AI-assisted engineering, not magic
Frontend, backend, mobile: AI দিয়ে কাজ করলেও ফল যাচাইয়ের দায়িত্ব তোমার।
Guess first
Pick an answer. You will see if you were right at the end of the lesson.
Does the model choose the next action, or does code always choose it?
Real-world analogy
What is it?
This course is for frontend, backend and mobile developers, from junior to senior. You need basic coding in one language and basic Git. Start with the junior exercises if AI agents are new to you; take the senior extensions when you can explain and verify the foundations. A model generates a candidate answer. A coding agent can choose tools and iterate. A workflow follows a predefined path. A harness is the surrounding execution system: tools, permissions, state, limits and checks. None of these removes engineering accountability.
Real-world relevance
One shared support-ticket product, three optional entry points. Frontend: Ask an agent to reproduce an inaccessible form error, then inspect its proposed fix. Backend: Ask an agent to trace a duplicate API request and propose an idempotency fix. Mobile: Ask an agent to reproduce an offline sync failure and inspect its tests.
Key points
- Name the job — Choose one support task: classify a synthetic ticket and propose a draft. Do not begin with an autonomous company.
- Separate three abilities — Using a coding assistant, integrating inference into a product, and training models are different skills.
- Keep a baseline — A deterministic keyword rule may solve a narrow category cheaply. Keep it as the comparison.
- Define failure — A fluent reply that exposes a private ticket is a failure even if its grammar is excellent.
- Own the outcome — The engineer chooses the requirements, checks evidence and accepts the change.
- Inspect uncertainty — Ask what evidence supports a claim and what input would change the result.
- Keep core skills — Offline sync, authentication, debugging and clear communication remain useful. No skill is guaranteed immune to automation.
- Frontend practice — Ask an agent to reproduce an inaccessible form error, then inspect its proposed fix.
- Backend practice — Ask an agent to trace a duplicate API request and propose an idempotency fix.
- Mobile practice — Ask an agent to reproduce an offline sync failure and inspect its tests.
- Start here and assess honestly — Read the course guide in Learning Resources. Choose one stack, complete the exercises and keep evidence. Quiz completion records practice, not professional competence. The offline lab uses Python fixtures; it is not a live AI service.
Quick check
Answer from memory — scrolling up first makes it stick less.
What artifact demonstrates this lesson?
Which statement is defensible?
Code example
function baseline(text) {
return text.toLowerCase().includes("crash") ? "bug" : "review";
}
console.log(baseline("App crashes on save")); // bug
console.log(baseline("No crash, need a receipt")); // bug: a false positiveLine-by-line walkthrough
- 1. JavaScript: this function works in a browser console or Node.js. It returns a category without sending anything. Both inputs contain crash, so the second is a deliberate false positive. This is deterministic code, not an agent. The same idea can be implemented in any language.
Spot the bug
Calling every cron job an agentNeed a hint?
Show answer
Explain like I'm 5
Fun fact
Hands-on challenge
More resources
- Start here: learning paths, capstone, assessment and limitations (Teamz Lab)
- Building effective agents (Anthropic)
- Download the free offline lab (Python, tests and instructions) (Teamz Lab)
Lock it in