Your AI's Memory Should Be Yours to Keep.

I love AI personal assistants. I wrote about them here a couple weeks ago.

Whether you use Meta’s Muse, OpenAI’s Dots, SpaceXSI’s Grok Bot, or any of the seemingly endless number of others to choose from, these tools can be incredibly useful.

But there is one thing about this wave of AI assistants that I decidedly don’t love.

When you use Muse/Dots/whatever, it learns about you and builds memory and context, thereby better meeting your needs over time. That's great.

But what's not great is that all of that context and learning is locked up in a black box. You can't easily inspect it or browse it, and most importantly, you can't take it with you to other apps. You're not building an asset you control that gives you compounding leverage over time.

I think you need to own your own second brain (not have it buried in the bowels of a mega-corp).

So today, I want to break down:

—@dharmesh

The first thing to understand about this new wave of AI assistants is that there’s an ongoing relationship with them that persists in the background even after you close the chat window.

With a classic chatbot, you open a window, ask a question, get an answer and close the tab. But the new personal assistants are built to stick around.

They check in with you and help you plan your day in a proactive manner, popping up to ask you questions so they can prepare for upcoming actions or events.

What makes that work is connectors and memory.

In most cases, one of the first things you’ll do with these personal agents is connecting them to your important apps (like Gmail, Gcal, Granola, etc.) This gives the assistant access to important systems so it can do more things for you.

It also learns about you and builds up memory and context about your people, your projects, your habits and what you care about right now. And because it keeps learning, it gets better at meeting your needs over time.

At first, one of these assistants feels like any other AI app.

A few weeks later, it starts to feel a little magical. It's no longer meeting you for the first time every time you open it. It already knows your routine, what you’re working on, and the default actions you tend to take when dealing with your email inbox, for example.

In my experience, each tool really does get better the more it learns about you. The problem is that those learnings are locked inside that specific tool.

Imagine you’ve spent a month or two getting comfortable using a certain AI assistant. It doesn’t matter which one. Then a competing product releases a big update, as is typical in this day and age.

If you’re anything like me, you’d want to try this new release.

But you’re faced with a hefty switching cost: if you can’t export your context outside of your existing tool, the new tool must get to know you from scratch.

It could take weeks (or even months) for this new tool to build up the level of context you had built with the previous tool.

And by then, a third competitor has released an even stronger product -- are you going to start over from scratch again?

Do that a few times and you end up split across a handful of apps that don't talk to each other. Each one holds its own piece of your context.

The way I think about it, you're putting real work into teaching these assistants who you are. But you're not building an asset you control that gives you compounding leverage over time.

Satya Nadella, the Chairman and CEO of Microsoft, has talked recently about how important it is for companies to own their enterprise context, instead of handing it over to the frontier model companies. I think he's right, and I'd go a step further.

This doesn't just apply to big enterprises. It also applies to regular people like you and me.

Imagine having a different personal assistant for each part of your life: one for work, one for your health, one for the family calendar. Each one keeps private notes on you that you're never allowed to read. And none of your assistants are willing to work together to share notes.

Nobody would sign up for that arrangement with human assistants. But with AI, it's pretty much the default setup right now.

I think you need to own your own second brain (and not have it buried in the bowels of some mega-corp).

Are you familiar with the concept of a second brain?

In short, the idea is that your second brain is a digital collection of your notes, ideas, and context. It’s where you offload all your short-term mental memories for long-term preservation, so you don’t have to rely on remembering them yourself later.

My conviction on the importance of owning your own second brain is one of the main reasons I'm working with the HubSpot Next team on YouSpot (the AI-native personal CRM). At the heart of YouSpot is a second brain that you can hydrate with information about you, your company and your customers. Bring in information from Gmail/Outlook, your calendar, LinkedIn, ex-twitter, Granola, Obsidian -- and anything that supports MCP.

Once you have your second brain built, you get a daily briefing email (that I think you will LOVE), you can chat with your brain using one of the frontier AI models (like GPT 6 Astra or Opus 5.5) and with MCP support, you can connect your YouSpot brain to any AI app/harness you want (like Claude or ChatGPT).

If the daily email YouSpot sends you surfaces one thing a month (which for me it does on a regular basis), it's well worth the $10. But if you use this link, you’ll get 20% off.

YouSpot isn’t for everybody (it’s primarily built for non-technical solopreneurs). If that doesn’t describe you, don’t worry. Let’s get into the options available to you.

Owning your own second brain doesn't require a new app or any technical skills. It comes down to three steps: Collect, Keep, Carry.

1. Collect. Start by pulling what each assistant knows about you into one place. Run this prompt in every assistant you use, including ChatGPT and Claude:

2. Keep. Paste the results into one plain document that you control. At one time, I may have recommended a Google Doc for this. But these days I like to work with Markdown (.md) files. Markdown files are the universal standard for storing context. But they’re very simple. Inside you’ll find text with a little formatting.

You can ask AI to create one for you in simple language with this prompt:

The challenge with a document like this is that it’s frozen knowledge: it’s a snapshot of something that was true on the day it was created. But what if something changes?

We can mitigate some of this risk by creating a scheduled task. Use this prompt to have your AI walk you through the process:

3. Carry. When you try your next AI assistant, bring your document along to skip the long process of getting to know each other.

You can drag and drop this document into any chat window. Better yet, you can ask AI to update your custom instructions to include the file path of your About Me document.

These assistants are going to keep getting better, and every few months another one will launch that’s smarter or just a better fit for how you work. That’s a good thing.

But switching to a new assistant shouldn’t mean starting over from scratch every single time. That’s where your personally maintained second brain comes in.

Some folks may find the process of keeping a second brain updated to be too much work. If that’s you, I completely understand. Just keep the tradeoffs in mind -- consider committing to one harness (like ChatGPT or Claude) so that all your context is kept by one company, instead of spread across several. You won’t get the benefits of ownership, but if you don’t plan on using tools from other companies, you might not care.

But if you like trying new tools as much as I do, getting into the habit of keeping your context portable is worth it. Think of portable context as an onboarding guide for AI agents you can re-use every time you work with AI.

When done right and properly maintained, this means less repeating yourself, and more magical moments where the AI seems to “just work” out of the box, with minimal additional guidance.

It’s a fun time to be experimenting with AI 🙂

—Dharmesh (@dharmesh)