AI Agents Becoming 'Hands and Feet'... Operating Systems Must Be Designed Before Organizational Adoption

While autonomous AI agents like Hermes or OpenClaw were previously difficult to install and set up, a recent environment has emerged where anyone can easily build multi-agents just by downloading an installation file, allowing anyone to make agents work 24 hours a day. Kim Min-jeong, Vice President of HRX Lab, appearing in a T Times TV video, compared agents to a person's 'hands and feet.' She explained that while previous AI tools played a role in making people comfortable by creating images or providing summaries, agents are becoming hands and feet, ultimately being able to do the work that humans do.

The Reason Why You Should Not Give Up After Trying Once

Vice President Kim pointed out that most people give up after trying an agent just once. In the beginning, the desired results do not come out at once, and no matter how good the skills provided are, they do not come out customized to the user. She explained that one must go through initial failures and continue to complete the agent through many conversations so that the agent understands the user well. The video description also emphasized that although many users are disappointed or give up in a single day, agents become smarter the more they are used, so one should not give up easily.

She advised expanding roles step-by-step. At first, assign tasks such as summarization, and then expand the role one by one, such as creating content to be posted on LinkedIn or Instagram based on those results. She explained that as roles expand from research, summarization, image production, to publishing inspection, accuracy increases further. She also mentioned that it is not a bad idea to first become familiar with personal interests like stocks or real estate.

If You Break Down Tasks, You Will See What to Assign

Vice President Kim emphasized that anyone working in a company must ultimately be able to design their own work. By setting large categories such as consulting advisory, education, or customer management and then breaking down the tasks within them, truly inconvenient and repetitive tasks can be discovered. Once identified, those tasks can be assigned to an agent. Examples provided included news briefing, and the task of creating meeting minutes every time, using the results to create a to-do list, and sending emails to persons in charge regarding action items. She explained that these are tasks frequently performed by junior employees.

She stated that for tasks like briefing, an agent can completely take over, leaving only the final inspection to the human, and added that if 80% of the work disappears from the person doing that job, they must think about what meaningful work they will fill that space with. Vice President Kim explained that the premise itself—that one must break down tasks well to assign them to AI—does not change just because Hermes has arrived.

Organizational Agents: Beyond Individual Productivity, Consider Corporate Value

When asked what is required for a personal agent to transition into a corporate agent, Vice President Kim distinguished between individual productivity, company productivity, and corporate value. She explained that it should not merely be good for the individual, but one must see whether it can increase productivity when it moves to the team and then to the organization; otherwise, companies will wonder why they should bother using it.

While seeing that agents are helpful to companies, she predicted that if adoption is pushed down from the top-down, there will be significant resistance at the employee level. Since an agent replaces the work one was doing—meaning a person is removed—members may have great concerns that their jobs might disappear. Vice President Kim warned that if agents are introduced without pre-designing governance, organizational culture, role distribution, and management/operation systems for agents, conflicts will inevitably arise. She cautiously suggested that since fast-moving companies have already begun the work of creating operating and management systems, individuals should continue to consider parts applicable to themselves, while companies should proceed with preparations from an organizational perspective.

She viewed that how a company uses agents will depend on its security policies. Some companies provide only "empty shell agents," which they call "can agents," to employees, allowing employees to assign roles within them to create their own agents. Other companies use automation screens where they click an agent on the left to move it to the right, input the model and system prompt, and then sequentially connect agents tasked with news clipping, research, and summarization. Vice President Kim explained that this process of specifying roles and designing exactly where inputs and outputs go is the first experience in designing an agent. Given security issues, she predicted that companies will create infrastructure and structures where employees can create agents at any time.

24-Hour Agents: Token Costs and Security Management are Key

Agents like Hermes continue to perform assigned tasks even while the user is away if a task is set. Vice President Kim introduced that she has assigned agents to monitor sites and detect what is trending on platforms like X. However, she pointed out that from a company's perspective, this must be considered as assigning work 24 hours a day results in a lot of wasted tokens. She explained that by distinguishing between tasks that truly need to run 24 hours a day and those that do not, and by setting the time well, token costs can be used efficiently, and it is necessary to clearly distinguish what tasks humans truly must do.

Regarding security, a question arose whether a main secretary like Hermes being hacked could reveal user information such as schedules. Vice President Kim replied that just as Telegram has individual IDs and tokens that must not be exposed, one simply needs to maintain user IDs and key values well, just as API key values are always maintained for security. She expressed the opinion that while information could be revealed if a laptop is stolen by someone, it likely would not be a major security issue otherwise.

Vice President Kim cited her own case, stating that the way of working has changed in the AX era. She takes on the role of inspecting, while Hermes acts as the main chief of staff, and sub-agents work underneath. She said that the change in the way of working moving into the AX era is the division into work done by humans, work done by humans and AI, work done by AI alone, and work done by agents and agents together. She added that within this, for tasks previously done only by humans, one must redesign work by breaking tasks down in detail and determining what the human will do. She explained that while consultants used to establish processes by working on projects for 1 to 3 months, now, an individual who knows the work well can accomplish it through conversation with the help of AI tools.