AI Everywhere, But Memory Is the Real Frontier

As Microsoft adds Copilot Chat to all the Microsoft 365 apps for even free tier users, there is such ubiquity around AI access seemingly everywhere that the barriers to entry are lower than ever. However, as most of the focus rests on what these tools can do, I believe the far more important frontline in the AI space is what they can remember.

Claude is the latest of the major AI chatbots to get memory, and, as with most things, Anthropic has approached this a bit differently (see blog post linked in the comments). In fact, there are now very real and clear differences between what you see of memory and how it impacts your use depending on whether you choose Claude, Copilot or ChatGPT.

The approach Anthropic has taken makes a lot of sense. Memory is shared, and within teams can be collaborative, but with silos created in logical places like projects to stop one memory context spilling into another. This is distinct from the ChatGPT conversational search approach with some isolation for projects, and so too from the Copilot purposeful building of user-linked memories in chat.

When you think about it, memory and associated contextual awareness is one of the most foundational capabilities that make human interactions with other humans work. A team of PhDs with no memory of the work you did last week or where it fits into broader goals would struggle to reach their potential; and this is why focusing so much on how smart AI models are somewhat misses the point.

It's great that AI tools are becoming more capable and are available in more places, but in order to deliver true agentic colleagues in the sense that most of us would understand that term, the place we need to look out for advances is this issue of memory.

How do AI memory capabilities enhance your work? Of the different approaches you've used, which one has worked the best for you?

First posted on Linkedin on 09/16/2025 -> View Linkedin post here