Thomas is the boss. He’s worked his way up over the years and has been the decision-maker for six. For a major project, he needs information by tomorrow. He calls Susanne, someone he’s worked with closely for a while and can rely on to deliver accurate, correct and understandable information. As usual, Thomas has no time. He throws a few keywords at Susanne. But because Susanne has worked with him so well for so long, she knows exactly what he wants and how he needs it.
Susanne thinks it through, pulls together some information, but needs support from the specialist department. She calls Jörg. Jörg is the undisputed expert for what she needs. She walks into his office, starts explaining that Thomas needs something, but Jörg interrupts her quickly and says: “Yes, yes, I know exactly what he needs, I’ll get it ready for you and how would you like it, in what format?” A few hours later Susanne gets the information, prepares it a bit more and sends it back to Thomas.
Why Thomas can rely on Susanne
Thomas can rely on both Susanne and Jörg knowing exactly what they’re talking about. He only needs to glance at the data and information, skim through it, copy it into his PowerPoint presentation and go into his important meeting tomorrow. This is a completely normal process that happens everywhere, constantly.
What happens now when AI comes into play? Are all involved parties at risk of being unemployed soon? Of course, one could say that artificial agents and Large Language Models will now take over everything. But that’s exactly how I don’t want to tell this story.
Thomas, Susanne and Jörg worry about their future
For many who don’t really know their field or want to learn something new, the results of AI generation are pure magic. Even the first result from a ChatGPT input reads impressively. But how does this actually work? That’s why I started with the example of Thomas, Susanne and Jörg. They are nothing other than human agents and Large Language Models.
In our case, Thomas is the person who needs information to make decisions with far-reaching consequences. He’s responsible when the decision was good or not. In times of artificial intelligence, there will always be a Thomas. Because a machine will never be responsible for the decision and certainly won’t be able to be held liable for the far-reaching consequences. How could it be? I’ve never seen a machine behind bars in prison.
Currently Thomas relies on Susanne. And that’s a good thing. Because he trusts her. He knows she delivers. Susanne could be a project manager, middle manager, an assistant or whatever. Anyone who has such good employees like Susanne can rely on them. If Thomas only gave a few keywords to Kevin, who started as an intern two months ago, he would never get his information. Even though Kevin is super clever and was top of his class everywhere. Kevin has little experience and limited people skills, wouldn’t know at all what Thomas wanted from him. Besides, he would quite underestimate Jörg, because Kevin believes he already knows everything better. Only with age do you realise how much there is that you’ll never learn.
Is Susanne hiding an AI agent?
In our case, Susanne is a kind of AI agent. She looks at what Thomas needs, translates it for herself and then thinks about where she can get it and who can help her. She understands the context. And that’s the most important thing for artificial intelligence. Without context, none of this can work. But without context, it can’t work in a normal company either. So she knows where to get the information. She finds some information herself. If Susanne were working in a RAG (Retrieval Augmented Generation) system, she would be the “Retriever” and also the “Generator”.
Jörg in our example is a well-trained Large Language Model. Like ChatGPT, Claude, but also very specialised models that you can freely download and run on your own computer. Jörg would be more like a local model that nobody else has access to. He has fed his Large Language Model with knowledge and experience for 30 years. He knows his stuff perfectly. He’s not a good handyman, would probably have to call an electrician if he wanted to change a power socket. But in his field, he knows his stuff perfectly.
Is Jörg perhaps an LLM himself?
And not only that, Jörg doesn’t just know his field, but he’s also able to translate that for his business. Because the company needs Jörg’s expertise to create products or develop other things. As an expert, Jörg has seen and heard everything. He attends all conferences, reads all the specialist papers, he just knows his stuff. And actually, life in the company is great for him, because most requests are pretty trivial.
When Susanne comes into his office and asks the question, all the synapses in his head light up and he immediately knows what she needs. That’s his experience, that’s what he’s trained himself for, he stays on top of things. So he’s a very valuable Large Language Model for his company. He doesn’t even need ChatGPT for it, because although ChatGPT also has this expertise, he simply has far more knowledge than the generic ChatGPT, which has collected knowledge from all experts. Jörg knows exactly what’s relevant and what’s important. That’s his value and his value will always remain, at least while Thomas is his boss. If it’s ever Kevin, it’ll be difficult. Kevin probably prefers new technologies and trends.
What Jörg knows, Kevin will never know. He only has access. Is that enough?
Sometimes Jörg even looks at ChatGPT and finds new aspects that he can then think about further. So he continues training his model. When Susanne starts with her questions, it rattles in his head. And she doesn’t even need to finish the question. He’s already coming up with an answer. And that’s also the characteristic of Large Language Models. Because they know so much, it’s very easy for Jörg to know what probably comes next. Then he can gather his knowledge and give it to Susanne either in a big list or he creates a kind of condensed paper from it, which Thomas can also understand, with a summary and the most important data.
This used to take Jörg a lot of time, but now he can use artificial intelligence to make it faster and more accurate. He’s not necessarily a great writer, but artificial intelligence helps him with that. He simply uses another model for that. But he is the main model and he will remain so. Susanne now takes everything, checks it again, looks to see if Jörg, who despite his expertise is sometimes a bit sloppy, a sloppy genius indeed, has done everything correctly, whether the commas are in the right place and whether it’s consistent. Because that’s also her job as assistant and as AI assistant and gives it back to Thomas.
Thomas, Susanne, Jörg: With AI knowledge you’re even more valuable
Thomas is pleased, that was quite quick and it was exactly what he needs. I’m not talking about Thomas now only working with AI agents or with ChatGPT and so on. He would never have figured that out, because for example many of the data that Jörg presented aren’t even in ChatGPT, but in other databases or the intranet, which only Jörg has access to.
But from this example it becomes clear that in many places in the company, processes can become faster. In many places AI can turn text and data and ideas into knowledge and then prepare it so that the people, who will always remain the impulse-givers and decision-makers, stay at the wheel - but only if they understand exactly what AI can do and what Jörg should continue to do.
From reinergaertner.de, est. 1997. Translated with the help of an AI that speaks better English than I do. Which isn’t saying much, after 25 years of Denglish.