Hermes: How to introduce it to your organization?

Initially, it should begin with simple tasks such as summarization, creating simple card news, and SNS posting, gradually expanding its roles. Kim Min-jung, Deputy CEO of HRX Lab, explained that as roles such as research, publishing, and review are progressively increased, the agent's sophistication improves.

Failure to prepare this in advance will inevitably lead to conflicts within the organization. Deputy CEO Kim added that rapidly changing companies are already minimizing such issues by concurrently operating agent systems.

Individual tasks should be categorized into large categories such as consulting, education, and customer management, and then sub-tasks should be further broken down to identify inconvenient and repetitive tasks. Examples suitable for AI agent utilization include news clipping for junior staff, meeting minute organization, and sending action items. Deputy CEO Kim Min-jung explained that AI agents can go beyond merely assisting tools to become hands and feet, replacing a significant portion of work.

AI agents can act as the hands and feet of work, completely replacing human tasks. If an agent replaces 80% of tasks, one must consider what meaningful work will be done with the remaining 20%. Deputy CEO Kim Min-jung emphasized the importance of strategic preparation for such a task transition and setting human inspection points. It is necessary to clearly distinguish between AI delegation and human intervention points according to the nature of the work to define roles suitable for the new era.

When building an agent within an organization, users can leverage existing agent design experience to assign roles to a 'barebones agent'. This method involves directly designing input and output data flows through interfaces similar to Make or Mpa to automate tasks. This allows users to create an agent optimized for themselves.

Exposure of these key values can lead to serious problems. Deputy CEO Kim Min-jung mentioned that if individual laptop security is ensured, system security is not a major concern, emphasizing the importance of building infrastructure that allows employees to create agents themselves.

Linking via Codex instead of a terminal to check gateways is recommended as an efficient usage method. This enables flexible application of various functions based on the agent's learning capabilities.

AI agents have the ability to automatically install software and self-diagnose systems through terminal access. When an agent requests terminal access permission, the user approves it, completing the installation process. This inter-agent inspection mechanism is ushering in an era of autonomous system operation.

In the AX (AI Transformation) era, humans and AI will collaborate to perform tasks. Humans will take on planning and inspection roles, while AI will handle execution. It is efficient to move away from past consulting methods and instead have individuals create bots, design tasks through dialogue, and have main and sub-agents collaborate to process tasks.

Rather than entrusting all tasks to the agent 24/7, it is more cost-effective to set operating hours according to the nature of the task. Wisdom is needed to clearly distinguish between what humans should do and what agents should do.

For example, it's possible to apply this to actual work by asking Claude to check calendars, ascertain available times, or recommend nearby cafes. This suggests a need for a shift in perception to utilize existing technology as agents.

AI agent beginners should aim to overcome their fear and start with simple tasks rather than targeting perfect automation from the outset. It's crucial to begin with everyday requests like schedule checks or research and let the agent learn to understand the user through dialogue. Deputy CEO Kim Min-jung emphasized the importance of meticulously reviewing outputs and progressively expanding functionalities.

This method involves sequentially expanding roles, such as taking a news article, creating card news from it, and posting it on LinkedIn and Instagram. As roles like research, summarization, image creation, publishing, and review are expanded, the agent's precision increases. During this process, it is necessary to establish clear human intervention points and enhance the agent's understanding of tasks.

AI agents are very intelligent but do not initially understand the user's tasks. Therefore, in the early stages, it is important not to give up and to repeatedly engage in dialogue to train the agent to fully grasp the user's work style and intentions. It is crucial to give detailed instructions step-by-step and maintain consistent communication so that the user and the agent achieve the same level of task understanding.

What a person would take a month to do, an AI agent can sufficiently grasp in about a week of daily work with it. Compared to a human new employee's OJT (On-the-Job Training) which usually takes about three weeks, an AI agent is overwhelmingly more efficient. After that, it will act as a perfect assistant who immediately understands the user's intentions and, unlike humans, silently accepts revision requests without emotional responses.

Deputy CEO Kim Min-jung pointed out that going beyond five tasks can lead to overload and suggested operating separate agents for about five major tasks to increase efficiency. This is a way to effectively assist with tasks by acting as the user's hands and feet.

One must be wary of the constant work environment created by AI agent integration, which can feel like working 24 hours a day. Deputy CEO Kim Min-jung stated that she does not recommend this approach. Instead, she emphasized that the free time gained through agent adoption should be invested in more meaningful activities or spent with family to secure work-life balance.

Kim Min-jung, Deputy CEO of HRX Lab, emphasized the convenience she felt using AI agents herself, despite not being a developer. She mentioned that general staff, team leaders, and executives who are not involved in development often show interest in AI agents but are hesitant to approach them, explaining that technological advancements now make it easy for anyone to use these tools.