Emergence of AI Agents Current Release Status of AI Agents by Major Companies Is This the Second Boom of AI Agents? Lessons fr om Past Failures 3 Key Reaso ns Behind the Current Boo m of AI Agents Understanding AI Agents What are AI Agents? Specific examples, applications, and value of AI agents Intelligence and autonomy as key factors for AI agents Toward Utilizing AI Agents for Real-World Applications Current Landscape of AI Agents in Business Key Featur es of Business Processes Most Suitable for AI Agents Ho w to Utilize AI Agents
presented a demo of ai agents at the developer conference 'Google I/O’. Agents operate various applications, such as placing orders, as user alternatives. After Nov 2024, ai agent services were released in Japan. NTT Data, Fujitsu, NEC, SCSK, CTC, Softbank, CyberAgent https://io.google/2024/explore/a6eb8619-5c2e-4671-84cb-b938c27103be/intl/ja/ Nikkei publishes articles about AI agents almost daily.
Big Tech has been releasing agents that connect various software. Aiming for a world where agents handle tasks without human control of software. Company Product Use case Google Project Mariner Web browser operation Project Astra Device assistant Jules Code development support Vertex AI Agent Builder Internal data, SaaS app integration Deep Research Market research, analytics Microsoft Copilot Studio Agent Builder Internal data, Office products, SaaS app integration SAP Joule Internal data, ERP, SaaS app integration Salesforce Agentforce Internal data, CRM, SaaS app integration UiPath Autopilot for Everyone UiPath Agent Builder Internal data, RPA, SaaS app integration Next-Generation RPA
Past Failures The first boom began with engineers publishing repositories for autonomous agents ChatGPT, initially designed as an AI dialogue system, has now been utilized as a reasoning engine for autonomous automation systems. Hackathons for developing agents using BabyAGI and AutoGPT, have been actively organized overseas. Why has the boom faded away? Fully autonomous systems had unstable task success rates. The code was complex and difficult to understand. blog:https://resources.parcha.com/agents-arent-all-you-need/ 2022年11月 2023年 3月 11月 2024年 3月 5月 9月 11月 2025年 Agents Announced by Google and Microsoft Business-specific Agents Salesforce agentforce ChatGPT AWS Agent Builder OpenAI GPTs Autonomous Agents First Boom Second Boom
in early 2024, completes tasks upon spoken commands. Examples of Tasks Ask AI Pin about the restaurant's reviews Request translations for local languages at travel destinations Send messages or take memos via voice commands Results (As of April 2024) Success rate: approximately 1 in 5 attempts Long response times Limited integration with external apps Reduced screen time, allowing users to focus more on the real world Task success rate, execution time, and the variety of supported tasks remain challenges. Despite ongoing issues, valuable insights into user needs and potential use cases were gained. Humane AI Pin Review: https://www.theverge.com/24126502/humane-ai-pin-review
1. Technical Factors: Improved Performance of GenAI Enhanced task success rate stability: structured outputs, stronger Function Calling capabilities. Improved reasoning abilities: OpenAI’s advanced models o1. 2. Business Factors: Limitations of DX via RAG Algorithms (Japanese Boom Factors) Increased demand for more advanced tasks leveraging internal documents, which RAG algorithms cannot fully address, such as retrieving top search results. 3. Business Factors: Added Value from Software Automation (Global Boom Factors) Agents operating software have become a popular research topic. Shift from human operation to agents utilizing SaaS platforms.
the environment to achieve goals. Key features include language understanding, dialogue, and reasoning. Intelligence and autonomy are essential. • Computer • Game • Home • Internal data • Search • Code Generation • API Calls Env Action Plan Perception Memory Conversation Knowledge Learning Cognition Reasoning AI Agents
a CSV and ask questions The agent plans and executes 1. Creates action plans independently 2. Executes actions autonomously 3. Understands information from the environment 4. Identifies issues and revises plans Key Features of AI Agents User Agent Env
humans from boring, time-consuming, and undesirable tasks. Improve the quality of the most valuable business operations. Replacing busy experts designers, developers, data analysts, mentors, and marketers. Accelerating Personalized Experiences AI agents manage user preferences to enhance customer experiences with services. Liberating Humans from Understanding and Operating Specialized Software AI agents navigate complex, multifunctional software to assist with onboarding and operational tasks for new employees.
The ability to understand, reason, and adapt knowledge effectively. Autonomy: The ability to independently perform tasks and make decisions without human intervention. Autonomy Intelligence Software development Data analysis Computer operations Review, Q&A Recommendations Decision support Routine tasks Simple app operations Requires advanced reasoning and flexible adaptability. Needs PDCA with a distant goal. Use search or retrieve APIs based on the situation Defined workflows with limited variability With no unexpected events, the goal is reached in just a few steps. Requires critical and logical thinking with contextual understanding of business operations and knowledge. high low high More employee-like behavior Toward more general-purpose goals. Product PoC R&D Google, Microsoft OpenAI, Anthropic Microsoft, UiPath SAP, Salesforce
are searching for practical use cases (A defining feature of 2025). Market pressures have fueled the hype and contributed to overpromising. Challenges in Replacing Employees in Japanese Companies • Agent technology is not yet mature for replacing employees. • LLMs lack the high-context understanding required for tasks, needing instructions every time. • Requires organized business documentation, experience reuse, and workflow standardization. Compromises for Practical Implementation • Workflow management and simplifying alternative tasks are essential. • Avoid high-stress tasks and start with single-department operations. Steps Toward Deployment • Focus on quickly analyzing business processes, building agents, and evaluating their limits. • Recognizing current technical boundaries and helping employees adapt to AI is key.
Diverse Patterns of Final Outcomes • Tasks are such as brainstorming ideas or creating research reports. • Agents generate diverse patterns of outcomes through trial and error in collecting and processing information. Need for Trial-and-Error and PDCA Cycles • Tasks that require setting new directions based on past successes and failures. • Activities such as experiments, development, training, satisfaction improvement, and knowledge management benefit from PDCA cycles. Business Processes with Many Exceptions and Variations • Tasks like inspections where criteria vary depending on the contents and require detailed checks. • Rules such as "if-then" are insufficient due to frequent changes in criteria, requiring agents to search documents or verify specifics. Expectation for Agents in Time-Consuming Development Processes • AI agents are expected to replace tasks that take significant time for testing and validation.
using generative AI agents: Use Existing Agent Products Examples: Cursor and Devin agents for development support, Deep Research and Perplexity for market research. Use an Agent Builder Easily integrate with internal data and deploy to understand usage scenarios. Recommended for those without development skills. May lack the ability to create highly specialized business solutions. Build from Scratch Effective for optimizing specific tasks but requires several months for development and validation. Involves designing agent architecture, prompt engineering, and tool development. Not as straightforward as RAG; delivering immediate business value is challenging, requiring significant engineering expertise.
DX department. Why Generative AI Agents Are Gaining Attention • In Japan, challenges with RAG PoC and advancements in LLMs have driven the AI agent market. • Recognition has been significantly influenced by marketing and news from companies like Microsoft and Google. The Current State of Generative AI Agents • Generative AI agents are intelligent systems that interact with their environment to make decisions. • Their applications often involve multi-step tasks that require "process" management. Considerations for Business Adoption • The key to 2025 lies in evaluating which business processes can be replaced by agents. Summary