DEV Community

Isaac Hagoel
Isaac Hagoel

Posted on

I’m a 20-Year Engineer – AI Coding Tools Are My New Oxygen (But They’re Toxic If You Breathe Too Deep)

Let’s be clear: I'm totally hooked on coding with AI and not looking back. Here’s how to use it like a stimulant, not a crutch

AI coding assistants aren’t just helpful – they’re a cognitive exoskeleton. I code 3x faster with Cursor, ship prototypes in hours instead of days, and offload the work that makes my brain feel like overcooked ramen. But here’s the catch: every line of AI code is a Faustian bargain.

Why I’m Addicted (And You Should Be Too)

AI isn’t “helpful” – it’s a force multiplier for:

  • Killing yak shaves (Bash scripts, config hell, regex, TypeScript puzzlers, SQL gymnastics)

  • Bypassing syntax paralysis (How do you format dates in Swift again? Why wouldn't this API cooperate?)

  • Prototyping at methamphetamine speed (Need a React table with client-side sorting? Generated in 8 seconds)

Last month, I prototyped a RAG tool for my AI agent with 87% AI-generated code (local vector DB, embeddings, ranking – the whole circus). 2 hours instead of 2 days. Was it perfect? Hell no. But it let me start iterating immediately. Now that we're shipping? I’m rewriting 87% manually – with AI as my WD-40 for stubborn code bolts.


The Tech Debt Paradox

AI-generated code isn’t “bad” – it’s strategically irresponsible.

I treat it like a high-interest credit card:

  • SWIPE FREELY for:

    • Throwaway code (one-off scripts, spike solutions)
    • Boilerplate (mocks, DTOs, CRUD skeletons)
    • “I just need this to work once” moments (local DB migrations)
  • PAY DOWN IMMEDIATELY when:

    • It touches business logic
    • Performance/security/privacy matter (read: production)
    • You’re past exploration and need precision

Example: Last month Claude wrote 5 Jest mocks + 10 unit tests for an unfamiliar codebase. 45 minutes saved. 2 hours of existential crisis avoided. Tech debt? 15 minutes fixing incoherent tests. Net win! 🎉


How to Inhale Without Choking

My rules for AI-as-breathing-apparatus:

  1. AI First, Human Always

    Generate → Review → Rewrite/Question. I refactor AI code like it owes me money. Sometimes I reject 100% of its suggestions – but even its wrong answers spark better solutions.

  2. The Complexity Alarm

    AI loves Rube Goldberg solutions. Ask: “Could a junior understand this sober at 3AM?”

  3. Architecture is Human-Only

    AI sees files, not systems. Never let it decide service boundaries – it’s like letting a golden retriever plan city infrastructure.

  4. Rubber Duck 2.0

    “Explain tradeoffs between JWT session strategies” → Then decide. Treat it like a Wikipedia article – useful overview, questionable details.

  5. Soul-as-a-Service

    Let AI borrow your coding soul like a library book – but always check it back in. Late fees hurt.


For Juniors: How to Level Up Without Getting Played

AI won’t make you a better coder – you make you a better coder. Here’s how to avoid becoming an AI puppet:

  1. Practice Raw Coding Like It’s the Gym AI is your protein shake, not your workout. If 90% of your code is AI-generated, you’re building prompt engineering muscles (useful!) but atrophying actual coding skills (disastrous).
    • Mandatory manual reps: Code one full feature/week from scratch. No autocomplete, no Copilot.
    • Read AI’s output line-by-line like it’s your ex’s text messages – with skeptical intensity.
  2. You’re the Architect, AI’s the Hammer Treat AI like a power tool:
    • “Why did you use a HashMap here?” → Make it justify every choice
    • “Show me 3 alternative approaches” → Then pick none of them and write your own
    • If you couldn’t explain the code to a intern, you don’t understand it
  3. Build Your BS Detector AI’s wrong answers are golden learning opportunities:
    • When it suggests an overcomplicated design pattern, ask: “What problem is this solving?”
    • When it makes a subtle mistake (timezone handling, API rate limits), dig into why it failed
    • Every AI error is a free lesson in critical thinking
  4. The “Could I Build This Blindfolded?” Rule Only use AI for:
    • Tasks you already understand (boilerplate, mocks)
    • Explorations you’ll immediately validate (spike solutions) If you’re using AI to hide from learning fundamentals, you’re just stacking debt.

Why Seniors Survive the AI Apocalypse

Experience lets me:

  • Spot “confidently wrong” code like a bloodhound (missed timezones, security holes, 🤮 code duplication)

  • Surgically extract value (Keep AI’s 80% SQL JOIN, fix its 20% N+1 query disaster)

  • Embrace strategic debt (This script dies Friday anyway – let AI write its epitaph)


Final Take:

AI is coding’s cheat code – but only if you’ve already beaten the game. Use it to:

  • Supercharge trivial work (Your brain’s for hard problems)

  • Buy speed with intentional debt (Like taking a loan to close a deal)

  • Avoid context-switching (No more 15-tab Google spirals)

But never forget: AI code is radioactive. Handle it with lead gloves.

I’m now 73% cyborg. Juniors – hit me with your AI horror stories. Seniors – share your power moves. Let’s code. 👇

Top comments (5)

Collapse
 
spo0q profile image
spO0q • Edited

Very nice read.

Indeed, you still have to supervise it, but it removes the hassle of writing the same structures over and over, especially in the early stages of the project.

I would recommend developers to be extra vigilant with the context they provide. While there are some hallucinations, it's not AI's fault if you miss critical details.

Collapse
 
ivanrivera profile image
ivan-rivera-projects

Great read. Thanks for sharing. As a newbie to development, it shows that those who don't skip on the fundamentals will be the ones best poised to use the tools and likewise those who'll be able to stick around the longest.

Collapse
 
cbreezier profile image
Leo Huang

One of the few actually good (and very pragmatic) takes on how to use AI effectively.

Not much to add, but I find it most useful to treat it as a very knowledgeable coworker with zero critical thinking skills.

In one case, I had a subtle bug and I refuted a dozen suggestions that AI made (some were good guesses but not the problem, some were straight hallucinations). But eventually, it did actually suggest something that led me to the root cause.

AI is great for inspiration and stimulating some of the magical "rubber duck" effect that comes from chatting with a coworker. Just remain vigilant and be very, very skeptical.

Collapse
 
ttsoares profile image
Thomas TS • Edited

You are hilarious... Love it. In fact, you seem to had stolen several of my own thoughts.
As a senior (in age) but junior (in coding) the AI leverage help me a lot about that ugly Typescript syntax in React realms...
Your wisdom at so early age renew the faith that humans will survive the "AI Apocalypse" ;-)
If I may offer a suggestion to add to yours:

  • Use only the free tier of all AIs. This will force using it as a limited resource. And prevent to upload even a medium size system to build context. The codebase as a hole will be to human eyes only.
Collapse
 
victorbolivar__ profile image
Victor Bolivar

Very useful, and appreciate the great humour!