames those efforts as evidence that the invisible-agent ideal has not been achieved and that integration demands remain high.

Why it matters

If Brockman’s vision wins, software could become less about learning interfaces and more about instructing persistent agents that manage context and state. That would shift investment from product feature design toward model reliability, continuous context handling, and operational integration. For businesses, the immediate implication is that reliable autonomous agents would reduce the need for custom app UIs, but only after models overcome the capability and reliability gaps Brockman highlighted.

What to watch

Watch improvements in model reliability and the emergence of practical, persistent agent prototypes. Also track whether companies that send teams on-site scale back that model as integrations become smoother; Anthropic, OpenAI, and Microsoft have already created such teams to help customers integrate AI.

The article anchors on two concrete points: Plugins launched in 2023 to extend ChatGPT with web search and third-party apps like Gmail, and Brockman’s comments on July 4, 2026 that those Plugins failed because models were not ready. He offered a clear alternative: build an invisible, context-aware agent rather than another app stuffed with features.