AI Agent + Shiji: A Step-by-Step Guide to Building an Automated Daily To-Do System

Why You Need an AI Personal Assistant

For content creators and entrepreneurs, the first thing you do every morning is often ask yourself: what's the most important thing I need to do today? This seemingly simple question actually consumes a significant amount of cognitive resources. You have to recall project progress, sort out priorities, and plan how to allocate your time — all of which can be fully automated with an AI Agent.

The phrase "cognitive resource consumption" isn't vague — it has solid theoretical backing in psychology. Nobel Prize-winning economist Daniel Kahneman, in Thinking, Fast and Slow, proposed that human decision-making operates through two systems: the fast, intuitive "System 1" and the effortful "System 2." Morning task planning is a classic System 2 activity that depletes your limited willpower — a phenomenon known as "Decision Fatigue." Research suggests that humans make roughly 35,000 decisions per day, and the sheer volume of low-value decisions crowds out the cognitive bandwidth needed for high-value ones. Delegating routine task planning to AI is essentially about freeing up System 2 resources, so entrepreneurs can reserve their deep thinking capacity for the core decisions that truly require creativity and judgment.

This article walks through a practical use case: how to deeply integrate an AI Agent with the Shiji task management tool to automatically generate and push daily to-do lists — shifting you from "figuring out what to do each day" to "reviewing what's already been planned for you."

Core Concept: From "Active Planning" to "Passive Review"

The Pain Points of Traditional Task Planning

Most people are still stuck in the "learning how to use it" phase with AI Agents — studying prompt techniques and testing various features. But the real value of an AI Agent lies in solving actual problems. For entrepreneurs juggling multiple business lines every day, the biggest pain point is task planning.

Take a typical entrepreneur's daily workload as an example. It might include:

• Advancing startup projects (market research, product development, etc.)
• Creating content for social media (public accounts, short videos, podcasts)
• Exploring new projects and benchmarking competitors
• Day-to-day operations and team collaboration

Spending time every morning or evening sorting through all of this is itself a drain on efficiency.

The AI Agent's Memory Capability Is the Key

AI Agents have a seriously underrated capability — memory and learning. You can give it a Scale (a capability configuration), allowing it to build a persistent understanding of how you work.

It's worth clarifying the fundamental difference between an AI Agent and a regular chatbot. A regular chatbot is stateless — every conversation starts from scratch with no retained context. An AI Agent, by contrast, has three core capabilities: perception (receiving information), planning (formulating a course of action), and execution (using external tools to complete tasks). The "memory and learning" capability mentioned here is typically implemented technically through a Vector Database — the AI converts your conversation content into high-dimensional vectors for storage, then retrieves relevant memories via semantic similarity during future conversations, enabling cross-session context understanding. This is distinct from the large language model's own context window, which has a length limit, whereas an external memory system can theoretically scale indefinitely.

Here's how to do it:

1. Initial conversation: Tell the AI what you do every day and what your role is (entrepreneur, content creator, etc.)
2. Feed in detailed plans: Share your project roadmaps, weekly plans, monthly goals, and so on
3. Set output rules: Ask it to push a list of your most important tasks for the day every morning at 8:30

After this kind of "training," the AI Agent becomes a personal assistant that truly understands you. It knows what your focus is this week, what your goals are next month, and can even dynamically adjust daily task priorities based on project progress.

Practical Guide: Complete Workflow for Connecting an AI Agent to Shiji

Step 1: Get the API Configuration from Shiji

Open the official Shiji website (you must use the web version), and find the OpenCloud integration entry point. The entire configuration process takes just two steps:

1. Download the Scale file: Find and download the corresponding Scale configuration file from the Shiji website
2. Configure the API Token: Go to Shiji's settings page, find the API Token option, and click to generate one. Copy and save it after generation — if it's ever compromised, simply revoke it and generate a new one

The Scale file is essentially a pre-defined capability package that tells the AI Agent how to interact with third-party tools. The underlying technical principle here is an API (Application Programming Interface) — a standardized protocol that allows different software systems to communicate with each other. An API Token is an authentication mechanism, like a digital key, ensuring that only authorized applications can access your data. Once the AI Agent reads the Scale file, it effectively gains the ability to call Shiji's API and perform actions like creating tasks, setting deadlines, and marking priorities. This pattern of connecting tools via APIs is known in the tech world as "Tool Use / Function Calling," and it's one of the most critical capabilities in today's AI Agent ecosystem.

Step 2: Have the AI Agent Read the Configuration File

Feed the downloaded Scale file to your AI Agent. If your agent is configured on the desktop client, simply have it read the file directly. The AI Agent will automatically complete the deployment within the OpenCloud configuration — the whole process is essentially plug-and-play, with no manual coding required.

Step 3: Have a Deep Conversation with the AI Agent

This is the most critical step. You need to have a thorough conversation with the AI Agent covering:

• Role definition: Who you are and what your work involves
• Task categories: What areas your work is divided into
• Specific plans: The current progress and goals for each area
• Output requirements: What format you want, when you want it, and where it should be delivered

Once the conversation is complete, explicitly tell it: "Please sync daily tasks to Shiji according to the Scale configuration."

Step 4: Daily Review and Continuous Iteration

Once configured, all you need to do each morning is open the Shiji app — your daily task list will already be automatically generated. Open "Today's Workspace" and you'll see the task list the AI has planned for you.

At that point, you only need to do three things:

• Review: Should this task actually be done today?
• Trim: Delete anything that doesn't make sense
• Add: If something's missing, continue the conversation with your AI Agent to fill in the gaps

This workflow — from AI generation to human review — falls under the broader category of Workflow Automation. Traditional RPA (Robotic Process Automation) executes repetitive tasks by simulating human actions, but it's rule-based and lacks comprehension. AI Agent-driven automation adds a layer of "understanding" and "judgment" — it doesn't just execute preset actions, it can dynamically adjust outputs based on context. This new paradigm, which combines the intelligence of large language models with tool-calling capabilities, is known in the industry as "Agentic Workflow," and is widely considered one of the most important directions for AI application in 2024–2025.

The Secret to Making Your AI Agent Understand You Better

Many people complain that their AI Agent is "dumb" and doesn't give precise enough recommendations. Here's a key insight: the smarter your AI Agent gets, the more you've talked to it.

If you've only had one conversation with it, its understanding of you will naturally be shallow. But as the number of conversations grows, the AI gradually learns your personality, work preferences, and decision-making patterns. This progressive process corresponds in machine learning to a shift from "few-shot learning" to "continual learning." Early on, the AI relies on general knowledge to make suggestions. As conversations accumulate, it builds a "user profile" through In-Context Learning — capturing your preferred work rhythm, your task prioritization logic, and how you allocate time across different types of work. This process resembles collaborative filtering in recommendation systems, but is far more personalized. It's worth noting that this learning doesn't modify the model's parameters (that would require retraining) — instead, it enhances the context available during each inference by continuously enriching the memory store, making outputs increasingly tailored to your individual needs.

In practice:

• Early stage: Task recommendations may be generic and require frequent adjustments
• Middle stage: The AI starts to understand your prioritization habits
• Later stage: The tasks it pushes become increasingly accurate, and you rarely need to make changes

This process is like training a human assistant — after the initial adjustment period, efficiency improves exponentially. Once fully calibrated, you won't even need to open the AI Agent's chat interface anymore; just check your tasks in Shiji each day.

Conclusion: The Right Way to Use AI for Automated To-Dos

The core value of this use case isn't in the technical complexity — it's in demonstrating a correct mindset for deploying AI in the real world:

1. Don't use AI for the sake of using AI — start from a genuine pain point
2. Fully leverage the AI Agent's memory capabilities so it truly "knows" you
3. Connect the AI's output with existing tools like Shiji to create an automated workflow
4. Keep iterating through conversation so the AI's output becomes increasingly precise

When an AI Agent is no longer just a chat window you have to actively open, but an intelligent assistant quietly planning your day in the background, it's finally delivering its true value. This may well be the ideal paradigm for AI integration into daily work — you handle the decisions, AI handles the planning.