AI Careers and Skills to Learn

AI has created new jobs and changed old ones. The good news: the most in-demand skills can be learned step by step, and many are covered in this book.

Three Paths

The first path is the fastest way into a technical AI job for most students in Pakistan. It does not require a PhD. The second path usually needs a strong university background. The third path applies to almost everyone. For more job titles, see Careers in AI.

The Skills That Matter

These skills come up again and again in AI builder job posts:

1. Programming: solid Python or JavaScript, and Git
2. Calling AI APIs: messages, settings, streaming, errors, costs (Part 2 of this book)
3. Prompting: clear instructions and examples (Prompt Engineering)
4. Structured output: getting and validating JSON
5. RAG and vector search: embeddings, chunking, retrieval quality
6. Tools and agents: function calling, MCP, safe agent loops
7. Evaluation: test sets, scoring, catching regressions
8. Web basics and deployment: a simple backend, a database, putting an app online
9. Safety and privacy: prompt injection, data protection, human approval

Soft skills matter as much: explaining technical things simply to clients, writing clearly, and saying "I am not sure, let me check".

A Realistic Learning Path

Track your progress honestly. A simple script can keep you on track:

plan = {
    "Python basics": 100,
    "Calling an AI API": 100,
    "RAG project": 60,
    "Evaluation tests": 20,
    "Deploy and share on GitHub": 0,
}
for skill, done in plan.items():
    bar = "#" * (done // 10) + "." * (10 - done // 10)
    print(f"{skill:<28} [{bar}] {done}%")
print(f"Overall: {sum(plan.values()) // len(plan)}%")

Python basics                [##########] 100%
Calling an AI API            [##########] 100%
RAG project                  [######....] 60%
Evaluation tests             [##........] 20%
Deploy and share on GitHub   [..........] 0%
Overall: 56%

Portfolio Projects That Impress

Employers and clients want proof, not certificates alone. Good projects solve a real local problem:

• An FAQ bot for a real shop, clinic or institute, built on their own documents, with their permission
• A tool that turns WhatsApp order messages into a clean spreadsheet
• An Urdu and English study helper that quizzes students from their textbook chapter
• An MCP server that connects an AI app to a small inventory database
• An eval report: "I tested my bot on 50 questions; the score went from 31 to 46 after these changes"

Write a short README for each project. A good README answers what the project does, how to run it, what you learned, and what its limits are:

# Shop FAQ Bot

Answers customer questions for a Karachi stationery shop using RAG.

Run:      pip install -r requirements.txt, then python app.py
Stack:    Python, an OpenAI-compatible API, Chroma
Tests:    50 eval questions, current score 46/50
Limits:   Does not handle refunds; sends those to a human.

Will AI Take the Jobs?

AI changes tasks faster than it removes whole jobs. Routine work (simple copywriting, basic data entry, first-level support) is shrinking. Work that combines a skill, judgment and responsibility is growing. The safest plan is to be the person in your field who uses AI well and safely.

Exercise