Dietmar Fischer

People have preferences.

And someone always has a painfully specific way they want the spreadsheet named.

A serious AI assistant cannot treat every conversation like a first date.

It needs memory.

But memory is not just one big bucket labelled stuff the AI knows.

That would be too messy.

If you tell an AI, for this report, make it very short, that does not mean every future report should become a tiny corporate haiku.

If a client says once, we are not sure about TikTok, that is not the same as a permanent strategy saying we never use TikTok.

One is an event, the other is a stable rule.

Mix those up, and your AI starts making confident business decisions based on one sentence someone said while fighting hotel Wi-Fi.

That is why AI researchers and agent designers increasingly separate memory into types.

Today, we focus on four of them.

Working memory, episodic memory, semantic memory, and procedural memory.

Working memory is the AI's desk.

It holds the current task, the current instruction, the thing it is trying to do right now.

Episodic memory is the AI's diary.

It remembers what happened before, a meeting, a decision, a client objection, a campaign result.

Semantic memory is the AI's knowledge base.

It stores stable facts, company policies, brand rules, product information, audience definitions.

Procedural memory is the AI's playbook.