Anthropic and OpenAI just made your AI know-how portable

Every few weeks someone in a lesson asks me a version of the same question: "What if I spend months setting up ChatGPT and my company moves to Claude?" Or Gemini. Or whatever wins next year.

It's a fair worry. Until now, the instructions, examples and templates you built for one AI tool mostly stayed in that tool. Yesterday, that got a lot less true.

What happened on 22 October?

The Agentic AI Foundation (AAIF) announced four new projects at its AGNTCon + MCPCon conference in San Jose. AAIF is the Linux Foundation's neutral home for the open standards that AI agents run on. It already looks after MCP, the protocol that lets AI connect to your apps, and AGENTS.md. It started last December with three projects; it now has ten.

The last two matter mostly to the people building AI. The first two matter to anyone using it at work.

Two of the fiercest rivals in AI just agreed that the plumbing should belong to nobody. That's good news for everyone who isn't a plumber.

Why should a non-developer care?

Because your real AI asset isn't the subscription. It's the know-how you've taught it: how you write proposals, how your weekly report is structured, which checks you run before an invoice goes out. That know-how takes hours to get right.

• Less lock-in. A skill you write today in Claude can be loaded by Codex, Gemini CLI, Cursor, GitHub Copilot and 30+ other tools that already read the format. Now that it sits with a neutral foundation, no single company can change the rules on you.
• Easier switching. Open Responses means the apps and automations your company builds can swap the model underneath with little rework. If a cheaper or better model appears, moving is a setting, not a rebuild.
• Shareable expertise. A skill is just a folder. You can hand it to a colleague, a new hire or a client, and their AI does the task your way.

What exactly is an Agent Skill?

A skill is a folder with one required file, SKILL.md. At the top of that file sit a name and a one-line description of when to use it. Below that come the steps, in plain language. Optional sub-folders hold examples, templates or small scripts.

The clever part is how it loads. At the start of a conversation the AI sees only the name and description of each skill, roughly 100 tokens each. When your request matches, it reads the full instructions. Only if it needs them does it open the files. You can install dozens of skills without filling up the AI's memory.

How to take advantage of this, starting this week

1. Pick three tasks you repeat every week. A weekly client update, a proposal, a meeting recap, social posts in your tone. Repetition is what makes a skill pay off.
2. Write down how you do them. The steps, what good looks like, two real examples, the mistakes to avoid. This is the hard, valuable part, and it's yours.
3. Turn each one into a SKILL.md with the prompt below. Save the folders somewhere you own, like a Google Drive or Dropbox folder, not only inside one AI app.
4. Install them where you work. In Claude you can upload a skill from Settings; Claude Code, Codex and Gemini CLI read them from a skills folder. Then test the same skill in a second tool and compare.
5. Share the winners with your team. A good skill is a process document that actually gets followed, because the AI follows it every time.

Copy this prompt: turn one of your processes into a skill

Paste this into Claude, ChatGPT or Gemini, then describe a task you do often.

You are helping me turn one of my recurring work tasks into an Agent Skill (the open SKILL.md format).

First, interview me. Ask me one question at a time, up to 8 questions, about:
- what the task is and when I do it
- the exact steps I follow
- what a great result looks like (ask me to paste 1-2 real examples)
- my tone, format and length preferences
- common mistakes to avoid
- any templates, checklists or files I use

Then write the complete SKILL.md file:
1. YAML frontmatter with:
   name: lowercase-with-hyphens (max 64 characters)
   description: one or two sentences saying what the skill does AND when to use it
2. A short "Goal" section
3. Numbered "Steps" in plain language
4. A "Quality check" list the AI must run before giving me the result
5. A "Mistakes to avoid" list
6. An "Examples" section (or tell me which examples to save in a references/ folder)

Keep the SKILL.md under 300 lines. Finally, tell me the folder name to use and which extra files to put next to it.

The task I want to turn into a skill:
[describe it here]

Also today in AI

OpenAI's Decisions API is now open to everyone

Also on 22 October, OpenAI opened its Decisions API to all customers. It's a "decision model": instead of writing text, it picks one answer from options you define (which team gets this ticket, is this lead hot or cold, approve or flag). OpenAI built it from its GPT-6 Luna model, and says responses dropped from 1–3 seconds to roughly 100–180 milliseconds. The language-learning app Speak cut the time between actions from 1.4 seconds to 0.13. It's OpenAI's answer to TypeSafe's Jev, which I explained in What is Jev?

How to use it: any step in your business that is really a multiple-choice question (tagging emails, routing enquiries, sorting invoices) no longer needs a slow, expensive chat model. If you use Make, n8n or Zapier, that step can now be near-instant and cheap. Test decision models on your own data before trusting any benchmark: they currently rank very differently depending on who measures.

DeepSeek raises about $15 billion

Chinese lab DeepSeek is raising around 100 billion yuan (about $15 billion), with Tencent and battery maker CATL among the biggest backers, according to Bloomberg. The target was raised twice along the way. The money goes to data centres and more open-source models, ahead of a possible IPO.

What it means for you: cheap, high-quality open models from China aren't going away, which keeps pressure on prices across the board. If you use them, check where your data is processed: running an open model through a European or US host is different from using a company's own app.

FAQ

Sources: Agentic AI Foundation announcement (22 Oct 2026); SiliconANGLE on the four projects; overview of the SKILL.md format and adopting tools; TechCrunch on the Decisions API (22 Oct 2026); Bloomberg via Yahoo Finance on DeepSeek's round. Recommendations from my own coaching work.