Your first AI agent that uses tools
Take this if you want to know what people mean by an agent, in enough detail to build one and to distrust it appropriately.
What you’ll be able to do
You can define a tool a model may call, run a control loop that decides when to call it, trace exactly why the agent did what it did, and identify the loop's failure modes.
What it assumes
Confident Python, comfort with JSON, and having called a model API at least once.
- Price
- ₹499
- Length
- 2 weeks
- Sessions
- 8
- Effort
- ~6 h/week · 12 h total
- For
- Class 11–12 · College · Working professional
Adds it to My courses on your dashboard so you can find it again. Payment is not open yet — nothing is charged.
What’s taught, in order
- 01
An agent is a loop, not a model
Separating the model's proposal from the code that decides whether to act on it. Getting this boundary right in the first session is what keeps the rest of the sprint debuggable.
- 02
Defining a tool the model can call
Writing a tool schema, handling the arguments the model gets wrong, and validating before executing. Most agent bugs are argument bugs, met here on purpose.
- 03
Deciding, acting, observing
The full cycle with every step logged, so you can read afterwards exactly why the agent chose a tool. An agent you cannot trace is an agent you cannot fix.
- 04
How it goes wrong
Loops that never terminate, tools called with plausible nonsense, and confident summaries of failed actions. Building the guardrails yourself is more instructive than being handed a framework's.
What you build
Build an agent with at least two real tools that answers questions about something in your own life — your timetable, your expenses, a folder of files. Ship the trace log of five successful runs and three failed ones, with an annotation on each failure saying which part of the loop was responsible.
Check it yourself against this
- At least two tools that do real work, not two prompts wearing tool costumes.
- Arguments are validated before execution, and you can show a rejected call.
- Trace logs make each decision readable after the fact.
- Three genuine failures diagnosed to a specific stage of the loop.
- A termination guard exists, and you can demonstrate it firing.
- You can state one task you would not let this agent do unsupervised.
Then answer this
Your agent called the right tool with wrong arguments and then confidently summarised the wrong result — which part of your loop should have caught that, and why did it not?
Two or three sentences, in your own words. If you cannot, the course is not finished — go back to the module it came from.
What this course does not do
One agent, two tools, no framework, no memory across sessions and no multi-agent coordination. Planning, persistent memory, agent evaluation at scale and the safety work that real deployment needs are the Agentic AI phase course rather than this sprint.
Where this leads
This sprint is the opening of the ten-week Agentic AI phase, which continues into planning, memory systems, the Model Context Protocol, multi-agent setups, evaluation and agent safety.
FAQ
- Do we use LangChain or LangGraph?
- Neither, deliberately. You write the loop yourself in about a hundred lines, because a framework hides exactly the control flow this sprint is about. The last session maps what you built onto what the frameworks provide, so you can pick one up afterwards knowing what it is doing.
- Will this cost me API credits?
- A small amount, and the sprint is sized to fit comfortably in a free tier. There is also a mocked mode for developing the loop without any model calls, which is both cheaper and a better way to debug control flow.
- Is an agent just a chatbot with extra steps?
- The difference is that an agent takes actions with effects, which is why the sprint spends a quarter of its time on validation and failure. A chatbot that is wrong wastes your time; an agent that is wrong changes something.
Other courses
This is a standalone course. It does not add weeks to the 78-week curriculum or change your roadmap. If you want the whole path instead, the full programme is ₹399/month · ₹2,000/year — see what it covers.