OpenAI Codex vibe coding took centre stage this week when Business Insider senior tech reporter Stephen Council used the tool to build a fully functional multi-slot clipboard manager for his MacBook, from a single 190-word prompt, in five minutes and 26 seconds.
Council had long wanted a clipboard tool that could store multiple strings of text simultaneously, letting him copy something new without losing what he had already saved. Off-the-shelf clipboard managers exist, but most either cost money or introduce extra clicks and pop-ups. His solution: describe the problem to Codex and let the agent write the code.
The result gave him nine independent copy-paste slots on his work MacBook. The standard Command-C and Command-V shortcuts remain slot one, unchanged. Each additional slot pairs a copy shortcut (cmd-c-2, cmd-c-3, and so on) with a corresponding paste shortcut, and a menu-bar icon lets him check what each slot is storing at a glance.
What OpenAI Codex Vibe Coding Actually Does
The exercise illustrates why vibe coding, writing software by describing intent in plain language rather than typing instructions line by line, has found a natural home in tasks that are too niche for commercial products and too small to justify hiring a developer.
Codex is now generally available, and OpenAI’s announcement confirms it ships with a Codex SDK that embeds the same agent into third-party workflows, tools, and apps. The SDK delivers performance on GPT-5-Codex without additional tuning, meaning developers can drop the agent into existing pipelines. OpenAI also confirmed that Codex can be tagged via @Codex in a Slack channel or thread, whereupon it automatically gathers context from the conversation, selects the appropriate environment, and replies with a link to the completed task in Codex cloud.
The practical implication is that the gap between having an idea and having a working piece of software has narrowed considerably for anyone willing to write a clear prompt.
The Broader Platform Shift Behind the Codex-ChatGPT Merger
Council’s clipboard experiment landed at a moment of considerable change inside OpenAI. On 16 May 2026, the company merged its consumer ChatGPT team and its agentic-coding Codex team into a single core product group, led by president and co-founder Greg Brockman, with the stated goal of building one unified agentic platform, according to Digital Applied.
That structural change fed directly into this week’s product releases. OpenAI merged Codex into the ChatGPT desktop app and simultaneously unveiled a new Work setting, which incorporates popular Codex capabilities including the ability to modify a computer’s files and to operate autonomously in a browser, Business Insider reported.
Alongside those changes, OpenAI revealed a new family of models, GPT-5.6. According to OpenAI’s API documentation, the family comprises three tiers: GPT-5.6 Sol for complex reasoning and coding tasks, GPT-5.6 Terra to balance intelligence and cost, and GPT-5.6 Luna for cost-sensitive, high-volume workloads. The tiering reflects a deliberate effort to let developers match model capability to task economics rather than defaulting to a single general-purpose option.
For Council, the platform mechanics were secondary to the outcome: a tool built precisely to his specification, requiring no coding knowledge and less time than it takes to make a cup of tea. OpenAI Codex vibe coding handled the environment setup, the keyboard-shortcut bindings, and the menu-bar interface without him writing a single line of code.
The episode also points to a shift in how reporters, lawyers, designers, and other knowledge workers think about software. When the cost of building a small bespoke utility drops to a paragraph of English text and a few minutes of waiting, the category of problems worth automating expands considerably.
Whether the clipboard app survives contact with a macOS update is the next question. Vibe-coded tools built on AI agents are functional but fragile, and maintaining them when the underlying operating system changes is a challenge no 190-word prompt currently solves.