Memory
Learning and memory are concepts that we need to understand distinctly when it comes to AI, as there is a relevant difference in comparison with humans.
In humans, it is financially viable to perform learning through direct training (sending to a college, to an MBA, to a postgraduate program) that impacts a change in the brain.
In AI, these trainings are generally only done by the large companies from which we are “renting” the brain.
As managers, we will normally not perform direct training on an AI’s brain using advanced techniques.
However, we can and should use and abuse external memory and auxiliary materials.
This is one of the great advantages of AI Agents: they can access and use information at a speed far superior to humans.
Think of a human trying to read and absorb an entire encyclopedia in just a few minutes.
It is simply impossible.
Our brain was not made for that.
But for an AI Agent, this task is not only possible but can be done in seconds.
They can search through gigantic databases, find relevant information, and apply it almost instantly.
It is like having an assistant with a perfect memory and a supersonic reading speed.
This difference completely changes the way we deal with knowledge and learning when working with AI.
With humans, we invest heavily in training, courses, and education.
With AI, our focus should be on building and organizing a robust external memory.
Instead of trying to teach everything to the AI, we need to provide access to the right information, in the right format.
This external memory can include company documents, internal guides, knowledge bases, frequently asked questions, process descriptions, information about organizational culture, and customer service history.
Basically, it is everything that the AI can consult at the moment it is executing a task or answering a question.
Let us consider a practical example.
Imagine that you hire a generic AI Agent, with no specific knowledge about your company.
It knows nothing about your products, services, processes, or culture.
But then you provide this AI Agent with the company manual, the customer service guides, the internal documentation…
Suddenly, it starts responding like an experienced employee of your organization!
This is a crucial insight for any manager dealing with AI: AI Agents do not come ready for your specific business.
They come with generalist knowledge.
It is up to you to couple your company’s specific knowledge to them.
Without this coupling, you will get generic and not very useful responses.
With it, the responses will be specific, relevant, and valuable for your unique context.
However, it is important to understand the limits of this approach.
External memory does not alter the underlying AI model.
It does not create new knowledge in the AI’s “brain”.
It merely provides context at the moment of response.
Another common mistake is thinking that more information always means better results.
There is no point in simply overwhelming the AI with a ton of disorganized data.
The true value lies in selecting, structuring, and providing access to the right information, at the right time.
This requires an intentional effort of knowledge management, something I will address in detail in the next topic.
For now, the key point is to understand that the way we approach learning and knowledge in humans and AI is fundamentally different.
While humans need to internalize knowledge, AI Agents can simply access it on demand.