AI Powered Engineer

Talk Information

Title: AI Powered Engineer

Speaker: Ms. Midha Tahir - Advisor SDE, BCT Consultants | IT Staff, City Educators

Duration: 45-60 minutes (including Q&A)

Audience: University students interested in tech careers and practical AI usage

🎯 Quick Intro

Hi, I'm Midha Tahir, a Software Engineer graduated from NED University in 2022. I work as an Advisor SDE at BCT Consultants on weekdays, IT Staff at City Educators on weekends and as a chef on night.

A year ago, I was a frontend engineer building interfaces. Today, I'm leading a team of around 7 engineers. From the past years I know I have to grow but I am just waiting for other things to work out like moving to abroad. This shift taught me something important: the biggest time wasters in tech aren't coding problems, they're repetitive tasks, unclear processes, and communication gaps.

AI became my productivity tool to handle repetitive tasks, so I could focus on the hard problems that actually need my attention.

📖 Full Personal Story

Let me start with a question: How many of you spend time doing tasks you know could be automated

That was me. For years, I worked as a frontend engineer-building React components, fixing UI bugs, making things look good. I was good at it. But I noticed something frustrating:

• I'd build a feature, but deployment would take forever because nobody had documented the steps properly.
• I'd review pull requests, but I was reading the same types of bugs over and over.
• I'd join meetings where people spent 20 minutes debating something that could've been solved with a clear process.
• I’d onboard to a project, but there was no README, no architecture overview, and no “how to run locally” guide.
• I’d fix a bug, but there was no logging or monitoring, so I had to reproduce it blindly every time.

From the start of my career, I take ownership of my work, but seeing the above problems, I didn't just want to build UI. I wanted to own the entire outcome-architecture, quality, delivery, and helping teams move faster.

After coming to London, I have seen people working on bus, on trains utilizing 100% of their times, our head of engineering starting to tell me that I work like I am on CD70.

How AI Fits Into My Work

My approach is simple:

1. Identify repetitive work - Tasks I do over and over that follow a pattern.
2. Draft a standard process - Document what "good" looks like.
3. Use AI for speed and structure - Let AI generate the boilerplate, then I validate and improve it.
4. Use infrastructure as code - Using terraform to make and deploy infrastructure as code

Real Examples from My Day-to-Day:

• Azure deployments: I used AI to help draft deployment pipelines and standardise the process. Then I validated and improved it myself. Now, deployments that used to take hours of manual work are consistent and documented.
• Diagrams: We first create the infrastructure on mind or paper then take help from AI to generate diagrams.
• PR reviews: We use AI to summarise code diffs, spot common issues, and suggest test cases. But I still make the final judgement. The AI speeds up the boring part; I own the decisions.
• Time of Delivery: The time we were quoting to the client was alot and nowadays clients are very intelligent they know we have AI coding tools, so we can't quote long time for the projects, we need to use AI at any cost to reduce the time and get the project done on time.

Real Team Example: The Localization Story

Let me share a recent example that shows exactly how this works in practice.

One of my colleagues said she wouldn't use Cursor or AI tools. She preferred doing things manually in order to grow her skills.

Then she worked on a repetitive task: adding Localization (translation support) to our application. She did it manually, section by section. After about a full day of work, she had completed only one page.

I observed her work and I really need to deliver more on that sprint, so the next day, we pair programmed together. I showed her how the same work could be done in a few hours using AI properly. She already knew the process - that wasn't the issue. The difference was using AI as a multiplier and she could focus on the hard problems that actually need her attention.

My Advice for This Situation:

1. Do 2-3 sections yourself first to learn and confirm the pattern. This is critical, research about libraries, processes, you can use chatgpt research feature, ask questions like which library to use, which one is latest and stable, which one consumes less memory etc. You have to do this research this is your part. After this you can either create a codesandbox on how to use the library or make a logic
2. Then rely on AI to replicate the pattern across the rest of the sections. It's following the structure you've already defined.
3. Still review carefully and keep ownership. AI speeds up the execution, but you validate the output. You need to use git diff to see the changes and you need to use git commit to commit the changes page by page.

This approach turns a full day of repetitive work into a few hours - not because AI is magic, but because you're using it strategically after understanding the problem yourself.

The City Educators Story: Where AI Doesn't Solve the Problem

Let me tell you about a real challenge I'm working on right now at City Educators.

They have a student attendance system. It already exists. It works. But the data is wrong.

When they pull a report of "absent students," it includes people who left years ago-some records go back to 2011. There's no proper check to distinguish between active students and students who've moved on. Who's sitting with his regular teacher and who's not. This creates confusion, wrong reporting, and no track

What will you do in this situation:

• The stakeholders are non-technical. They don't think in terms of databases or logic-they think in terms of "I need to know who's actually here today, who's sitting with his regular teacher and who's not."
• The real challenge isn't technical. It's human: understanding their actual workflow, aligning multiple people, explaining the problem clearly, and designing a process they will actually adopt.
• AI can help me build the solution faster once I know what to build. But it can't tell me what the right solution is. That requires conversations, empathy, and understanding context.

Leadership Under Deadline: The 2-Week Project

Here's another example that shows how AI fits into real project delivery.

I was given a project to complete in 2 weeks. The task: convert an existing app into a web app and make additional frontend changes. I had a small team of 2-3 people working with me.

The deadline was very short. Many people would panic and start coding immediately. Instead, I spent the first 2 days carefully analyzing the codebase and writing a Cursor rules file a set of project-specific AI guidelines.

What's a Cursor Rules File?

It's a document that tells AI tools like Cursor/Copilot:

• Our coding standards and patterns
• Component structure we follow
• What to avoid (anti-patterns in our codebase)
• Project-specific context and constraints

Think of it as "training" the AI on your team's standards before it helps you code.

Bonus: MCP (Model Context Protocol)

I also used MCP integrations to give AI direct access to tools and data:

• GitHub MCP: AI could read repo, check PRs, and understand project structure
• ShadCN MCP: AI had access to component library patterns and best practices
• Figma MCP: AI could reference designs directly for accurate implementation

This meant AI wasn't just following rules - it had real context about our tools, designs, and codebase.

With that rules file and the structured approach, the team completed the project in 1.5 weeks - half a week ahead of schedule.

Key Lesson:

Speed comes from clarity + standards + good AI usage, not from rushing blindly.

The 2 days spent on analysis and setup saved us much more time during execution. The team had clear guidelines, AI had proper context, and we avoided inconsistent code that would need refactoring later.

🎓 3 Memorable Takeaways for Students

1

Don't Use AI to Skip Learning-Use It to Remove Boring Work

AI won't make you a better engineer if you don't understand the fundamentals. But once you know what good code looks like, AI can help you write it faster.

Bad use: "Write me a sorting algorithm" (when you don't know how sorting works).

Good use: "Generate boilerplate test cases for this function I wrote" (because you already know what to test).

2

Identify Your Time-Wasters, Then Systematically Eliminate Them

Ask yourself: "What tasks do I do repeatedly that follow the same pattern?" Then build a process, document it, and automate it.

Examples:

• Research for a project → Identify a problem in the surroundings and use AI to brainstorm solutions
• Summarizing notes from classes → Use AI to summarize the notes, generate flashcards that you could read while travelling
• Generate diagrams for lessons → Use AI to generate diagrams because its faster to learn from visual content
• Writing the same email templates → Create a library
• Manually checking code diffs → Use AI-assisted PR reviews

The goal isn't to be lazy, it's to free up your brain for harder problems.

3

The Hard Problems Are Always Human Problems

Technology is the easy part. The hard part is understanding what people actually need, aligning stakeholders, designing solutions they'll adopt, aligning things according to the budget and resources of the project.

AI can't do that for you. But if you get good at understanding people and processes, AI becomes an incredibly powerful tool to execute faster.

Bottom line: Learn to think like a lead, not just a coder.

✅ ❌ Where AI Helps vs Where It Doesn't

✅ HELPS

Generating boilerplate code: Repetitive structure you already understand.

❌ DOESN'T

Teaching you how to code: You still need to learn fundamentals yourself, you can take help from AI to learn the fundamentals but in order to keep it in your memory you need to write it down yourself.

✅ HELPS

Summarising large code diffs: Spotting patterns and common issues quickly.

❌ DOESN'T

Making architectural decisions: You need context, judgement, and experience.

✅ HELPS

Drafting deployment pipelines: Once you've defined the steps, AI speeds up writing config files.

❌ DOESN'T

Understanding stakeholder needs: This requires real conversations and empathy.

✅ HELPS

Creating bots for repetitive workflows: Following rules you've clearly defined.

❌ DOESN'T

Designing processes people will actually use: This is about human behaviour, not code.

💼 Practical Tips for Students

How to Start Using AI Effectively Today

Identify one repetitive task you do every week (writing similar code, formatting reports, etc.) Document the process in plain English before asking AI to help Use AI to generate a first draft, then validate and improve it yourself Never copy-paste AI code blindly-always understand what it does Track your time savings-see where AI genuinely helps vs where it wastes time

Tasks That Drain Energy (Where AI Can Help)

• PR reviews: Use AI to summarise changes and spot patterns
• Making slides: Use AI to generate initial structure and bullet points
• Manual admin tasks: Build simple automation scripts
• Writing documentation: Use AI for structure, but add your insights

Skills That Matter More Than Ever

• Critical thinking: Knowing when AI is right vs when it's nonsense
• Communication: Understanding what stakeholders actually need
• Process design: Creating workflows that people will follow
• Judgement: Making decisions with incomplete information

🎯 The 1-Slide Framework You Can Teach

"Most bad results happen because people give AI a sentence instead of giving it a brief."

Simple Prompt Formula:

Prompt = Context + Task + Constraints + Output format

1

Context

What you're doing + who the audience is

Example: "I'm building a React app for university students to track their assignments."

2

Task

What you actually want

Example: "Write a function that calculates how many days are left until a deadline."

3

Constraints

Tech stack, tone, length, what not to do

Example: "Use JavaScript. Keep it under 20 lines. Don't use external libraries. Don't invent features I didn't ask for."

4

Output Format

How you want the result structured

Example: "Give me the code in a single code block with comments explaining each step."

🔑 Why This Works

When you give AI a complete brief instead of a vague sentence, you get better results faster. It's the same principle as briefing a colleague - the clearer you are, the better the outcome.

Remember: AI is a tool that follows instructions. Good instructions = good results.

✅ The Most Important Part: Verification

🔑 Critical Insight

"Prompting is easy. Verification is professional."

Anyone can ask AI for code. What separates a professional from someone who's just copy-pasting is verification. This is how you avoid becoming lazy and stay in control.

1

Explain It Back in Your Own Words

If you can't explain what the AI-generated code does, you don't understand it well enough to use it.

Try this: Read the code, then explain the logic to yourself (or a colleague) without looking at it. If you get stuck, you need to study it more.

2

Test It / Run It

Never deploy AI-generated code without testing it yourself. Run it locally, check the output, make sure it works as expected.

Ask yourself: Does it handle the happy path? What about errors? Does it match the requirements?

3

Check Edge Cases + Security

AI often gives you the "happy path" solution but misses edge cases, security issues, or performance problems.

Common issues to check:

• What happens with empty inputs?
• What if the user enters invalid data?
• Are there SQL injection risks?
• Is sensitive data being logged?
• Will this scale with more users?

4

Review Its Changes Through Git

Use version control to see exactly what changed. Review the diff before committing.

This helps you:

• Spot unintended changes AI might have made
• Understand the full impact of the code
• Write meaningful commit messages
• Have a clean history you can explain to your team

This Stops the "AI Made Me Lazy" Fear

When you verify everything AI gives you, you're not being lazy - you're being efficient. You're using AI as a tool to go faster, but you're still thinking, checking, and making decisions.

That's what makes you valuable. Not the code itself, but your judgement about whether it's good code.

🎯 Final Thought: AI-Assisted Engineering vs "Vibe Coding"

Let me leave you with an important distinction that will shape how you use AI in your career.

1

AI-Assisted Engineering (Professional Approach)

As a Software Engineer, you use AI as a smart assistant:

• AI generates boilerplate, tests, and structure for speed
• You maintain control and understanding of all code
• You ensure production readiness through verification
• You own the decisions and the quality

Best for: Production systems, team projects, code you'll maintain long-term

2

"Vibe Coding" (Exploratory Approach)

Letting AI generate most code with high-level prompts:

• Accept AI output rapidly for prototypes or learning
• Often without deep review - prioritizing flow over rigor
• Great for exploration, experimentation, and quick demos
• Risky for complex, stable, production systems

Best for: Prototypes, learning new tech, hackathons, personal experiments

🔑 Know When to Use Each

Both approaches have their place. The professional knows when to be rigorous (AI-assisted engineering) and when to move fast (vibe coding).

When building production systems, working with a team, or creating code that matters - be an AI-assisted engineer. Maintain control, verify everything, and own the quality.

When exploring, prototyping, or learning something new - vibe code freely. Move fast, break things, and iterate quickly.

The key is knowing which mode you're in, and never confusing the two.

⚠️ The Reality Check: Who Gets Replaced?

AI won't replace "engineers." It will replace tasks - and it will shrink the demand for engineers who only do low-judgement, repeatable tasks.

Most Replaceable (High Risk)

• People who mostly copy/paste patterns without understanding - CRUD pages, basic localization, simple UI wiring. If AI can generate it in seconds, why hire someone who just types it?
• People who can't debug, can't explain tradeoffs, can't own failures - When things break (and they will), you need someone who understands what happened and how to fix it.
• People who rely on one narrow tool and avoid fundamentals - If your only skill is "I know this one framework," you're vulnerable. Tools change; fundamentals don't.

Least Replaceable (Low Risk)

• Engineers who own outcomes - Turn messy goals into working systems. This requires understanding business context, not just writing code.
• People strong in debugging + production - Logs, monitoring, incidents. When production breaks at 2 AM, AI can't fix it alone.
• Engineers who can design, not just implement - Architecture, performance, security. These require judgment, experience, and tradeoff analysis.
• People who can work with non-technical stakeholders and drive adoption - Building the right thing building the thing right.

Frontend Isn't the Enemy - "Low Judgement" Is

There's a misconception that "frontend will be replaced by AI." That's wrong. Low-judgement work will be replaced, regardless of stack.

Frontend can be very high value when it includes:

• Performance: Core Web Vitals, bundle strategy, rendering patterns
• Accessibility: Design systems, UX consistency
• State management complexity: Offline/real-time, reliability
• Security: Auth flows, XSS, CSP, observability, error handling
• Product thinking: What actually improves conversion/retention?

So the split isn't "frontend vs backend" - it's:

Repetitive Implementer vs Outcome Owner

The Bottom Line:

Just like calculators didn't kill math, AI won't kill engineering. It kills low-leverage work.

The winners are the people who move up the value chain: from typing code to owning decisions.

🚀 What Companies Will Need in 2-3 Years

In the next 2-3 years, most companies won't mainly need "people who can code fast." They'll need people who can ship reliably, safely, and cheaply - with AI in the workflow.

What Companies Will Need from Engineers

1

AI-assisted delivery (with ownership)

Use AI to speed up dev, but still understand, test, and maintain what ships.

2

Strong fundamentals

Debugging, data structures basics, networking basics, databases, clean code - because AI outputs still break.

3

System thinking

How components connect (frontend, backend, DB, queues, caching), and how changes affect production.

4

Production readiness

Logging, monitoring, error handling, performance, incident response (basic SRE mindset).

5

Security by default

Secrets, auth, permissions, OWASP basics, supply-chain security (dependencies).

6

Cloud + cost awareness

Deploy confidently (CI/CD), manage environments, and avoid wasting money.

7

Automation mindset

Remove repetitive work (scripts, pipelines, internal tools), not just "do tasks."

8

Communication & documentation

Explain decisions, write clear PRs, create runbooks, align with non-technical stakeholders.

🎯 The Practical Takeaway for Students

If you want to be "future-proof," aim to become:

A developer who can build features + deploy + observe + iterate - using AI responsibly.

It's not about being the fastest coder. It's about being someone who can own the entire delivery process, make smart decisions, and ship value consistently.

💙 Thank You

To NSC Community

I want to take a moment to thank the NSC Community, which we founded back in 2019 with Wania Rajpoot.

We helped a lot of students back then - organizing sessions, sharing knowledge, and building a supportive tech community. But as life got busy, I had to step back.

I deeply thank Wania Rajpoot for still working on her vision and keeping the community alive. Your dedication and consistency are inspiring, and I'm grateful for everything you continue to do for students.