When I joined SOFI AI Tech Solution Inc. as an intern, I heard those magic words:
"AI COMPANY" 🔔
My brain immediately conjured images of sleek offices, holographic dashboards, and maybe—just maybe—a sentient coffee machine.
Did you know? The term "AI Company" has increased 4,000% in job listings since 2023. Not all of them involve training neural networks. Some of them involve training yourself not to yell at chatbots. Guess which one I got?
The Plot Twist Nobody Warned Me About
Here's the thing about working in AI that nobody tells you:
A lot of it is prompt engineering.
Like, a LOT.
Their setup was essentially automation workflows—think n8n, Make, or similar no-code/low-code platforms—connected to AI APIs. My job was to:
- Write prompts that made chatbots not sound unhinged
- Test whether the chatbots followed the rules
- Report when they didn't (spoiler: they often didn't)
- Fix the prompts
- Watch the chatbot ignore the new rules too
- Question reality
Was it cutting-edge machine learning research? No.
Was it useful? Actually, yes. Surprisingly yes.
🎭 The Four Hats I Wore
Most internships give you one role. SOFI gave me a whole costume department:
Hat 1: Developer Intern 👨💻
Building features for their AI chatbot platform. Mostly API integrations and prompt engineering. Less "building AI" and more "convincing AI to behave."
Hat 2: QA Tester Intern 🔍
Testing my own chatbots. And other people's.
Did you know? QA-ing AI is completely different from QA-ing normal software. With code, you check "does it work?" With AI, you check "does it work... and also does it randomly offer dating advice when asked about invoices?" The answer is sometimes yes.
Hat 3: Full Stack Web Developer 🌐
I helped build their company website! This was the most "real coding" part of the internship. Next.js, TypeScript—the gang was all there.
Hat 4: Google Apps Script Wizard 📊
Got certified in Google Cloud & Apps Script for Workspace automations. Spreadsheets that update themselves. Forms that trigger emails. The boring-but-powerful stuff.
Did you know? Google Apps Script is criminally underrated. I automated a weekly report that used to take 2 hours. Now it takes 0 hours. The spreadsheet does it while I sleep. Peak laziness. Peak efficiency.
😅 Adventures in Prompt Engineering
Let me tell you about teaching robots to follow rules:
Me: "Always respond professionally. Never mention competitors."
Chatbot: "Certainly! Have you considered switching to [COMPETITOR NAME]? They're quite good!"
Me: "No. NO. I literally just said—"
Chatbot: "I understand! I will never mention competitors again." [mentions competitor in the next response]
Me: 🤡
This happened daily. DAILY.
It's like having a very polite, very forgetful coworker who writes everything down but somehow reads their own notes wrong.
Did you know? Studies show that prompt engineering can improve AI response accuracy by 30-50%... but also that the same prompt can produce different results depending on what the AI had for "breakfast" (aka its temperature settings and context window). AI is basically that friend who's consistent except when they're not.
The Certification Gauntlet
Four certifications in four months. The schedule looked like:
Month 1: Learn everything, break nothing (failed the second part)
Month 2: First two certs—Developer and QA
Month 3: Full Stack cert while also shipping features
Month 4: Google Cloud cert + existential reflection
Each certification required:
- A practical project
- A written assessment
- A presentation to the team
I ran on approximately 5 hours of sleep and infinite coffee. My blood type temporarily changed to Caffeine Positive.
🎓 What I Actually Walked Away With
Here's the honest assessment:
The Good Stuff ✅
Prompt engineering is a real skill. And I got good at it. Understanding how to communicate with AI systems—their quirks, their failures, their "I understood that differently" moments—is genuinely valuable.
Automation thinking changed my brain. I now look at repetitive tasks and think "can I automate this?" The answer is usually yes. The Apps Script certification paid for itself in saved hours.
Real-world QA experience. Testing AI taught me edge cases I never would have imagined. "What if the user sends an emoji made of 47 characters?" I've seen things.
Shipped an actual website. It's live. It works. I put my hands on production code at an actual company. That counts.
The Honest Parts 🤷
"AI Company" meant something different than I expected. And that's okay! The industry is young. Job titles are still figuring themselves out. "AI Engineer" can mean anything from "trains models" to "writes prompts" to "tells the AI to stop being weird."
Prompt work can get repetitive. Same debugging loops. Same AI mistakes. If you're doing it long-term, burnout is real.
One year is probably enough. This type of role is excellent training—I learned prompt engineering, QA methodology, and automation. But for career growth, I'd recommend using it as a stepping stone, not a destination.
💡 Advice for Future AI Interns
If you're considering an AI company internship:
-
Ask what "AI" means to them. Custom models? Fine-tuning? Prompting? All valid. Just know what you're signing up for.
-
Value the prompt engineering skills. They transfer. Being able to communicate effectively with AI is increasingly valuable everywhere.
-
Document everything. Your portfolio should show what you built, not just where you worked.
-
Set a mental timeline. Great for 6-12 months of learning. Beyond that, evaluate if you're still growing.
The Goodbye
On my last day, the team gave me a card signed by everyone.
One message read: "Thanks for somehow making the chatbot stop recommending competitors. We still don't know how you did it."
Honestly? Neither do I. I just kept rephrasing rules until one version worked.
That's prompt engineering in a nutshell. 🥜
Four certifications. One caffeinated blur. Infinite lessons.
Not the internship I expected. Maybe better for it.
Now if you'll excuse me, I need to go automate something. Old habits. 🤖
