Introduction to Agentic AI and AI Agents

What lets an agent read information, choose a next step and change a file? A shirt that still needs folding shows the difference between describing an action and doing it.

Separate what a model can explain from what an application can actually do, using a folded shirt and an edited spreadsheet.

See how tools let an agent use a connected service. Inspect the actions a calendar connection offers, tell reading from changing, and check that the event was really created.

How the lectures connect, and how the summaries, prepared files and quizzes help you practise each idea.

Follow a calendar request from plain-language appointment details to the event Google Calendar returns, and understand an API as the agreed route software uses to ask for information or actions.

Meet the 4 moves of every agent run on a short project-software comparison, and see how a failed check sends the work back for another step.

Follow Daniel through 2 unpaid-invoice decisions with Look, Think, Do and Check, and see the evidence he compares with each draft.

Tell the information supplied for today's client reply from an instruction saved for later. Understand the context window, tokens and compaction, and recover a lost detail by reading the source again.

An LLM writes language one token at a time from learned patterns and the current context, which is why a fluent factual answer can still be wrong.

Match text, image, audio or video inputs with the result you need, then check what the model supports and what the application can reach.

Compare models on performance, cost or plan usage, and whole-job speed. Use benchmarks to shortlist, then decide on your own checked work.

Run one invoice-review task on 2 models with the same files, request and checks, and compare accuracy, corrections, waiting and plan usage.

Check what stays available when you switch models inside one app, and what you must supply again when you move to another app.

Connect models, tools, APIs and context on the examples from this section, and explain how a model can change while the source facts stay the same.

Agent, chatbot, workflow and automation get used as if they mean the same thing. Set the goal for this section: recognise the shape of the work and explain it in ordinary words.

Tell an answer written from existing knowledge from a run that uses tools and chooses its next actions, and judge the work performed rather than the product's label.

Compare one supplier-price job done with prepared software steps and with model-chosen actions, and see how fixed calculations and agentic research can work together.

A request, a schedule or an event can start a job. Separate what starts the work from how its steps are chosen, and a notification from finishing the job.

A customer-order email connects assistant, agent, agentic AI and automation, and shows why preparing a reply is separate from permission to send it.

12 useful sentences for talking about AI agents at work, each tied to the lecture where the idea became clear.

Use the shop's order email to tell an answer, a prepared workflow and model-chosen actions apart, and separate the trigger from the steps that follow.

A customer asks to cancel order 24. Decide which company rule, live order fact and approval matter before the lectures answer each one.

Priya wants staff questions answered without finding each file herself. Go from an approved knowledge base to embeddings and a vector database, then test the answer against the policy passage behind it.

Maya checks whether her contract allows a supplier's price rise. Learn what training changes, how supplying a file differs, and what to check about storage, access and retention.

Trace an invoice-status question through a model's tool request, the app and the accounting API. Understand function calling and MCP, and why being connected is not the same as having permission.

Build a first set of 20 checked questions, including 3 hard cases, and compare answers with expected results. See how instructions hidden in outside material can redirect an agent, and how to test for it safely.

Short customer replies show why wording varies. Understand temperature and top-p without any maths, and how fine-tuning differs from answer settings.

Follow a weekly supplier-price job through where it runs, how it is built and who maintains it, from LangChain and CrewAI to n8n, Zapier and Make.

Compare 2 support offers on one request to cancel order 24, and find the rule, the live order status, the reply draft, the cancel permission and the proof that it happened.

Resolve the order-24 request with the company rule, live order status and human approval, and connect RAG, APIs, MCP and testing to that one decision.