Original video in Chinese.
Key Takeaway
- Agent Designer (intelligent agent designer) is the most valuable job over the next three years, because realizing the value of AI technology depends on Agents.
- An Agent can be compared to a Gundam’s body, the large language model is the power source, tools are weapons, and the user is the pilot.
- An Agent Designer needs to understand both AI and business, be able to “translate” human intentions for AI, and design the Agent’s workflow.
- An Agent Designer needs to connect software and databases, turn them into the Agent’s tools, and design Agent collaboration from a systems perspective.
- Agents and RAG will become standard features of AI-native applications.
Over the next three years, the most valuable job is: Agent Designer, intelligent agent designer.
As the name suggests, it is someone who designs Agents.
Why is it valuable?
Because for AI technology to create value, and for large language model vendors to recoup their huge development costs, they must rely on Agents.
So, what is an Agent?
I’ve found that many people mix this concept up with the large language model concept and can’t keep it straight. Let me use Gundam as an analogy, and you’ll definitely understand.
An Agent is like a Gundam’s body, the mecha we see.
For this mecha to move, it needs power, right? The large language model is the Gundam’s most powerful power source — the Minovsky Drive.
Anyone who has watched Gundam knows that having the Minovsky Drive and not having it are worlds apart. It’s the same for Agents.
Long before large language model technology came out, researchers in Computer Science and AI had already been studying Agents for many, many years. It wasn’t until large language models became popular that this mecha finally got its Minovsky Drive installed.
But for Gundam, power alone still isn’t enough — what are you going to fight with?
So the various weapons in Gundam’s hands are equivalent to an Agent’s tools:
Depending on the weapons it uses, a Gundam may be a melee type, a sniper type, or a heavy armor defense type.
Agents are the same. Some Agents are equipped with search tools and are responsible for looking up information online; some Agents are equipped with scraping tools and can rip all the content off a webpage, and so on.
For a Gundam to act, it needs the pilot’s control. Agents are the same and need user instructions. Of course, both Gundam and Agents can to some extent run autonomously.
Since the pilot is human, there must be differences in skill level. In Gundam’s setting, the most powerful pilots are Newtypes. They have stronger analytical abilities, and so on.
Likewise, on the Agent side, users also differ clearly in ability. I mentioned in my previous video that only a small minority of less than 5% can use AI well at this early stage. And I named my Knowledge Planet Newtype, taken from Gundam, with the goal of bringing together the super individuals of the AI era.
So if you look at the concept of Agents again through my theory, it becomes much clearer. And I am very sure that my understanding is correct, because I came to this from the code.
Take the code of CrewAI as an example. When writing each Agent, in addition to describing its responsibilities and background, you also set what large language model it is equipped with and which tools it can call — my first reaction when I saw this at the time was Gundam.
That’s why in my previous video I recommended that everyone must learn programming. If you can read code, your understanding of AI will definitely be deeper than other people’s, and you won’t be fooled by those half-informed media figures and business tycoons.
OK, now everyone understands the definition of an Agent. So why do I say Agent Designer is the key to AI technology creating value?
Last month in the Knowledge Planet Newtype, I mentioned:
“As for how to build a Multi-Agent System. Technically, it’s not difficult at all. What’s difficult is that you have to think clearly about how you want the Agent to [do things].
The value of an Agent lies in the Workflow. And designing that Workflow requires you to understand both AI and business.“
An Agent Designer is the person who understands both AI and business.
He has to know how to ‘communicate’ with AI and ‘translate’ human intent for AI to understand; he also has to know how to communicate with people in different business units, define requirements and problems; and he has to know how to communicate with the people who control the budget, such as the boss, and manage the other side’s expectations.
He needs to know how to connect various software and databases and turn those channels into tools to equip the Agent with.
He needs to know how to design workflows from a systems perspective, so that different Agents can divide up the work and cooperate, and each stage delivers something with a clear definition that can be evaluated and measured.
Such a person spans the human world and the AI world, connecting the two by designing and building an Agent system. Think about it: with that kind of ability and value, how much should he be worth?
If you’re interested in this kind of job or this kind of role, now is the best time to act. Everything is just getting started.
I’ll also keep sharing content related to Agents. Just as I said in the Knowledge Planet Newtype: Agents and RAG will definitely become standard features of AI-native applications. If you’re interested, remember to follow. What I share is very valuable.
OK, that’s it for this episode. See you next time!