- OpenAI co-founder Greg Brockman said the company’s 2023 Plugins effort failed because the models “weren’t ready.”
- He argued people should not have to learn software, and ChatGPT should become an invisible layer for handing off digital tasks.
- His stated goal is a persistent, context-aware agent that acts on its own — “you want almost no interface, you want no product.”
- OpenAI’s current products, including Codex, still contradict that vision, and reliability gaps require heavy prompt work and custom integrations.
What Happened
OpenAI co-founder Greg Brockman said he wants ChatGPT to become an “invisible layer” for handing off digital tasks rather than an app packed with ever more features, in comments reported by The Decoder on July 4, 2026, via journalist Alex Kantrowitz. He also conceded that OpenAI’s 2023 Plugins launch “didn’t work at all because the models weren’t ready.”
Why It Matters
The admission is notable because OpenAI marketed Plugins confidently at launch — a reminder to weigh the company’s product hype against results. It also reframes the roadmap away from feature-rich apps and toward autonomous agents that act on a user’s behalf.
Plugins, introduced in 2023, added web search and third-party apps such as Gmail to ChatGPT. Brockman’s account positions that effort as ahead of the models’ actual capability, and his “no interface” framing is a bet that improved models will make dedicated apps and learned workflows unnecessary.
Technical Details
Brockman’s stated goal is a persistent, context-aware agent that acts on its own, not an app with a growing feature set. “You want almost no interface, you want no product,” he said. The vision runs against OpenAI’s own current products: a tool like Codex is, in the report’s framing, far from an invisible interface. AI models also remain unreliable enough that closing the gap requires heavy prompt work and custom integrations, which is the opposite of a hands-off layer. That gap between ambition and current capability is the core tension in the remarks.
Who’s Affected
The direction affects ChatGPT’s users, who Brockman suggests should not have to learn software, and the developers and enterprises building on OpenAI that would need to adapt to an agent-first rather than app-first model. It also aligns OpenAI with rivals: Anthropic, OpenAI, and Microsoft have all spun up separate companies that send teams on-site to help businesses integrate AI — an acknowledgment that the autonomous-agent vision still depends on substantial human setup today.
What’s Next
The near-term test is whether OpenAI narrows the distance between the “no interface” goal and shipping products like Codex that still require significant configuration. The on-site integration companies suggest the invisible-agent future remains a direction rather than a released capability. Brockman described where he wants to take the product, not a launched feature, so the claim will be measured against what OpenAI actually ships next.
