Hermes leads the popularization of AI agents

AI agent technology, which was previously perceived as a developer's domain, has reached a stage where it can be easily used by the general public. This is because a desktop version has emerged that can be used immediately by simply downloading an installation file, eliminating complex initial setup processes.

Vice President Kim Min-jung mentioned, “We are actually already using those agents,” explaining that agent technology, which is already integrated into daily life, will be more widely utilized as its accessibility improves.

While chatbots focus on answering questions, AI agents perform actual actions such as accessing local files, writing code, and generating documents. The core of an agent is its ability to explore files and perform tasks without direct user manipulation.

Vice President Kim Min-jung emphasized the expanded capabilities of agents, explaining, “It’s not just simply giving answers; it accesses our laptop, reads and summarizes files, and based on that, it can write code and create documents. These are also agents.”

Hermes's differentiator lies in its 'self-improvement' function. While existing AI required manual modification of skill files each time, Hermes learns through user conversations and upgrades its operational manuals autonomously.

Vice President Kim Min-jung stated, “Hermes makes me think of self-improvement. If you ask it to save the changes to the skill file every time you make a correction, Claude Code updates the skill file and allows you to download it again. But Hermes, even without me specifically asking, through conversations with me... upgrades its own operational manual.” She emphasized the autonomy of Hermes in understanding requirements and improving outputs without explicit user requests.

While traditional chatbots performed only single tasks as instructed by the user, agents perform sequential tasks, such as accessing a designated folder, opening and exploring files, organizing content, and then sending emails.

Vice President Kim Min-jung explained, “I describe it as a difference in action. Without me having to move my hands, the AI directly accesses my folder, opens files, explores content, organizes it, checks my email, and sends it to people on that email list.” She emphasized that AI performing physical actions is the decisive difference between agents and chatbots.

AI agents are created in a 'bot' format and assigned specific tasks. To aid intuitive understanding, appropriate names are given, and guidelines (Prompt) and skills (System Prompt) define the identity of each bot.

Vice President Kim Min-jung explained, “You can think of a bot as having created each individual agent. Here, you can create bots,” adding that agents perform individual roles through the bot creation process.

Creating too many bots from the outset without sufficient work processes can increase management points and be inefficient.

Vice President Kim Min-jung shared her experience, saying, “I asked if I should create a separate bot for AX, and it said, 'Not yet.'“ She advised that indiscriminate bot creation could lead to management difficulties.

Typical AI tools can create confusion by making a new folder for each task and not sharing memories. However, Hermes ensures that memories of different tasks do not get mixed even when all work is performed by the main bot.

Vice President Kim Min-jung explained, “Those who used Claude Code or Codex probably created a new folder, a new space, for each task. But Hermes, even if you do all tasks with the main bot, has a very good memory, so things don't get tangled.” She mentioned Hermes's strengths in independently recognizing each task and automatically saving it locally.

Hermes recently introduced a bot-to-bot conversation feature, allowing for mutual calls between the main bot and sub-agents. This is a significant advancement that overcomes the limitations of existing AI, which could not move between task folders.

Vice President Kim Min-jung explained, “Hermes recently enabled bot-to-bot communication, allowing calls from the main bot. While the instruction was given to the chief secretary, it says 'Message from Hermes',” highlighting a structure that clearly identifies message sources, supporting efficient collaboration. This feature allows agents to maintain their respective roles while influencing each other's work.

Finally, it provides content in an HTML file format that can be immediately uploaded.

Vice President Kim Min-jung explained, “I specifically asked for the importance of this article from an AI and AX perspective. So it summarizes why it's important. This is the key summary, and if I ask it to include important FAQs from this article, it provides FAQs, additional links, and tags every day.” She presented specific examples of results generated by the agent.

Using tools like Buffer for approval before publishing is recommended.

Vice President Kim Min-jung said, “If you completely trust it, you can link it to blog automation. But for now, I think it's better to check it yourself. If you just want simple card news automation, you can do that.” She emphasized that simple content can be automated, but important content requires human verification.

Agents for various roles such as news research, insight extraction, and writing operate, and their conversations and work progress can be checked through group chat rooms.

Vice President Kim Min-jung explained, “You can think of the agent named 'Chief' as managing five agents currently operating in the production room. This is the task that Hermes handed over to Cron earlier.” She added that once agent roles stabilize, individual agents can perform tasks independently, maximizing work efficiency.

The core of agent design is to break down job functions into detailed tasks and clearly define segments requiring human intervention. It is most efficient for users to design agents themselves rather than purchasing ready-made ones externally.

Vice President Kim Min-jung emphasized, “Ultimately, the best person to design an agent is oneself, and a suitable agent cannot be bought from outside. In other words, I have no choice but to design and create what I need.” As design experience accumulates, the completeness of task automation increases, making customized agents more sophisticated.

This allows for easier bot creation, bypassing the conceptualization stage.

Vice President Kim Min-jung explained, “With older agents, I had to thoroughly think about how to create an agent for what I needed, conceptualize it, try it, and it might not work. But with Hermes, I can just have a conversation, ask what parts of my work can be automated, and it recommends. So I can receive recommendations and create bots like that to automate.” She emphasized that the barrier to entry for design has been lowered.

In the past, installing Open Clou incurred costs, but Hermes can be easily started by simply installing the desktop version.

Vice President Kim Min-jung explained, “You can use Hermes's free models. I'm using the free version now. It's all open source, so it's free. That's why many people use it. In the past, people charged 150,000 to 200,000 won just to install Open Clou for us, but there's no need for that; you just install the desktop version.” She highlighted Hermes's strengths in terms of accessibility and cost-effectiveness.

Changes are reflected immediately through conversation without skill resets, and the history of execution is recorded, providing live feedback to enhance the completeness of the final output.

Vice President Kim Min-jung explained, “With Claude Code, you had to fix it and then update the skill. But here, if you just keep conversing, it updates itself continuously, leading to gradual development and a final decision. And you don't have to go back to the profile to update it.” This method eliminates the need for users to constantly update profiles and allows for transparent tracking of the agent's development process.

While agents can most easily approach repetitive automation, they can be expected to play a wider variety of roles beyond that.

Vice President Kim Min-jung advised, “At the corporate level, you can't just create toy agents. You should test what you can do in your work. While agents can most easily approach repetitive automation, I don't think that's all there is to agents.” She emphasized a strategic approach to prepare for the era of one agent per person.