Key Takeaways
- Greg Brockman says OpenAI Plugins failed because AI models weren't capable enough in 2023
- His vision: ChatGPT becomes an invisible layer handling digital tasks with 'almost no interface'
- Current OpenAI products like Codex contradict this vision, and AI reliability gaps persist
Greg Brockman, OpenAI cofounder, has admitted that ChatGPT Plugins failed outright. 'That didn't work. It didn't work at all because the models weren't ready,' he said. The admission is notable given how confidently OpenAI marketed Plugins when they launched in March 2023 with partners like Expedia, Instacart, and Shopify.
Now Brockman is pitching a different direction: software that disappears. 'You want almost no interface, you want no product,' he said. The goal is a persistent, context-aware agent that acts on your behalf rather than an app stuffed with features.
Why did Plugins fail?
Plugins launched to add web search and third-party apps like Gmail to ChatGPT. The pitch was compelling: turn the chatbot into a universal interface for the internet. But the 2023-era GPT-4 couldn't reliably orchestrate multi-step tasks across external services. It hallucinated API calls, lost context, and frustrated users who expected the slick demos to translate into daily use.
OpenAI quietly deprecated Plugins by late 2024, folding some functionality into native features like code interpreter and browse. The company never issued a post-mortem, so Brockman's blunt admission is the closest we've gotten to an official explanation.
Worth remembering next time OpenAI hypes a new product: they shipped Plugins knowing the models weren't ready.
What does 'almost no interface' actually mean?
Brockman's vision is that ChatGPT becomes an invisible layer for handing off digital tasks. You express intent in natural language. The agent handles the rest: scheduling, research, email, code, purchases. No menus, no buttons, no learning curve.
This idea isn't new. UX designer Golden Krishna argued in his 2015 book 'The Best Interface is No Interface' that screens and apps are often poor solutions to problems better solved by ambient, automated systems. Brockman is essentially applying that philosophy to AI agents.
The implications for product teams are significant. If AI agents commoditize software interactions, the value shifts from UI/UX design to backend integrations and data access. Apps become APIs. Brands become skills an agent can invoke.
The gap between vision and reality
Here's the problem: OpenAI's own products contradict Brockman's vision. Codex, the company's coding agent, is light-years from invisible. It requires developers to structure prompts, manage context windows, and debug hallucinations. The same applies to ChatGPT's current interface, which keeps adding features like custom GPTs, memory toggles, and canvas modes.
AI models still aren't reliable enough for true autonomy. Closing that gap demands heavy prompt engineering and custom integrations, work that companies can't do themselves. That's why OpenAI, Anthropic, and Microsoft have all spun up separate consulting units, sending teams on-site to help enterprises integrate AI. If the models were ready, you wouldn't need a squad of prompt engineers babysitting them.
For teams building AI workflows today, tools like Zapier, Make, and n8n remain essential for bridging the gap between AI outputs and real business processes. These platforms handle the orchestration logic that AI models still fumble.
Disclosure
Some links in this post are affiliate links — Logicity earns a commission if you sign up, at no extra cost to you. We only link products we have used or actively recommend.
What this means for AI builders
If Brockman's vision materializes, the competitive moat shifts from user-facing design to agentic reliability and tool integration. Companies that control the data and APIs agents need will hold leverage. Those building pretty dashboards may find their interfaces bypassed entirely.
But the timeline matters. Plugins were premature by two years at minimum. A truly invisible interface requires models that can chain dozens of actions without human intervention, handle edge cases gracefully, and maintain state across long sessions. Current models fail routinely at these tasks.
The pragmatic takeaway: build for today's reality while watching for the shift. Design your systems to be agent-accessible. Expose clean APIs. Structure your data for machine consumption. But don't bet your product roadmap on invisible AI replacing your interface within 18 months.
Logicity's Take
Brockman's admission that Plugins failed because models weren't ready is more interesting than his vision pitch. It reveals OpenAI's willingness to ship half-baked products while marketing them as transformative. Product teams should calibrate their expectations accordingly. The 'no interface' future may arrive, but OpenAI's track record suggests their timeline estimates are optimistic by 2-3x. Build agent-compatible systems, but keep your UI team employed.
Another view into how major AI labs handle organizational tensions
Frequently asked questions
Frequently Asked Questions
Why did OpenAI Plugins fail?
According to Greg Brockman, the AI models in 2023 weren't capable enough to reliably orchestrate tasks across third-party services. Plugins launched with 70+ partners but couldn't deliver consistent results.
What does Greg Brockman mean by 'almost no interface'?
He envisions ChatGPT as an invisible layer that handles digital tasks through natural language. Users express intent; the AI agent executes across apps and services without requiring users to learn traditional software.
When will AI agents replace traditional software interfaces?
No clear timeline exists. Current AI models still require heavy prompt engineering and human oversight for multi-step tasks. OpenAI, Anthropic, and Microsoft all maintain consulting teams to help enterprises integrate AI, suggesting full autonomy is years away.
How should product teams prepare for an agent-first future?
Build agent-accessible systems with clean APIs and machine-readable data structures. But maintain your user interface, current models can't reliably replace human-facing products yet.
Need Help Implementing This?
Building AI agent integrations or planning for an interface-optional future? Reach out to the Logicity team for consulting on AI product strategy and technical implementation.
Source: The Decoder / Matthias Bastian
Manaal Khan
Tech & Innovation Writer
Produced with AI assistance and reviewed by the Logicity editorial team. Learn more in our Editorial Policy.
