No-Code Conversational AI Building Service 'miibo' Releases 'Thinking Mode,' a New Feature Where AI Agents 'Autonomously and Repeatedly Deliberate'

miibo Inc. (Headquarters: Minato-ku, Tokyo; Representative Director: Masashi Kunugi) is pleased to announce the release of 'Thinking Mode,' a new feature for our no-code conversational AI service 'miibo' that allows AI agents to 'autonomously and repeatedly deliberate.'

What is Thinking Mode? Conventional AI chatbots typically provide a single, immediate response to a user's question. However, this approach is often insufficient for complex questions or tasks requiring multi-stage processing.

With the newly implemented Thinking Mode in miibo, the AI agent autonomously makes internal judgments and processes tasks step-by-step, as shown in the example below:

User: "Help me prepare for my business trip to Tokyo next week."

◼︎ AI Agent's Thought Process: ├─ Step 1: Confirm the business trip dates ├─ Step 2: List necessary reservations (transportation/accommodation) ├─ Step 3: Create a packing checklist ├─ Step 4: Check the weather forecast and provide clothing advice └─ Completion: Organize all information and provide a response

AI Agent: "I have summarized the following preparations for your Tokyo business trip next week..."

Features of Thinking Mode 1. Planned Task Execution In Thinking Mode, the AI agent first creates a to-do list and executes each item reliably. It can handle complex requests without missing any details.

2. High-Quality Responses Through Deep Thinking Instead of simple Q&A, the agent generates more accurate and comprehensive answers by repeatedly examining the request from multiple perspectives.

3. Transparent Thought Process The log function in the management console visualizes how the AI agent thinks and what steps it takes. You can review the AI's decision-making process and use it for further improvements.

4. Deeper Responses via Integration with Knowledge, MCP, and Connectors Thinking Mode reaches its full potential when combined with other miibo features: 'Knowledge Data Store (RAG),' 'MCP (Model Context Protocol),' and 'Connector' functions.

A. Deep Responses Across Multiple Information Sources Traditional RAG generated answers using only information retrieved from a single search. In Thinking Mode, the agent searches its knowledge base multiple times as needed, building up information to construct a final answer.

User: "Create a competitor comparison report for new product A."

◼︎ AI Agent's Thought Process: ├─ Step 1: [Knowledge Search] Retrieve specifications and features of new product A ├─ Step 2: [Knowledge Search] Retrieve information on competitor products B, C, and D ├─ Step 3: [Knowledge Search] Check the format of past comparison materials ├─ Step 4: Compile collected information into a comparison table ├─ Step 5: Organize the competitive advantages of our product └─ Completion: Generate the competitor comparison report

AI Agent: "I have created the competitor comparison material. New product A is..."

B. Autonomous Task Execution Linked with External Systems By combining it with MCP and Connectors, you can automate a series of processes from information retrieval to processing and action execution, as shown in the example below.

User: "Calculate this month's sales and share the report on Slack."

◼︎ AI Agent's Thought Process: ├─ Step 1: [Connector] Retrieve sales data from the sales management system ├─ Step 2: Analyze data and calculate month-over-month trends ├─ Step 3: [Knowledge Search] Check past report formats ├─ Step 4: Create a report including graphs and summaries ├─ Step 5: [MCP] Post the report to the designated channel via Slack API └─ Completion: "I have created the sales report and shared it in the #sales channel."

How to Switch to Thinking Mode Thinking Mode can be easily configured from the sidebar of the management console. The setup takes only two steps:

1. Open 'Model/Prompt' in the sidebar and check 'Use Thinking Mode' under 'Thinking Mode Settings.' 2. In the 'Thinking Mode Prompt,' set instructions regarding agent behavior, such as tool usage or multi-step processing.

That completes the setup. No coding or specialized knowledge is required. Anyone can immediately start building AI agents that utilize Thinking Mode without writing a single line of code.

Use Cases It is particularly useful in the following scenarios:

1. Complex Research Requests Example: Automatic generation of market research and competitor analysis reports. By searching and integrating competitor information, past reports, and industry trends from the Knowledge Data Store multiple times, it generates a comprehensive analysis report from a single instruction.

2. Multi-stage Business Support Example: Automation of recruitment or internal application flows. The agent autonomously plans and executes a series of steps: checking applicant info, verifying against criteria, assisting in pass/fail decisions, generating result text, and sending notifications via Slack.

3. Professional Consultations Example: Internal expert consultation desks for legal or accounting. By cross-referencing internal regulations, past contracts, and relevant laws, it provides specific advice rather than just generalities.

4. System Integration Tasks Example: End-to-end automation from sales report aggregation to Slack sharing. It can handle data retrieval from CRM, analysis, report formatting, and posting to Slack in one continuous flow.

Summary Thinking Mode is a feature that allows AI agents to autonomously and repeatedly exercise their 'power to think.' - Autonomously plans and executes complex tasks - Searches knowledge multiple times to generate deep, accurate answers - Operates external systems via MCP and Connectors - Visualizes transparent thought processes

Thinking Mode allows AI agents to think and judge autonomously to lead to the optimal answer. Expand the possibilities of your AI agents with miibo's Thinking Mode.