On 6 August 2026, something rare happened in the AI industry. OpenAI, Microsoft, Amazon, Cursor and Vercel, companies that spend most of their time competing with each other, published a shared standard called Agent Plugins 1.0.0. GitHub was in the room too. If you do not write software, that headline probably sounds like it belongs to someone else. It does not. This is the moment the AI tools you already use started agreeing on how to share skills.
The problem it solves, in plain words
Think about what you have already taught your AI. Maybe you have written out how your firm formats a management summary. Maybe you have a carefully worded instruction that turns a messy trial balance into a clean schedule. Maybe you have a prompt that drafts client emails in your voice.
Now think about where all of that lives. It lives inside one tool. If your office moves from one AI assistant to another, or if the good model for your work changes next quarter, you rebuild everything from scratch. That is not a small annoyance. It is the single biggest reason people stop investing effort in AI: the fear that the effort evaporates the moment the tool changes.
Agent Plugins is a packaging format. You build a skill once, and it installs into any AI client that supports the standard, without rewriting it for each one.
What was actually announced
Vercel proposed the format, and representatives from Amazon Web Services, Anysphere (the company behind Cursor), GitHub, Microsoft, OpenAI and Vercel refined it together into version 1.0.0. At launch it is supported by ChatGPT, Codex, Cursor, GitHub Copilot, Kiro and VS Code.
Technically, a plugin is just a folder. It holds a small manifest file that names the plugin, an optional folder of Skills, which are reusable instructions and resources, and an optional configuration for MCP servers, which are the connectors that let an AI reach your tools and data. You do not need to remember any of that. The part worth remembering is that instructions and connections now travel together, in one portable bundle.
The governance detail matters more than it sounds. The project is openly licensed, its maintainers and decisions are public, and the steering committee has people from AWS, Cursor, Microsoft, OpenAI and Vercel. No single company's roadmap steers the format. That is what makes a standard a standard rather than a marketing announcement.
What changes for you, realistically
Nothing changes tomorrow. You will not see a new button in ChatGPT this week. What changes is the direction of travel, and there are three consequences worth planning around.
The effort you put into AI becomes portable. The prompts, formats and checklists you have refined stop being tied to one subscription. Over a year, that is the difference between building an asset and renting one.
Good skills will become shareable. The moment a format is common, people start distributing skills the way they distribute templates today. Expect a CA to publish a plugin that knows a particular disclosure format, and expect a firm to hand new joiners a folder of skills on day one instead of a training deck.
Switching tools gets cheaper. Lock-in was the quiet reason many organisations delayed choosing an AI tool at all. When moving costs less, choosing gets easier, and that is good for everyone who has been stuck waiting for the dust to settle.
What to do this month
You do not need to touch the standard to benefit from the shift. You need to have something worth packaging when the packaging arrives. Start writing your instructions down properly, outside any single chat window.
Keep the result in a plain document, not just in a chat. Note which tool you tested it in and what good output looks like. That document is the thing that will slot into a plugin later, and it is useful today regardless of what any standards committee does.
The honest caveat
Standards are announced far more often than they are adopted. Agent Plugins has unusually strong backing and real clients supporting it on day one, which is more than most attempts manage. It is still version 1.0.0, and support in the AI apps that ordinary professionals use every day, rather than the developer tools on the launch list, is the thing to watch over the next few months.
So treat this as a signal, not an instruction. The signal is that the industry has decided your accumulated AI work should be movable. Start accumulating it deliberately, and write it down somewhere it can travel.