Vibe Engineering with AI Coding Agents
Aik practical hands-on course jo aapko “vibe coding asal mein kya hai?” se le kar AI agent ke sath real apps ship karne tak le jata hai — aur sath hi yeh samajhata hai ke background mein asal mein kya ho raha hai. SMIT mein Muhammad Uzair Khan ka sikhaya gaya course.
Aapka class-by-class roadmap
Shuru se aakhir tak padhein, ya kisi bhi topic par jump karein. Har page mukammal hai aur agle page se juda hua hai.
Jadu dekhein, phir usay samjhein
Sentence se 3D game generate karein, phir mental map banayein: Missing Manual, vibe coding vs. agentic engineering, AI adoption ke 8 stages, aur paisa kahan kharch hota hai.
Engine ke andar kya chal raha hai
Tokens, memory, reasoning, tools aur loops — woh 4 tricks jo autocomplete ko aik agent bana deti hain. Phir context engineering, agents.md, aur professional workflows ka irteqa.
Asal mein build karein
Tools install karein, aik behtareen agents.md likhein, aur Cursor ko YOLO mode mein chala kar aik working Kanban board banayein — planning, generating, testing, aur feedback ke sath iterate karna.
Model chunein, phir YOLO karein
Sahi model select karein, 5 “be the boss” usool seekhein, OpenRouter set up karein, aur AI digital twin ke sath portfolio site vibe-code karein — phir doosre model se review karwayein.
Claude Code — aur iske competitors se miliye
VS Code mein Claude Code install karein, CLI seekhein — /init, /context, tests run karna — phir OpenCode, AMP, OpenRouter aur local Ollama ke zariye free aur cheap models chalayein.
Commands, workflow aur autonomy
Rozmarrah ki commands aur shortcut keys, sessions, checkpoints aur Git kaise aapko mehfooz rakhte hain, real
--dangerously-skip-permissions YOLO mode, aur ghanton tak build karne wale Ralph loops.
MCP, Skills aur Plugins
Claude Code ko supercharge karne wale big three — doosron ke tools MCP se connect karein, skills se expertise add karein, plugins se sab bundle karein — aur kab kya use karna hai iska wazeh rule.
Jira, MCP aur FeatureDev
Mukammal development lifecycle chalayein: Jira se kaam lein, Anthropic ke disciplined FeatureDev process ke zariye Claude Code chalayein, reviewed GitHub PR ship karein — aur pro level debugging karein.
Pre-Legal shuru se aakhir tak banayein
Project CLAUDE.md aur custom Cerebras skill likhein, V1 ko Jira tickets ke tor par plan karein, aur multi-user AI legal-document SaaS — FastAPI, Next.js aur SQLite — ticket by ticket build karein.
Poore course ko yakja karein
Pehle do hafton ko consolidate karein — eight stages, control↔autonomy dial, LLMs asal mein kaise kaam karte hain, pro toolkit, aur finale se pehle master cheat-sheet.
Controlled chaos: multi-agent par jayein
Custom slash commands, sub-agents (Codex cameo ke sath), event-driven hooks, aur aapka apna plugin + marketplace — sab kuch FiNALLY trading-app capstone project par.
Pehle kaam karte dekhein — phir samjhein ke kya dekha
Instant gratification pehle, theory baad mein. Dissect karne se pehle aap iski taqat mehsoos karenge.
Welcome & aapka pehla build →
Cursor se miliye aur aik single prompt se playable 3D shooter game generate karein — baghair koi code likhe.
The Missing Manual →
Yeh course kyun wajood mein aaya, kiske liye hai, vibe coding vs. agentic engineering, aur teen surfaces.
Eight stages & cost ki baat →
ChatGPT se agent orchestration tak Steve Yegge ki ladder — aur AI spend par imandaranah guftagu.
Theory ka aik aham din — jo har lehaz se qeemti hai
Mechanics ko samajh lein to aage ki har trick samajh aa jayegi. Experts: chahein to skim kar sakte hain.
LLMs asal mein kaise kaam karte hain →
Tokens, memory ka illusion, reasoning, tools aur loops — aur agent ki precise tareef.
Context engineering →
System prompts, context window, compacting, aur intehai aham agents.md file.
Workflows & model ka intekhab →
Micromanagement se Ralph Loops aur swarms tak — sath artificialanalysis.ai samajhna.
Baatein kam, kaam zyada
Tools ka tour karein, phir apne sath AI agent ke zariye aik real app banayein aur usay improve karein.
Tools, setup & agents.md →
Cursor, Copilot, Codex, Antigravity — Node install karein, repo clone karein, aur real agents.md line by line samjhein.
Kanban app banayein →
Plan mode, YOLO mode, agent loop ko dekhna, aur targeted feedback se behtar banana.
Khud banayein →
Seedha hands-on lab par jayein aur apne hathon se build karein.
Model chunein — phir asal build karein
Real YOLO build ke sath jaan boojh kar rules torein, phir wapis aakar har line ko own karein.
Model ka intekhab →
Do faislay: tool vs model. Fast frontier vs top frontier — aur YOLO kab safe hai.
Be the boss →
Kamyab vibe coding ke 5 usool — spec, start simple, iterate, challenge, frustration handle karna.
Responsible YOLO →
Do counterpoints, “own the code” mantra, aur OpenRouter API key setup.
The YOLO build →
AI digital twin ke sath Next.js portfolio site ship karein — phir doosre model se review karein.
Karpathy & the MVP →
Vibe coding ko naam dene wale tweets, workflow shift, aur real build ke liye balanced mindset.
Claude Code — aur koi bhi model kaise chalayein
Is sab ko shuru karne wale tool ko install karein, CLI master karein, phir har free aur cheap alternative explore karein.
Claude Code ka urooj →
2024 ke side project se Opus 4.5 tak ki tareekh — sath pricing aur VS Code mein installation.
CLI: init, context & testing →
/login, /init aur /context — phir Claude se poora test suite chalwayein.
OpenCode & free models →
GLM 4.7 jaise free models chalayein, aur apni pasand ka provider connect karein.
AMP, OpenRouter & Ollama →
Ad-supported credits, Claude Code ko OpenRouter par point karna, aur locally chalne wale models.
Command line par control hasil karein — phir usay chalne dein
Rozmarrah ki commands, project mein mehfooz tareeqe se aage barhna, aur YOLO aur Ralph loops ke sath autonomous coding.
Commands, shortcuts & config →
Commands discover karein, three C’s (context/compact/clear), shortcuts, .claude folder aur @ trick.
Sessions, checkpoints & Git →
Project mein aage barhne ke 3 tareeqe — aur aisa workflow jo aapko safe rakhe.
Rewind & YOLO mode →
Real rewind chalayein, phir bina kisi jhijhak --dangerously-skip-permissions run karein.
Ralph Loops →
Agent ke gird loop lagayein taake woh musalsal build karta rahe — autopilot par autonomous coding.
MCP, Skills & Plugins — aik naya mor
Doosron ke tools connect karein, simple files ke zariye expertise add karein, sab bundle karein — aur wazeh taur par janein ke kab kya use karna hai.
Big three & tools →
MCP, Skills aur Plugins ka naqsha — jo tools ke bunyadi concept par mabni hai.
MCP under the hood →
Hosts, clients aur servers; local vs remote; aur qabil-e-aitmad server dhoondna.
MCP servers add karna →
Claude Code ko Context7 docs aur live market data se equip karein — phir unhein remove karein.
Skills — aik asan tareeqa →
Progressive disclosure ke sath markdown instructions — light, simple aur shareable.
Skills marketplaces & Agent Browser →
Skills dhoondein, phir Claude Code ko apne computer par real browser chalate dekhein.
Plugins & aapka setup →
Marketplaces ke zariye sab kuch bundle karein — aur MCP vs Skills vs Plugins ka usool.
Jira ticket se merged PR tak
Jira aur GitHub ko Claude Code se jorein, FeatureDev process shuru se aakhir tak chalayein, aur pro ki tarah debug karna seekhein.
Workflow & Jira setup →
Slides chhoriye — do din ka project, free Jira board, aur pehla ticket set up karein.
Jira connect karna (MCP) →
Atlassian MCP server add karein, OAuth se login karein, aur re-auth issue solve karein.
GitHub repo, token & MCP →
Least-privilege token aur GitHub MCP server — phir issue → PR → merge loop.
FeatureDev & 7 stages →
Anthropic ka plugin jo Claude Code ko aik disciplined, staged build ke zariye le jata hai.
Autonomy: Jira to PR →
Do tickets end-to-end — dataset curate karein, phir 76 tests ke sath Next.js NDA app banayein.
Debugging strategies →
Asal playbook: snapshot, copy-paste loop, debug.md tareeqa, aur pro tips.
Pre-Legal ship karein, ticket by ticket
Week 2 ka ikhtitam: aik multi-user AI legal-document platform, jo aapki custom skill ke sath end-to-end banaya gaya hai.
The SaaS blueprint →
Skills asal mein kaise kaam karti hain, aur do CLAUDE.md files — bashamool aik lean global file.
CLAUDE.md & custom skill →
Mazboot project CLAUDE.md likhein aur apni fast Cerebras skill create karein.
V1 tickets ki planning →
Product ko clean backlog tickets (PL-4 se PL-7) ke tor par stakeholder ki aawaz mein plan karein.
Features build karna →
Foundation aur AI chat ship karein — architecture review aur real design debate ke sath.
Cerebras ke sath testing →
Confirm karein ke skill use hui, user ki tarah test karein, aur har document type tak expand karein.
Multi-user, polish & victory →
Sign-in add karein, freemium feature, imandaranah review — aur week 2 ki kamyabi declare karein.
Finale se pehle consolidate karein
Koi naye tools nahi — aik wazeh mental map. Week 1 aur 2 ka nichor aur rabt.
Ab tak ka safar →
Eight stages, control↔autonomy dial, aur do mindsets jo sab ko jor kar rakhte hain.
Foundations recap →
LLMs asal mein kaise kaam karte hain, agent banane wali 4 tricks, context engineering aur CLAUDE.md.
The pro toolkit →
Claude Code CLI, models aur providers, safety net, aur big three: MCP, skills, plugins.
Workflow & cheat-sheet →
Jira→FeatureDev→PR loop, jo humne ship kiya, aur master cheat-sheet + self-check.
Sub-agents, hooks, swarms & controlled chaos
Week 3 ka aaghaz: real capstone project par multi-agent banein, hooks se automate karein, aur apna plugin ship karein.
Controlled chaos →
Finale ka mindset: chaos aur control ka balance, aur pro techniques ka map.
FiNALLY project →
Multi-agent trading capstone set up karein — plan.md, boundaries aur simple architecture.
Custom slash commands →
$ARGUMENTS ke sath apni /commands banayein — aur skills kaise free mein command deti hain.
Agents & sub-agents →
Isolated context ke sath sub-agents ko kaam dein — yahan tak ke Codex ko reviewer ke tor par chalayein.
Hooks & events →
Events par actions automatically fire karein — aik aisa stop-hook banayein jo aapke kaam ko review kare.
Plugins & marketplaces →
Commands, skills, agents aur hooks ko plugin mein bundle karein aur marketplace par share karein.
YOLO ko safe banayein: sandboxes & remote execution
Claude ke gird safety boundary lagayein taake woh bina kisi risk ke run kare — aapke laptop par, Anthropic ke cloud par, ya rented sandbox par.
Sandboxing & ahmiyat →
Sandboxing ka asal matlab, approval fatigue se nijat, aur 3 approaches ka map.
Native sandbox →
/sandbox ke zariye Claude Code mein built-in fast OS-level box — modes, demo, WSL point.
Web par Claude Code →
Managed cloud sandbox aur remotely chalane ke 5 tareeqe — bashamool @claude tagged GitHub issue.
Sprites.dev sandboxes →
Fast, isolated cloud box rent karein aur kisi bhi agent ko safe full-YOLO mein chalayein.
Recap & cheat-sheet →
Blue/purple/yellow map, decision table, har command, aur self-check.
Bare codebases, aur agents ki frontier
Bari team repos ke usool — phir agent frontier ka tour: Agent SDK, Cowork aur OpenClaw.
Recap & zero-shot demo →
Toolkit ko consolidate karein, phir live market-data demo jo pehli baar mein perfect chala.
Large team codebases →
Bari multi-person repo par coding agents use karne ke 7 rules taake kaam kharab na ho.
The Agent SDK →
Apne code se Claude Code ko drive karein — aik chhota Python script working game banata hai.
Claude Cowork →
Non-coders ke liye agentic AI: receipts ke folder ko formatted expenses spreadsheet mein tabdeel karein.
OpenClaw sidekick →
Aapke computer par personal AI, jo Telegram se chalta hai — powerful, hyped, aur waqai risky.
Aasteenein charha lein
Do guided, self-paced labs. Koi grading nahi, koi pressure nahi — sirf practice. Dono free tiers par kaam karti hain.
Aik prompt, aik app →
Aik sentence se working app generate karein, follow-ups se behtar banayein, phir wohi prompt dobara chala kar divergence dekhein. Non-determinism ka pehla tajruba.
agents.md likhein, YOLO mein build karein →
Real agents.md likhein, starter clone karein, aur Cursor ko plan → build → test → feedback ke zariye drag-and-drop Kanban board banane ke liye chalayein.
✓ In classes ke ikhtitam par aap…
- Aik single prompt se working app generate kar sakte hain aur conversation ke zariye usay behtar bana sakte hain.
- Wazeh kar sakte hain ke LLM kya kar raha hai — tokens, memory, reasoning, tools, loops — aur "agent" ki precise tareef bayan kar sakte hain.
- Aik tight aur mo-assar
agents.mdlikh sakte hain aur samajh sakte hain ke context window results ko kaise shape karta hai. - Tools aur models ke darmiyan soch samajh kar chunao kar sakte hain, aur jaan sakte hain ke YOLO kab safe hai.
- Real applications ko build, review aur debug karne ke liye AI-native IDE (Cursor) ko plan aur YOLO modes mein drive kar sakte hain.
- OpenRouter ke zariye AI digital twin ke sath portfolio site ship kar sakte hain — aur uski har line ko own kar sakte hain.
- Terminal se Claude Code install aur drive kar sakte hain — aur OpenCode, AMP, OpenRouter aur Ollama ke zariye free ya cheap models chala sakte hain.
- Rozmarrah ki commands, shortcut keys aur **sessions, checkpoints aur Git** ke safety net par command hasil kar sakte hain — aur YOLO aur Ralph loops ko zimmedari se chala sakte hain.
- Claude Code ko MCP servers, skills aur plugins ke sath supercharge kar sakte hain — aur wazeh taur par jaan sakte hain ke kab kis cheez ki zaroorat hai.
- Poore course ko aik mental model mein consolidate kar sakte hain — eight stages aur control↔autonomy dial.
- Claude Code ke sath multi-agent setup chala sakte hain: custom commands, sub-agents, hooks, aur aapka apna plugin + marketplace.
Yahan har cheez beginners aur tajrubakar engineers dono ke liye mufeed hai. Agar koi section aapke liye aam hai to 2× speed par skim karein — har kisi ke liye kuch na kuch zaroor hai. Light aur dark mode switch karne ke liye oopar ☾ button use karein, aur navigation ke liye left sidebar use karein.