이미지: Microsoft AI
Summary
- On August 10, 2026, Microsoft posted an article on the Microsoft Source blog explaining how to build AI agents without writing code
- The guide lays out a four-step process using Microsoft 365 Copilot's 'Agents' panel: defining the problem, checking for prebuilt agents, and creating an agent through a natural-language description
- The guide offers concrete use cases such as automatically blocking compromised accounts detected in the early morning hours, sorting shared mailboxes, and drafting weekly status reports
- 발행처
- Microsoft Source — AI (공식 블로그)
- 발행일
- 2026-08-10
- 핵심 도구
- Microsoft 365 Copilot 내 'Agents' 패널 → 'New Agent' 버튼
- 제작 방식
- 코드 작성 없이 자연어 설명으로 에이전트 초안 자동 생성
- 제시된 단계
- 문제 정의 → 기존 에이전트 확인 → Copilot에서 제작 → 지식·출력 추가
Microsoft has published a guide on its blog, Microsoft Source, arguing that non-developers — ordinary office workers — can now build their own AI agents. The article opens with an example: at 2:13 a.m., an employee clicks a phishing link, and an AI agent automatically detects and locks the compromised account, checks whether malware has spread, then alerts the security team and drafts an incident summary. The guide's proposed process boils down to four steps. First, clearly define the problem you want to solve, then check Microsoft 365 Copilot's 'Agents' panel to see if a prebuilt agent already exists. If nothing suitable is available, click 'New Agent' in the same panel and describe what you want to accomplish in plain sentences — Copilot then drafts an agent for you. From there, you use 'Configure' to refine the tone, behavior, and scope of tasks, adding any necessary knowledge and output formats.
Understanding the difference between an agent and a chatbot clarifies what this guide is really about. A chatbot only answers questions, but an agent actually performs tasks — sorting emails, locking accounts, drafting reports. This shift isn't unique to Microsoft. On August 8, Moonshot AI unveiled Kimi Agent Swarm, which can run up to 100 sub-agents simultaneously, and around the same time AWS open-sourced Agent Skills, letting a coding agent handle the entire Bedrock policy verification process. Both of those examples were largely advanced automation aimed at developers. Microsoft's guide takes a different direction, targeting users with no coding experience at all — ordinary office workers who just need to organize emails or draft reports.
Ultimately, this guide reflects a broader trend: lowering the barrier to building AI agents. The assumption that working with large language models requires APIs and code is being challenged. Microsoft 365 Copilot subscribers can now build agents that automate repetitive tasks using just a few lines of natural language, with no separate development knowledge required. Whether such no-code tools will hold up reliably in real enterprise settings, however, remains to be seen.
