Recruitment and Onboarding of People vs Construction of Agents
In the world of Human Resources, one of the first responsibilities is recruitment.
When a need arises, the company opens a position and seeks someone who can perform that function.
Resumes are received, candidates are interviewed, and among the available options, the most suitable person is chosen.
This process is limited by supply.
The company does not create candidates, it only chooses from existing ones.
After hiring comes onboarding.
The new employee needs to understand the company, its processes, rules, and culture.
He receives materials, learns in practice, and, over time, improves his performance.
Now, let us look at the world of Artificial Resources.
Here, the logic is different.
There is no “recruitment” in the traditional sense.
You do not wait for resumes.
Instead, you build the agent you need.
The limitation ceases to be the supply of candidates and becomes the clarity with which you define the problem and the function of that agent.
For this reason, the construction of the agent is equivalent to both recruitment and onboarding.
You not only “hire” the agent, but create it with its role, objective, and initial context already defined.
To do this, you can use specific frameworks, such as CrewAI, which allow you to structure agents and even teams of agents.
There is a technical layer in this construction, but the central point is definition.
Before any implementation, you need to be clear about the need.
What problem needs to be solved?
Should this task be done by a human, by an agent, or by a hybrid team?
From this decision, construction begins.
And this construction follows some fundamental elements:
- Identity: You give a name and a position to the agent. This helps organize and guide its behavior within the system.
- Objective: What does the agent need to do? What result should it deliver?
- Background: It is the context that guides how the agent should think and behave. A generalist model can respond on many topics, but when you direct the role (like a legal analyst or a financial specialist), the focus changes.
- Input and Output: What does the agent receive? And what does it need to deliver? This definition avoids generic responses and makes the work more predictable.
- Tools and Memory: Tools allow the agent to interact with external systems. Memory, often structured as reference material, allows it to use company-specific information, such as documents, processes, and internal rules.
This is where the equivalent of onboarding comes in.
Just as a new employee receives materials and learns how the company works, the agent also needs this context to operate correctly.
The difference is that, for the agent, this access is immediate and continuous.
Let us see a practical example:
Imagine that you need an agent to analyze legal contracts.
You could build it like this:
- Name: ContractBot
- Position: Legal Analyst
- Objective: Analyze contracts, identify risks, and suggest improvements
- Background: Graduated in Law, with 10 years of experience in commercial contracts
- Input: Contracts in text format
- Output: Report with risk analysis and improvement suggestions
- Tools: Access to the company’s contract management system
- Memory: Knowledge base with contract templates, standard clauses, and relevant jurisprudence
With this construction, you not only “hire” a legal analyst, but you also prepare him to perform his function efficiently and aligned with the company’s context.
For this reason, building an agent is not just configuring a tool.
It is defining, with clarity, a function within the company.
And the clearer that definition is, the better the result will be.
→ Next: 3.2.2 Management of Human Teams vs Orchestration of Agents