Build AI Workflows You Can Take Anywhere

Nicole Leffer had a workflow she couldn’t wait to show off — an AI agent living inside an Excel workbook, tracking her consulting hours, writing client recaps, running her dashboards. She opened the file, fired up Claude, and got nothing.

Claude was down.

For most marketers, that’s where the story ends. The demo dies, the work stops, and you wait for the status page to turn green again.

Nicole opened ChatGPT inside the same workbook instead. It read the instructions, read the history, and picked up exactly where Claude had left off. Same agent. Same memory. Different brain.

That’s the whole point of this guide. Here’s how to build workflows that don’t care which platform is up.

You picked a platform, learned its quirks, built your projects. That loyalty feels efficient — but everything you built lives somewhere you don’t control. Three things will eventually test that arrangement:

{{ r.title }} {{ r.icon }}

{{ r.body }}

There’s an upside too: the models leapfrog each other constantly. When your workflows travel, you go where the best tool is this week — instead of waiting for your platform to catch up.

A skill is surprisingly simple: a zip file containing a Markdown file — plain-text instructions for one specific task, plus any templates or assets it needs. Mike compared it to Neo downloading kung fu in The Matrix — your AI absorbs the capability instantly. Why skills beat projects for portability:

{{ s.tag }}

{{ s.title }}

{{ s.body }}

And you don’t hand-write these. Claude, ChatGPT, and Codex each include a skill-creator skill — paste in a recurring prompt, and it builds the skill for you. Then test, fix, repeat.

Keep a copy of your AI infrastructure somewhere the AI companies don’t control. Download your skills to Drive or Dropbox. Export your project and GPT instructions into folders you own. Those instructions are nearly interchangeable — paste your best GPT’s instructions into a Claude project and you’ll get close to the same output.

If your platform has an outage or a price hike, you recreate your entire setup elsewhere in five minutes, not five weeks. Thirty seconds per asset — and almost nobody does it. Do it this week.

The myth: AI does better when it sees everything — your whole Drive, every doc your company ever made. Nicole’s experience says the opposite. The AI does dramatically better when you hand it exactly the right document than when it burns effort — and your tokens — searching your archive. You already know what’s relevant. The portable version of context:

{{ c.text }}

Update the doc once and every platform is instantly current. Further along? MCP connectors apply the same idea to live data — one connector, wired into multiple platforms, so your data access travels too.

Nicole’s favorite system — an AI agent that lives inside an Excel workbook and runs identically on Claude, ChatGPT, or Copilot. Here’s how she builds one, in four moves. Tap each step:

The two moves that make it portable: the prompts live in tabs (instructions travel with the file, not the platform), and a history log tab gives the agent memory — so whichever AI opens the file next knows everything that’s happened. That’s why ChatGPT could pick up the demo the day Claude went dark.

The exact prompts from Nicole’s method — copy and adapt:

{{ pr.label }} {{ pr.btn }}

{{ pr.text }}