I Let AI Replace Me for a Week as a (Kinda Junior) AI Engineer 😅

Sujeeth Sundarajan Rajkumar — Wed, 27 May 2026 03:05:58 +0000

I recently tried using only ChatGPT + Claude for most of my day-to-day work as someone building RAG/chatbot systems, APIs, debugging random issues, and doing way too much prompt engineering.

Short answer?

AI did not replace me.

But it definitely made me faster.

Also… it occasionally made me question reality.

Why I Even Tried This

Everywhere online, I keep seeing:

“AI will replace engineers.”

And then the other side saying:

“AI is useless.”

Meanwhile, I’m sitting there building chatbot/RAG projects thinking:

“Okay but… what actually happens if I seriously try this?”

So for a week, I decided to do something simple:

Before doing anything myself, ask AI first.

Not blindly copy-paste.

Just genuinely try using it as much as possible.

My setup was pretty simple:

ChatGPT
Claude

No fancy agent workflow.

No 12-monitor productivity setup.

Just me, AI, and a questionable amount of debugging.

My Rules

I gave AI permission to help with:

✅ RAG / chatbot development

✅ API/backend coding

✅ Debugging

✅ Prompt engineering

✅ Documentation

✅ Resume/portfolio improvements

✅ Random “why is this not working” moments

But I refused to let AI do:

❌ Production decisions

❌ Blind copy-pasting into production

❌ Anything security-related without checking

❌ Thinking for me

That last one became important later.

Very important.

Day 1 — This Felt Illegal 🚀

I’m not gonna lie.

Day 1 felt amazing.

I started throwing everything into ChatGPT and Claude.

Things like:

API errors
FastAPI backend stuff
Weird JSON issues
Prompt engineering ideas
RAG architecture questions
Documentation

And suddenly…

Stuff that normally took me 30–40 minutes sometimes got solved in 5 minutes.

It honestly felt like:

pair programming with someone ridiculously smart… who occasionally forgets reality.

Or like:

having an intern who somehow read the entire internet overnight but still needs supervision.

For example:

I had one annoying API issue that would've normally taken me forever to debug.

AI didn't instantly solve it.

But it pointed me toward the exact place I was messing up.

That alone saved me a lot of pain.

Day 2–3 — Productivity Mode Activated

This was where I started understanding why engineers are obsessed with AI tools.

1. Debugging Became Less Painful

Not perfect.

But definitely better.

Normally my debugging process looks like this:

Google
Stack Overflow
Random GitHub issue
YouTube video
Confusion
Existential crisis
Finally fix bug

With ChatGPT + Claude it became:

Paste error
Get possible causes
Test solutions
Fix faster
Still confused but slightly happier

Huge improvement.

Especially for weird backend issues.

2. Prompt Engineering Was Weirdly Fun

Since I’ve been working on chatbot/RAG stuff recently, I started using AI to improve prompts.

Instead of asking:

“Write a prompt.”

I started asking:

“Why is this prompt failing?”

“How would you redesign this for better responses?”

This helped way more than I expected.

Turns out:

Good prompting is less about magic words and more about clarity.

Which sounds obvious…

But I definitely learned it the hard way 😅

3. Documentation Became 10x Easier

I’ll be honest.

I hate writing documentation.

Probably more than debugging.

Before AI:

“I’ll write docs later.”

Later never came.

After AI:

I would dump messy notes into ChatGPT and say:

“Make this readable.”

And suddenly I had something decent.

Honestly one of my favorite use cases.

4. Resume & Portfolio Stuff

This one surprised me.

I recently spent time updating my portfolio/resume, and AI became weirdly helpful.

Not for lying.

Not for buzzwords.

But for:

making descriptions cleaner
improving wording
removing awkward sentences
making projects sound more readable

Because apparently:

“Built cool AI thing”

is not recruiter-friendly language.

Who knew 😭

Day 4 — Reality Hit Hard 😅

This is where things got interesting.

I gave AI a more complicated engineering problem related to backend logic.

At first glance?

Everything looked great.

The explanation sounded smart.

The code looked clean.

The confidence level?

10/10

The actual solution?

More like…

4/10

After testing it properly:

The logic was wrong.

Not obviously broken.

Which is actually worse.

Because bad code that crashes is easier to catch.

Bad code that looks correct?

That’s dangerous.

This was my biggest realization:

AI is dangerously useful.

Useful enough that you trust it.

Dangerous enough that you absolutely shouldn’t trust it blindly.

Day 5–6 — I Changed How I Used AI

This was the biggest mindset shift for me.

At first I treated AI like:

“Do this work for me.”

That worked sometimes.

But better results came when I switched to:

“Think through this with me.”

Instead of:

“Build this.”

I started asking:

“What edge cases am I missing?”

“Why might this architecture fail?”

“What are better alternatives?”

“Challenge my approach.”

And weirdly…

The responses got much better.

This stopped feeling like replacement.

And started feeling like:

having an extremely fast collaborator.

One who occasionally hallucinates.

But still useful 😅

What AI Was Surprisingly Good At

Debugging

Honestly?

Probably my favorite use case.

Not always right.

But often good enough to save me from wasting an hour.

Documentation

Massive time saver.

No debate.

Brainstorming

Especially for:

RAG ideas
chatbot flow logic
architecture thinking
API approaches

Sometimes I didn’t even need the answer.

Just someone (or something?) to think through the problem with.

Explaining Things

Way faster than opening:

15 Stack Overflow tabs
random Medium articles
confusing docs

What AI Was Bad At

Complex Logic

The harder the problem got…

The more careful I had to be.

Real Context

AI doesn’t know:

your codebase
business requirements
weird production problems
hidden dependencies

And that matters more than people think.

Being Confidently Wrong 😭

This one hurts.

Because sometimes AI sounds SO convincing.

And then after 40 minutes you realize:

“Wait… this makes absolutely no sense.”

The Biggest Thing I Learned

At the start of the week, I thought the question was:

Can AI replace me?

By the end, I realized the better question is:

Can engineers using AI move faster than engineers ignoring it?

And honestly?

I think the answer is:

Probably yes.

Not because AI replaces engineers.

But because it removes a lot of repetitive frustration.

Especially if you’re still learning (like me).

Final Verdict

Could ChatGPT + Claude replace me?

No.

Could they make me faster?

100%.

My honest takeaway:

AI feels less like replacement and more like a really smart collaborator who occasionally says nonsense with extreme confidence.

And learning how to work with AI feels like a skill worth developing early.

Especially if you're building things in AI already.

Curious About Your Experience

If you're a developer or engineer:

Has AI actually improved your workflow?

Or does it just create more problems?

Genuinely curious.

Drop your experience below 👇

DEV Community: Sujeeth Sundarajan Rajkumar