Skip to main content
floppy disk
Written by Jason Montoya on . Posted in Organizing Systems.

AI Has a Memory Problem. Here's How to Work Around It.

While generative AI has many limits and issues, one of the big ones is a loss of memory. If you've had long threads with AI on the same topic, especially when you shift gears, you'll have experienced the frustration when AI forgets. Since part of the value of AI is the ability for it to process large volumes of content to detect patterns, this is quite a problem.

floppy disk

Life is Hard. Business is Challenging. The World is Uncertain.

Leaders, freelancers, and entrepreneurs: Get stories & systems, for navigating the challenges, in your inbox.

It's such a problem that there are now apps popping up to solve it. I stumbled upon a company called MemVid, which is an app that connects with your AI tool to retain that forgotten memory. They even posted a bounty of $800 for people to test it and bring it to its limits.

What's wild is how full circle this technology is operating. MemVid uses video files to store the melee and feed it to the AI. It's like a new form of digital data tapes! Here's how MemVid stores memory.

“Memvid is an experimental Python library that stores millions of text chunks inside an MP4 as QR-coded frames + a FAISS index, enabling fast, offline semantic retrieval without a database.”

This is such an unexpected way to store data, but I suspect there are some great technical reasons why it works so well.

With AI, we're entering a new world. It's one where we'll discover tools that we didn't even know existed that could transform our lives or work. It's also an opportunity to create those things AI can even do when we have no skill set to do so. Expect to see micro apps in the future, to solve small problems for individuals or small groups of people (where it was too costly before to develop with humans).

Regarding the memory problem, I've created a prompt for you to create memory blocks (if you're not yet ready to sign up for Memvid). I'm now creating these blocks on my long-term persistent threads on important topics (saving in Evernote). Structured data like this is also an important way to maximize the benefit of generative AI. This little prompt gives you four buckets (goals, decisions, context, next steps) for AI to populate in the memory block. Feel free to alter or add structure to this for your own memory blocks. Using Evernote, I'm creating a note for a thread and then adding each block into it as a backup I can give back to the AI or start a new thread should I need to remind it of anything when its memory fades.

Memory Block Prompt

We've covered a lot. Please provide a concise 'Project Memory Block' that summarizes:
  • Core Goals: What are we trying to achieve?
  • Key Decisions: What have we already finalized?
  • Current Context: Any specific names, variables, or constraints?
  • Next Steps: Where were we just about to go?
Please format this as a clean Markdown block so I can save it for later.

Wrapping Up

In many ways, leveraging AI is about using our imagination. It's about simply thinking of the thing you could do or could streamline. And it's discovering the AI tools that exist, that we had no clue existed. In addition to MemVid, I've also stumbled upon Replit and Google AI Studio, both of which let you build fully functional apps using AI. It's pretty wild.

How about you? What are some powerful AI tools you've stumbled upon that are not just cool entertainment but actually problem-solving resources you couldn't do without?


Related Content


Last Updated: April 02, 2026