AI Agents
Put an AI model inside an automation so it can decide and act — and learn where that is genuinely useful and where it is the wrong answer.
- 6 lessons
- about 17 minutes (estimate)
- Free
- Free BvLogic Certificate of Completion
- Video in اردو
- Exercises in اردو
- 18 quiz questions
- 6 exercises
- 1 graded lab
Earn your free BvLogic Certificate of Completion.
A digital credential issued by BvLogic Academy to recognize successful completion of this introductory learning activity.
Not a government qualification or regulatory certification.
That was a one-minute preview. The full course video is for signed-in learners. Sign in to watch it here, or create a free BvLogic ID; it takes a minute, and this course is added to it.
What you'll be able to do
- Separate the real idea from the marketing around it.
- Define tools an agent can call, and understand that the tool list IS the permission list.
- Manage what an agent knows, and understand why long conversations degrade.
- Constrain an agent with mechanisms that actually hold.
- Recognise the many cases where a plain automation is better, and say so.
- Build a real, safe, read-only agent end to end.
Course content
- 1 What an agent actually is Separate the real idea from the marketing around it.
- 2 Tools: giving a model hands Define tools an agent can call, and understand that the tool list IS the permission list. Sign in to open
- 3 Memory and context Manage what an agent knows, and understand why long conversations degrade. Sign in to open
- 4 Guardrails, and why they belong in code Constrain an agent with mechanisms that actually hold. Sign in to open
- 5 When an agent is the wrong answer Recognise the many cases where a plain automation is better, and say so. Sign in to open
- 6 Building a simple agent Build a real, safe, read-only agent end to end. Sign in to open
Read what each lesson covers
- What an agent actually is. An agent is an automation where a model chooses the next step: a goal, a list of permitted tools, and a loop. It earns its place only when the steps cannot be known in advance.
- Tools: giving a model hands. A tool is a function the model may ask you to run; your code decides whether to run it. Keep tools narrow, because the tool list is the permission list.
- Memory and context. A model remembers nothing between calls, so every call resends the history. Cost grows with each step and long runs drift, so critical rules are restated rather than trusted.
- Guardrails, and why they belong in code. A rule in the prompt is a request; a check in the code is a guarantee. Limit tools, validate every argument, cap the loop, rate-limit actions and require a human for anything irreversible.
- When an agent is the wrong answer. Known steps, exact answers, high volume or serious consequences all call for plain automation or a single model call. Knowing when not to build an agent is what makes you worth hiring.
- Building a simple agent. A read-only enquiry assistant with five guardrails that drafts replies for a person to send. You test it on real past enquiries, including awkward ones, and count sendable, needed editing and wrong.
Course lab
Graded lab A read-only enquiry assistant with five guardrails, measured Open the lab →Who this course is for
- Developers and automation builders who want a model to decide steps, not just follow them.
- Business owners who have been sold an 'AI agent' and want to know what it can safely be trusted with.
- Anyone who finished AI Fundamentals and Prompt Engineering and wants the next practical step.
Try the first exercise (in Urdu)
کسی ایسے کاروبار کا ایک کام لکھیں جسے آپ جانتے ہوں اور جہاں اقدامات پہلے سے واقعی طے نہ ہو سکتے ہوں۔ پھر لکھیں کہ ایک agent کو کون سے tools درکار ہوں گے۔ پھر ہر tool کے لیے لکھیں کہ اگر وہ غلط استعمال ہوا تو بدترین نتیجہ کیا ہو گا۔
Using this in Pakistan
- AI automation entry roles advertised in September 2026 (Junior AI Automation Engineer, AI Automation Specialist, AI Video Content Creator) are mostly in Lahore; one Lahore junior role offered Rs 50,000 to 80,000 a month. Source
- The data is still thin: on 30 September 2026 Rozee.pk returned about 110 'AI automation' results across Pakistan and 1 in Faisalabad. Source
- The tools these adverts name: n8n, Make, Zapier, OpenAI and webhooks. Source
Questions
Is it really free?
Yes. No payment and no card; lesson 1 needs no account.
Is there a certificate?
Yes, when you complete every lesson.
Can I learn in Urdu?
Every lesson has an Urdu exercise alongside the English reading.
How long does it take?
About 17 minutes of reading, plus the exercises.
What comes next?
The paid Artificial Intelligence programme, in Faisalabad and live online.
Artificial Intelligence
AI & Generative AI
This free course is the introduction. The programme is taught by a trainer, marked by the academy, and carries the syllabus in full: the part an employer asks about.
See the full syllabus and feeRelated free courses
- AI Ethics, Disclosure and SafetyUrdu video
- AI for AccountantsUrdu video
- AI for Clinics and DoctorsUrdu video
- AI for Everyday WorkUrdu video
- AI for TeachersUrdu video
- AI FundamentalsUrdu video
AI courses in Pakistan · All free courses · Government free courses open now