Bilibili's Most Complete AI Agent Tutorial: From Zero to Advanced in Seven Days
Bilibili's 748-episode AI Agent course builds a systematic learning path through Fundamentals, Advanced, and Practice modules.
This article outlines the "748-episode" systematic AI Agent course launched by a Bilibili creator. Addressing scattered tutorials that leave learners unable to independently develop agents, the course designs three progressive modules: Fundamentals (Agent principles, Prompt engineering, workflow design), Advanced (RAG knowledge bases, multi-Agent collaboration, tool calling, mainstream frameworks), and Practice (personal knowledge base assistant, intelligent customer service Agent, automated office assistant). It also provides a learning roadmap, source code, and prompt templates. The article advises viewing the "seven days from novice to expert" marketing phrase rationally and treating the course as a starting point rather than an endpoint.
In an era of rapid AI iteration, AI Agent technology has become a hot topic of interest for everyone from developers to working professionals. A Bilibili creator has launched a systematic AI Agent course billed as "748 episodes in total," aiming to solve the common problem in the market: scattered tutorials that leave learners unable to independently build agents after finishing them. Based on the course's public introduction, this article outlines its content architecture, design approach, and target audience, offering reference for readers seeking a learning path.
Course Design: Rebuilding the Learning Path from Pain Points
In the opening, the creator mentions that before making the course, they revisited a large number of Agent tutorials on Bilibili and found a common phenomenon: whether viral hits or niche tutorials, most content consists of scattered knowledge points, and learners find it hard to independently develop agents after watching. This observation points to a typical shortcoming of current AI Agent learning resources—a lack of systematicity and a closed loop of hands-on practice.
According to the Bilibili creator, the course took three months to complete, organizes the core knowledge system of Agents, and explains it with real-world cases. The course emphasizes "outputting valuable content every second," attempting to balance beginner-friendliness with practical skill development. For zero-foundation learners, this progressive design from principles to projects is more actionable than simply piling up concepts.
Three Modules: The Progressive Logic of Fundamentals, Advanced, and Practice
The course is clearly divided into three modules—Fundamentals, Advanced, and Practice—forming a stepped learning path.
Fundamentals: Building an Agent Cognitive Framework
The Fundamentals module focuses on Agent principles, Prompt engineering, and workflow design. These three areas constitute the foundational capabilities for agent development: understanding how an Agent perceives its environment, calls tools, and makes decisions; mastering Prompt design techniques to control model behavior; and learning to orchestrate complex tasks with workflows. For zero-foundation learners, the goal at this stage is to get started quickly and build a complete cognitive framework rather than getting lost in details.
Prompt engineering refers to the systematic method of carefully designing the prompts input to large language models to guide them toward desired outputs. It is not simply "asking a question" but involves a series of techniques such as role setting, task decomposition, few-shot examples, chain-of-thought prompting, and output format constraints. In Agent development, Prompt engineering directly determines the reliability of model tool calls, instruction parsing, and multi-step reasoning, making it a core link between natural language intent and programmatic behavior. Mastering Prompt engineering means developers can significantly improve an Agent's task completion rate and stability without modifying model parameters.
Advanced: Mastering Core Development Capabilities
The Advanced module covers RAG knowledge bases, Agent architecture, multi-Agent collaboration, and tool calling, with hands-on development using mainstream frameworks. RAG (Retrieval-Augmented Generation) is currently a key technology for enabling Agents to access external knowledge; multi-Agent collaboration corresponds to the trend of decomposing and dividing complex tasks. The design intent of this module is to move learners from "knowing how to use" to "being able to develop," mastering the core skills for building practical Agent systems.
RAG (Retrieval-Augmented Generation) is a technical architecture that combines external knowledge retrieval with large model generation. Its basic process is: first split documents and vectorize them into a vector database; when a user asks a question, the system first retrieves the most relevant text chunks, then feeds these chunks as context into the large model to generate an answer. RAG effectively alleviates the problems of outdated knowledge, hallucinations (fabricated facts), and inability to access private data in large models, enabling Agents to answer accurately based on enterprise documents or personal knowledge bases. For building knowledge-base assistant Agents, RAG is an essential foundational capability, involving key engineering steps such as text splitting strategies, Embedding model selection, vector retrieval, and reranking.
Multi-Agent collaboration refers to decomposing complex tasks and having multiple Agents with different roles and capabilities work together. Common patterns include: an orchestrator Agent responsible for planning and dispatching tasks, and multiple executor Agents handling subtasks such as retrieval, computation, and writing, sharing intermediate results through message passing. This architecture mimics how human teams work and can overcome the limitations of a single Agent in context length, tool variety, and reasoning depth. Typical application scenarios include automated research report generation, complex customer service ticket handling, and software development assistance. Understanding multi-Agent collaboration requires mastering design considerations such as task decomposition, communication protocols, state management, and conflict resolution.
Practice: Turning Projects into Experience
The Practice module adopts a hands-on project-based approach, including cases such as a personal knowledge base assistant, an intelligent customer service Agent, and an automated office assistant. The project selection covers three high-frequency scenarios—personal productivity, enterprise services, and office automation—and is well representative. By completing the full project workflow, learners can turn the knowledge from the first two stages into demonstrable hands-on experience.
Supporting Resources and Learning Support
To lower the barrier for zero-foundation learning, the course provides resources such as a complete learning roadmap, project source code, prompt templates, deployment tools, course notes, and practical case notes. The creator says these materials can be obtained by leaving a comment in the comments section. The completeness of supporting resources largely determines the effectiveness of self-study; especially for beginners lacking engineering experience, ready-made prompt templates and deployment tools can significantly reduce time spent stumbling over pitfalls.
Who It's For: Clear Boundaries and Expectations
This course is suitable for learners who want to systematically study AI Agents and quickly improve their practical skills through projects. The creator emphasizes "job-ready upon completion" in the introduction, but this should be viewed rationally: the course provides a knowledge system and project experience, while actual employment competitiveness also depends on individual practice intensity, technical foundation, and market conditions. For readers with existing programming foundations who want to enter the Agent development field, this course's structured path is worth referencing; for complete beginners, it is advisable to supplement basic programming knowledge in parallel.
A Rational View of the "Quick Mastery" Narrative
"From novice to expert in seven days" is a common marketing phrase in courses. Judging from the course's content volume, 748 episodes clearly cannot be fully absorbed in seven days; a more reasonable understanding is that seven days can complete the introductory Fundamentals module and establish a learning rhythm, while Advanced and Practice require sustained investment. AI Agents involve multiple technical points such as Prompt engineering, RAG, and multi-Agent collaboration, each of which requires hands-on practice to master. Treating the course as the starting point of a systematic learning path rather than the endpoint is perhaps a more pragmatic attitude.
Overall, the value of this course lies in integrating scattered Agent knowledge in a structured way and providing actionable project cases. For readers looking for an entry point to learning AI Agents, it can serve as a reference framework, with a learning plan tailored to their own foundation.
One-line Summary
Bilibili's 748-episode AI Agent course builds a systematic learning path through Fundamentals, Advanced, and Practice modules.
Paragraph Summary
This article outlines the "748-episode" systematic AI Agent course launched by a Bilibili creator. Addressing the problem of scattered tutorials that leave learners unable to independently develop agents, the course designs three progressive modules: Fundamentals builds a cognitive framework around Agent principles, Prompt engineering, and workflow design; Advanced focuses on RAG knowledge bases, multi-Agent collaboration, tool calling, and hands-on work with mainstream frameworks; Practice turns knowledge into experience through projects such as a personal knowledge base assistant, an intelligent customer service Agent, and an automated office assistant. The course also provides supporting resources such as a learning roadmap, source code, and prompt templates to lower the barrier to entry. The article reminds learners to view the "from novice to expert in seven days" marketing phrase rationally and suggests treating the course as the starting point of systematic learning rather than the endpoint.
Related articles
VERS: Editing Activation Explanations to Halve Reasoning Model Tokens
VERS guides reasoning models by editing activation explanations, cutting 45.8% of tokens on GSM8K with only a 0.82 pp accuracy drop and transferring to larger models, but losing accuracy on competition problems.
Finding Errors Is Not Enough: Selective Model Escalation for Entity-Targeted Sentiment Analysis
This study explores selective model escalation for entity-targeted sentiment analysis, comparing data collection ordering, uncertainty-driven re-examination, and model routing—finding that minority-class priority only helps at very small data volumes, the oracle upper bound far exceeds actual gains, and escalation must be tailored to each dataset.
WAS-PO: Using Whitened Advantages to Solve the Correlation Trap in Multi-Reward Reinforcement Learning
WAS-PO paper proposes whitening advantages to correct correlation bias in multi-reward RL: when rewards are correlated, naive weighted summation effectively applies Σ^{1/2}w instead of w. Experiments show training reward improves steadily, but held-out accuracy does not improve in sync, and under two rewards it actually falls behind.