How to Future-Proof Your AI Workflows

By now, it probably feels like your AI knows you at least a little bit. Maybe even a lot bit.

It knows what you’re working on, what you’re interested in, plus all kinds of little quirks and patterns it’s noticed from the messages you’ve exchanged with it.

(Side note: Try this prompt to see what I mean -- “Based on everything we’ve ever talked about, what are 25 interesting things you know about me?”)

This context is crucial in returning helpful answers back to you. The more context you provide, the more accurately AI can deliver the result you’re looking for.

This is why it’s unfair to compare, say, a naked ChatGPT to the Claude account you’ve used for the past year and a half. The context gap is dramatically more impactful than any difference in ability between the premier models themselves.

But what happens when your go-to AI tool goes down? Or when you want to know whether a competing tool could be better than the one you use today?

The solution here is portable context, and it’s saved me plenty of times. Today, I want to break down:

• Why does portable context even matter?

• How to store your context in Markdown files

• The 80/20 compromise for beginners (much simpler)

—@dharmesh

Why Does Portable Context Matter?

When Anthropic’s Claude (in all its various forms) went down last week, it could have been enough to derail my entire day had I not been prepared.

Thankfully, because I keep my context stored in plain files that every AI tool can read, I was able to supply a model I’d never tried before with everything it needed to continue where Claude left off.

Right now, most of what your AI knows about you lives inside one company’s app (or what’s sometimes referred to as a “super app”). The preferences you’ve set, the backstory behind your projects, the way you’ve slowly trained it to talk to you: it’s all in there. And it’s only accessible to that one app.

Portable context flips that. By storing key information and personalization details in an external file outside the app itself, it becomes accessible to any AI model.

That one change unlocks four things:

Outage insurance. When your main AI goes down, you’re not stranded. You open another, connect your context, and you’re back to work in under a minute.

The right model for the job. Different models really are good at different things. Maybe one writes more like you and another is sharper at digging through a long document. With portable context, you can hand the exact same background to each and use whichever fits the task, without re-explaining yourself every time you switch.

No re-teaching tax. Think about how many times you’ve re-introduced yourself to a fresh chat: your role, your company, what you’re trying to get done. With a file you own, you mostly stop.

Ready for next month’s model. There’s always a newer, more capable model on the way. When the next one shows up, you don’t start from zero. You hand it your context the day it launches and see whether it beats what you’re already using.

What Portable Context Looks Like

Creating portable context starts with creating a Markdown file (.md). This file will steer how the AI behaves on every future answer.

Markdown files are the universal standard for storing context. This is good news because markdown is just text with a little formatting. Very simple.

This is what a markdown file supplying context looks like:

That’s a real, usable context file. The lines with # symbols are headers, and the lines with - are list items that provide context on the headers. This is how most context-bearing files are structured.

The most popular context files are CLAUDE.md and AGENTS.md files used by Claude Code and Codex, respectively. These specific files are loaded before the model even processes the message you send it to make sure it handles the request in accordance with your preferences.

A more practical-at-work context file could be marketing-june-2026.md, which contains all the key data I pulled in from HubSpot’s marketing team. Instead of copying and pasting all the numbers into every conversation, I could just reference the marketing-june-2026.md file to any AI model and it would immediately have the full picture.

Or say you want to maintain a slightly different writing voice across your social media channels to better cater to each algorithm. You could keep a writing-voices.md file with a couple examples of what “good” and “bad” writing looks like on each channel.

The bigger idea here is that portable context is anything you’d otherwise keep re-explaining (your preferences, your data, your writing voice), written down once in a plain file you can hand off to any AI.

Open Notepad (Windows) or TextEdit (Mac) to create your first markdown file using one of the templates above.

How you use these files will depend on how you interact with AI on a daily basis.

If you use Claude.ai or ChatGPT.com: Look in the sidebar for “Projects” and create a new one if you don’t already have a living one. Inside your project, you’ll see “Sources” (ChatGPT) or “Files” (Claude). Drag and drop your relevant markdown file(s) there. The same instructions apply to the official apps. Once completed, all conversations within that project will have access to your context files.

If you use Claude Code or Codex: Drag and drop your markdown file(s) along with this prompt: “I want to add the following context to this project”.

A Simpler Option

The race to build the smartest AI models isn’t slowing down. Establishing a portable context frees you to tinker with today’s best AI models without any lock-in. The moment a better model lands, you’re free and ready to switch.

If creating markdown files feels like more than you’re ready to take on right now, here’s a plan B for you: Custom Instructions. ChatGPT and Claude both have a spot in their settings where you can provide extra context.

Take the 80/20 approach: make sure both ChatGPT and Claude are loaded up with your own custom instructions. That way, you’d still be protected if one of the apps went down. If you’ve never used custom instructions before, here’s a short guide.

Remember the golden rule: context is queen. It takes a bit more manual work, but copying the custom instructions from your AI of choice over to a new app is not just a good idea, it’s essential if you want a fair shot at equal or better output.

I’m curious -- what AI apps are you using these days? Has your answer changed over the past 6 months? I use OpenAI’s Codex on a regular basis, but I like to dabble across a handful of different models every now and then.

Hit reply and let me know. Reading the responses to this newsletter is what keeps me writing it 🙂

—Dharmesh (@dharmesh)