/
Agentic AI Systems and Workflow Automation
Save to my account
Sign up
Agentic AI Systems and Workflow Automation
Agentic AI Systems and Workflow Automation
Study
1
Question
What distinguishes an agentic AI system from a simple chatbot?
42:59
Answer
An agentic AI system can pursue a user-defined task with minimal human guidance. It plans actions, selects tools, adapts to changes, and requests help when necessary, whereas a simple chatbot primarily answers questions without autonomous tool selection or workflow reasoning.
2
Question
How does an AI agent convert a user prompt into completed work?
1:13
Answer
The agent interprets the prompt as a goal, divides the goal into multiple tasks, creates a plan, selects appropriate tools, executes the tasks sequentially or iteratively, tests the results, and seeks human input when required.
3
Question
Why does a large language model need external tools for current information?
20:47
Answer
A large language model has a knowledge cutoff determined by the data used during training. Information created after that cutoff is unavailable from the model’s internal knowledge, so an external tool such as internet search must retrieve current information.
4
Question
What does knowledge cutoff mean for a language model?
19:47
Answer
Knowledge cutoff is the specific date up to which a model’s training data extends. Information appearing after that date is not part of the model’s internal knowledge.
5
Question
How does retrieval-augmented generation address outdated model knowledge?
24:14
Answer
Retrieval-augmented generation connects a language model to a knowledge base containing newer information. The system retrieves relevant data and supplies it to the model, which analyzes, reorganizes, and presents the information in a response.
6
Question
Why can retrieval-augmented generation be easier than fine-tuning?
22:11
Answer
RAG updates the connected knowledge base rather than retraining the language model. Adding new information therefore avoids the substantial computational resources, budget, and time associated with fine-tuning large models.
7
Question
How does a RAG system answer information already known by its language model?
25:41
Answer
The input is first evaluated against the language model’s internal knowledge. When the requested information is already available, the model can respond directly without retrieving material from the external knowledge base.
8
Question
How does a RAG system answer information absent from its internal knowledge?
26:21
Answer
The system sends the request to the knowledge base and performs semantic or similarity search. Relevant results are returned to the language model, which analyzes and refines them before producing the response.
9
Question
Why does a static RAG knowledge base struggle with continuously changing data?
28:02
Answer
A static knowledge base may not contain information that changes continuously, such as weather, temperature, or breaking news. Manually collecting and adding every new update is impractical, so the system can return incomplete or outdated answers.
10
Question
How does an agent improve upon a RAG system for real-time questions?
29:03
Answer
An agent can decide to use a live search tool when the requested information is current or unavailable internally. It retrieves fresh information, passes it to the language model for refinement, and returns the resulting answer.
11
Question
What makes tool use agentic rather than merely programmatic?
30:24
Answer
The application itself reasons about whether a tool is needed and selects when to call it. The developer provides the prompt and available tool access, while the agent determines the appropriate action for the request.
12
Question
How does an AI agent use a search tool for current news?
29:54
Answer
For a current-news request, the agent recognizes that internal knowledge may be insufficient, calls an internet search tool, retrieves information from current websites, and gives the language model the retrieved content for summarization and refinement.
13
Question
How does an AI coding agent build a Python game from one prompt?
34:01
Answer
It interprets the prompt as a goal, creates a multi-step plan, identifies required resources and tools, executes the plan one step at a time, and tests the developing application for problems.
14
Question
Why does an AI agent sometimes request human approval during execution?
35:43
Answer
Human approval provides control over consequential actions and allows a person to review or modify the proposed plan. After approval, the agent can continue executing the remaining workflow automatically.
15
Question
How does human-in-the-loop control differ from continuous human operation?
14:07
Answer
Human-in-the-loop control requires only limited intervention at selected points, such as reviewing a plan, answering a question, or approving a command. The agent performs the remaining workflow autonomously rather than requiring a person to direct every step.
16
Question
How do large language models support agent reasoning and tool selection?
34:14
Answer
The language model functions as the agent’s reasoning component. It interprets the request, determines whether a tool is needed, selects an appropriate tool, and helps transform tool results into a useful response.
17
Question
How did generative AI applications evolve before agentic systems?
17:40
Answer
Earlier applications used language models for prompt-based text generation, translation, summarization, chat, and other natural-language tasks. Multimodal systems later added image generation, while agentic systems added planning, tool use, workflow automation, and adaptive action.
18
Question
How do LangChain and LangGraph relate within agent application development?
4:08
Answer
LangGraph was developed as a product of the LangChain team. LangChain provides utilities such as language-model loading and prompt templates, while LangGraph is used to build and organize agent workflows.
19
Question
Why is asynchronous programming useful for agent workflows?
2:59
Answer
Agent frameworks use asynchronous programming so multiple agent operations can run in parallel. Parallel execution can support workflows in which independent tasks do not need to wait for one another sequentially.
20
Question
How can an agent automate a hiring workflow from one goal?
48:01
Answer
Given a hiring goal, the agent can draft and post a job description, monitor applications, revise the strategy when applications are insufficient, screen candidates, schedule interviews, prepare interview questions, draft an offer letter, and initiate onboarding with limited human approval.
21
Question
Why does an agent revise a job description when applications are insufficient?
53:47
Answer
A low application rate indicates that the current recruitment strategy may not meet the hiring goal. The agent can adapt by broadening the role description or promoting the job on platforms such as LinkedIn or Indeed, subject to approval.
22
Question
How does an agent use APIs as tools in workflow automation?
52:57
Answer
An API provides access to an external platform’s operation, such as posting a job on LinkedIn or Indeed. The agent receives access to the API and decides when to call the appropriate tool during execution.
23
Question
How can calendar access extend an agent’s hiring capabilities?
56:54
Answer
With calendar access, the agent can check the employer’s availability and propose an interview time. After receiving approval, it can schedule the interview and prepare an invitation email for the candidate.
24
Question
How does an agent maintain control while completing a multistep workflow?
58:14
Answer
The agent executes the plan step by step rather than acting randomly. At meaningful checkpoints, it can present a draft, request confirmation, incorporate feedback, and then continue with the next approved operation.
25
Question
How does the hiring example demonstrate adaptation to changing conditions?
49:34
Answer
The agent monitors recruitment results after posting the job. When applications fall below expectations, it changes the job description or activates promotion, then continues monitoring to determine whether the revised strategy works.
26
Question
What categories can an agent use when screening candidates?
56:11
Answer
The hiring workflow divides candidates into strong candidates, partial matches, and weak matches. The agent selects strong candidates according to the company’s requirements for further consideration.
27
Question
How does an agent support onboarding after a candidate accepts an offer?
60:01
Answer
After acceptance, the agent can initiate onboarding, send a welcome email, submit access requests, provision a laptop, and schedule an introductory meeting after receiving the necessary approval.
28
Question
How do RAG and agentic systems differ in their use of external information?
29:11
Answer
RAG retrieves information from a connected knowledge base, usually through semantic or similarity search. An agent can use a broader set of tools, such as search, storage, calendars, or cloud drives, and decide dynamically which tool is appropriate.
29
Question
Why does a language model remain central inside an agentic system?
34:14
Answer
The language model supplies the reasoning capability needed to interpret goals, create plans, select tools, analyze retrieved information, and refine outputs. Tools provide external actions or data, but the model coordinates their use.
30
Question
How can an agent reduce manual work in a business workflow?
39:47
Answer
An agent can automate repeated operational tasks such as customer support, email handling, data collection, recruitment, and scheduling. It performs the workflow automatically while involving a person at selected approval or feedback points.