Agentic AI: 다음 조 달러 규모 효율성 전략

LLM에서 AI 에이전트로

The performance of AI has changed how we perceive artificial systems’ ability to interact with people, in large part thanks to the almost human speech from LLMs (Large Language Models) like ChatGPT.

이미 이러한 수준의 역량을 갖춘 AI는 번역, 데이터 검색, 프로그래밍 등과 같은 작업에서 인간을 돕거나 심지어 대체할 수 있습니다. 그러나 AI 에이전트를 통해 새로운 수준의 AI 능력이 열리고 있습니다. AI 에이전트의 핵심 아이디어는 주어진 환경에서 독립적으로 작동할 수 있는 AI를 만드는 것입니다. 이를 통해 지속적인 확인이나 감독 없이 스스로 행동할 수 있게 됩니다.

이는 주로 인간이 만든 프롬프트에 반응하는 LLM이나 이미지 생성기와 같은 생성 AI와는 매우 다른 실용적 역할을 부여합니다.

이 맥락에서 “환경”은 실제 세계의 특정 상황—예를 들어 자율 주행 에이전트를 위한 도로 위의 자동차—또는 특정 소프트웨어 제품군이나 디지털 인터페이스와 같은 완전한 가상 “장소”를 의미할 수 있습니다.

이 단계가 AI가 “채팅”에서 “실행”으로 진화하는 순간입니다. 일반적인 LLM과 달리 AI 에이전트는 종종 더 제한적이고 좁은 범위를 가집니다. 이는 자율적인 결정을 내릴 때 더 효율적이고 신뢰할 수 있게 해 주며, 일반적인 AI는 예상치 못한 방향으로 흐르기 쉽습니다.

AI 에이전트는 수많은 프로세스의 효율성을 향상시키는 데 AI를 유용하게 만드는 다음 단계입니다.

AI 에이전트 설명

스크롤하려면 스와이프 →

AI 에이전트는 새로운 돌파구인가, 아니면 진화인가?

These features put AI agents one step above previous iterations of AI tools, like assistants and bots, thanks to more proactive abilities, autonomy, and the ability to handle complex, multi-step tasks.

Like a true intelligence, they can be self-refining: they learn from experience, adjust their behavior based on feedback, and continuously enhance their performance capabilities over time.

So while bots and AI assistants can fulfill parts of the tasks given to an AI agent, the autonomy, proactive approach, and high level of complexity set agents apart from previous levels of automation. This makes them much closer to an actual human worker, at least for the specific task they are trained for.

자율 AI 에이전트가 산업 전반에 걸쳐 확장되는 방법

Modern life is full of repetitive tasks that are slightly too complex for simple automation scripts but tedious for humans. This makes AI agents especially relevant for highly repetitive, judgment-based workflows—from walking a customer through a troubleshooting algorithm to driving trucks on a highway.

Unlike humans, such AI agents can work 24/7 and scale instantly without additional overhead.

There are many ways to classify the level we have reached in AI abilities. Overall, metrics tend to compare the ability of AI to the general human population. The newest AI agents are reaching the skills of 50-90% of the population in specific, narrow-domain tasks.

These “Level 2-3” AI agents are usually considered a mid-point in AI progress, and just the beginning for AGI (Artificial General Intelligence).

The emerging architecture is to build many specialized AI agents and let them each handle a specific task at a level comparable to a human worker. For more complex workflows, a series of AI agents will act successively, collaborating to perform the larger job by breaking it down into smaller sub-tasks.

Custom agents, developed internally with the help of AI platforms, are also likely to become more common as coding agents reduce the complexity of developing enterprise apps.

AI와 인간의 효율성을 동시에 높이기

Another advantage of AI agents over more general AI is efficiency. While they excel at one task, they are not burdened by extraneous capabilities.

For example, a dedicated AI agent might be able to drive a car like a human while having none of the other human capacities for reasoning. It might “understand” roads, but will not make for good conversation, know how to generate an image from a prompt, or handle large databases.

As such, the model requires less memory and computing power to work. In turn, this reduces the hardware requirements and the energy consumed to perform its task.

얼마나 많은 자율성?

The greater autonomy of AI agents is their greatest strength, but it can also be a barrier to adoption.

Even a highly competent and reliable AI agent might not be trusted entirely to make decisions that have a high impact on the real world. It is one thing to set an improved chatbot to deal with customer service; it is another to let it handle payroll for thousands of employees.

It is likely that the parallel growth in the quality of AI decisions and increasing familiarity with them will let authorities give more latitude to AI decision-making. This, however, opens interesting legal and ethical questions regarding the responsibility of AI actions.

A clear legal framework will need to be determined. For example, if a self-driving car crashes, is the provider of the AI agent responsible? As autonomy increases, decisions impact real people and become expensive liability questions.

This also covers the issue of misuse, such as identity theft or automated fraud. These are legislative questions, but technological progress often outpaces regulation.

AI 에이전트의 미래

AI 에이전트가 좁은 범용성을 가질 수 있을까?

As explained, early AI agents are narrow to be efficient and trustworthy. However, higher-level AI agents will require an understanding of context, memory of past decisions, and task continuity.

At first, this might be done with the help of a human, who becomes more of a “conductor” of the agentic AIs rather than doing the task themselves. Of course, removing human labor from the equation is the ultimate goal to improve efficiency.

For example, an AI performing a diagnosis in a hospital will need to analyze medical images, understand text or voice describing symptoms, integrate medical test results and patient history, and find relevant scientific literature—all at once. It must then combine this data intelligently.

재무 적용 사례

Some sectors are skeptical of removing humans from the decision loop, particularly manufacturing or healthcare where mistakes can be deadly. However, one sector is embracing AI agents enthusiastically: finance.

Most of the financial world already uses high levels of automation, from trading systems to fraud detection. Fintechs are even more open to agentic AI, as their existence depends on automating financial efficiency. Improved efficiency in an industry handling trillions can quickly turn into profitable margin expansion.

For example, an agent can target the time-consuming task of reconciliation (matching bank statements, spreadsheets, and ledgers). Mid-sized companies can spend over 300 hours a year just on bank reconciliation. While spreadsheets can automate parts of this, they are brittle. Agentic AI offers higher flexibility and reasoning capabilities to handle exceptions and unstructured data.

Agentic AI에 투자하기

ServiceNow

ServiceNow is a cloud computing platform founded in 2003, dedicated to the creation and management of automated business workflows. From an established base of business automation clients, the company has fully switched to agentic AI.

It allows companies to utilize its AI agents, as well as customize them or create new ones from scratch using low-code and “vibe coding” (letting an AI write the code following prompts by a human).

The key selling point of ServiceNow is that it is not “married” to any AI technology in particular and can integrate into the existing digital tools and workflows of companies. It also provides a reliable interface to centralize the management of a growing number of AI agents.

AI governance is redefined, with a central hub to manage, monitor, and optimize AI Agents—whether native or third-party. And unlike closed ecosystems, ServiceNow is LLM-agnostic and deeply integrated with NVIDIA, hyperscalers, and a thriving AI ecosystem—giving businesses full control to future-proof their AI strategy.

The focus of these agents is to improve companies’ margins by making them more efficient—automating IT tasks, simplifying HR, handling routine customer requests, and speeding up app development.

The 20+ year-old company is still growing quickly, with more than 20% year-over-year revenue growth at the end of 2025. Remarkably, existing client cohorts are consistently growing their usage, leading to expanding ACV (Annual Contract Value). Renewal rates consistently stay in the 95%-97% range, making revenues highly predictable.

The company has managed to create solid operating margins and free cash flow, reflecting its relatively low cost base compared to its recurring revenues.

최신 ServiceNow (NOW) 주식 뉴스 및 개발