Forem: Learn AI Resource

Stop Getting Mediocre Answers from AI Coding Assistants

Learn AI Resource — Sat, 02 May 2026 15:01:16 +0000

Stop Getting Mediocre Answers from AI Coding Assistants

Your AI coding assistant is probably smarter than you think. You're just asking the wrong way.

I spent the last 6 months using Claude, ChatGPT, and Copilot daily. Noticed something weird: some developers get incredible help from AI, while others get generic garbage that barely compiles.

The difference? Not the AI. It's how you ask.

Here are 5 techniques that actually work.

1. Give Context Like You're Onboarding a Junior Dev

Bad prompt:

How do I add authentication to my app?

Good prompt:

I'm building a Next.js 14 app with TypeScript and Prisma. 
Currently using API routes for backend. Need to add 
JWT-based authentication with refresh tokens. Users already 
exist in the database (User model has email/password fields).

How should I implement this?

Why it works: AI isn't psychic. It doesn't know your stack, your constraints, or what you've already tried. Give it the same context you'd give a new teammate.

Pro tip: Keep a "context template" for your project. Paste it at the start of new conversations:

Stack: Next.js 14, TypeScript, Prisma, PostgreSQL
Deployment: Vercel
Constraints: Need server-side rendering, keep bundle small
Current auth: None yet

2. Show Your Failed Attempt First

Instead of asking "how do I do X," show what you tried:

I'm trying to debounce this search input but it's not working:

[paste your code]

The search still fires on every keystroke. I tried using 
setTimeout but the previous timeout isn't getting cleared.
What am I missing?

Why it works: You get a targeted fix instead of a generic tutorial. The AI can see your actual problem, not guess what it might be.

This is how I debug now:

Try to fix it myself (5 min)
Paste broken code to AI with specific error
Get solution in 30 seconds
Move on

Saves me hours of StackOverflow rabbit holes.

3. Ask for Explanations Like You're Five

When you get code back, don't just copy-paste it. Ask:

Explain this line by line like I'm a junior developer.
Why did you use useCallback here? What would break without it?

Real example from yesterday:

AI gave me this React optimization:

const memoizedValue = useMemo(() => computeExpensiveValue(a, b), [a, b]);

I asked "why useMemo instead of just calling the function?"

Got back: "computeExpensiveValue runs on every render. With 60fps that's 60 calls/sec even when a and b don't change. useMemo caches the result."

Now I actually understand why, not just what.

Side benefit: You'll spot when AI is wrong. It happens. Understanding the "why" is your BS detector.

4. Iterate in the Same Conversation

Don't start a fresh chat for every follow-up question. Build on what you've established.

Session example:

Me: How do I validate email input in React?
AI: [gives validation code]

Me: Now add password validation with these rules: 
    8+ chars, 1 number, 1 special char

AI: [adds to existing code]

Me: Show me how to display validation errors below each field

AI: [extends the solution]

Each response builds on the previous context. By message 3, you have a complete, working solution that fits together.

Reset when: You switch topics or the AI starts hallucinating. Otherwise, ride that context.

5. Use "Rubber Duck Mode" for Complex Problems

Sometimes I'm not even sure what question to ask. That's when I rubber duck with AI:

I'm building a file upload system. Users can upload images, 
and I need to:
- Validate file types
- Compress images before upload
- Show upload progress
- Store in S3
- Save metadata to database

I'm not sure where to start or what could go wrong. 
What's the best architecture for this?

Why it works: By explaining the full problem (not just "how do I upload files"), you get a solution that considers the whole picture.

AI will often spot edge cases you missed:

"What happens if upload fails halfway?"
"How will you handle large files (>50MB)?"
"Should compression happen client-side or server-side?"

These are questions that would've bitten you later.

The Meta-Skill: Treat AI Like a Pair Programming Partner

Here's the real unlock: stop treating AI like Google.

Google: Query → Answer → Done

AI Assistant: Problem → Discussion → Solution → Refinement

Best developers I know use AI like they'd use a senior dev sitting next to them:

"Here's what I'm thinking, does this make sense?"
"I'm stuck on this part, what am I missing?"
"Is there a better way to do this?"

You wouldn't ask a teammate "how do I center a div" and walk away. You'd have a conversation.

Same with AI.

Bonus: The Copy-Paste Test

Before you paste AI code into your project, ask:

What could go wrong with this code in production?
What edge cases am I not handling?

Good AI will roast its own solution. You'll get:

"This doesn't handle null values"
"Memory leak if component unmounts during fetch"
"No rate limiting on API calls"

Fix those before shipping. Your future self will thank you.

The Compound Effect

These techniques are small. Each one saves maybe 5 minutes per coding session.

But if you code daily, that's 30+ hours saved per year. Plus, better code. Plus, you actually learn instead of just copy-pasting.

I went from "AI tools are overhyped" to "can't work without them" by changing how I asked questions.

Your mileage may vary, but give these a shot for a week. You'll notice the difference.

Level Up Your AI + Dev Game

I write about AI tools, productivity hacks, and developer workflows every week. Real techniques that work, not generic "AI is amazing" content.

Subscribe to LearnAI Weekly — practical AI tips delivered to your inbox.

What's your best trick for working with AI assistants? Drop it in the comments — always looking to level up.

5 Terminal Tools That Actually Changed How I Code

Learn AI Resource — Fri, 01 May 2026 15:01:26 +0000

You know that feeling when you discover a tool and think "where has this been all my life?"

These five terminal tools did that for me. Not the obvious ones everyone talks about (yeah, we all know about git and npm). These are the productivity gems that actually changed my daily workflow.

1. `fzf` - Fuzzy Finder That Feels Like Magic

If you're still typing out full file paths in 2026, we need to talk.

fzf is a fuzzy finder that lets you search through ANYTHING — files, command history, git branches, processes. Type a few letters, get instant results.

Real example from today:
Instead of git checkout feature/user-authentication-refactor-v2, I just type:

git checkout $(git branch | fzf)

Then type "auth" and boom, done.

Install:

# Mac
brew install fzf

# Linux
sudo apt install fzf

Pro tip: Add this to your .bashrc or .zshrc:

# Ctrl+R for command history search
[ -f ~/.fzf.bash ] && source ~/.fzf.bash

Now Ctrl+R gives you a beautiful searchable history instead of that janky default.

2. `bat` - Cat on Steroids

cat is fine, but bat is cat with syntax highlighting, line numbers, and git integration built in.

Before:

cat server.js  # wall of monochrome text

After:

bat server.js  # beautiful highlighted code

It automatically pages long files, shows git modifications in the sidebar, and even integrates with fzf:

# Preview files while searching
fzf --preview 'bat --color=always {}'

Install:

brew install bat
# or
sudo apt install bat

3. `ripgrep` (rg) - Grep But Insanely Fast

grep is slow. ripgrep is stupid fast.

I used to wait 5-10 seconds searching a large codebase. Now it's instant.

Search all files for "TODO":

rg "TODO" --type js

Find all API endpoints:

rg "app\.(get|post|put|delete)" --type js

Case-insensitive search with context:

rg -i "database" -C 3

It respects .gitignore by default (huge), supports regex, and shows results in a clean format.

Install:

brew install ripgrep
# or
sudo apt install ripgrep

4. `tldr` - Man Pages for Humans

man pages are comprehensive but overwhelming. tldr gives you practical examples instead.

Compare these:

man tar  # 3000 lines of dense documentation
tldr tar # 8 examples that actually help

Output from tldr tar:

  tar

  Archiving utility.

  - Create an archive from files:
    tar -cf target.tar file1 file2 file3

  - Extract an archive:
    tar -xf source.tar

  - Create a gzipped archive:
    tar -czf target.tar.gz file1 file2

You get the idea. It's the StackOverflow answer you were going to Google anyway.

Install:

npm install -g tldr
# or
brew install tldr

5. `jq` - JSON Swiss Army Knife

If you work with APIs, jq will change your life.

Pretty-print API response:

curl https://api.example.com/users | jq

Extract specific fields:

curl https://api.example.com/users | jq '.[].email'

Filter and transform:

# Get all users with admin role
curl https://api.example.com/users | jq '[.[] | select(.role == "admin")]'

I used to pipe JSON to a file and open it in VS Code. Now I just use jq and move on.

Install:

brew install jq
# or
sudo apt install jq

The Compound Effect

Here's the thing: each tool saves maybe 30 seconds per use. Doesn't sound like much.

But I use these tools dozens of times a day. That's 15-20 minutes saved daily. Over a year? 100+ hours of productivity gained.

Plus, they make coding more enjoyable. When your tools feel smooth, you stay in flow longer.

Want More Dev Productivity Tips?

I write about tools, AI workflows, and productivity hacks for developers every week. No fluff, just practical stuff you can use immediately.

Subscribe to LearnAI Weekly — my newsletter where I share what's actually working.

What terminal tools are in your daily rotation? Drop a comment — always looking to level up my setup.

5 No-Code Tools That Devs Actually Use (And Why You Should Too)

Learn AI Resource — Thu, 30 Apr 2026 15:01:21 +0000

5 No-Code Tools That Devs Actually Use (And Why You Should Too)

I used to roll my eyes at no-code tools. "Real developers code everything from scratch," right?

Wrong. So wrong.

After years of building everything manually, I finally tried some no-code tools. Not because I couldn't code it myself, but because I wanted to ship faster and focus on the interesting problems.

Here's what I wish someone had told me earlier: no-code tools aren't about replacing coding—they're about not wasting time on solved problems.

1. Zapier/Make - Because Writing Glue Code Sucks

What it does: Connects apps and automates workflows without writing integration code.

Why devs use it: Because writing another OAuth integration or webhook handler for the 47th time is not how I want to spend my Tuesday.

Real example: I built a system that:

Watches my GitHub releases
Posts to Discord automatically
Updates my changelog website
Sends me a Slack notification

Total time with Zapier: 15 minutes. Time to code it from scratch: probably 3 hours, plus maintenance when APIs change.

When to code it yourself: If you need complex logic, custom error handling, or you're processing sensitive data that shouldn't go through third parties.

When to use no-code: Literally everything else. Your time is worth more than proving you can write another webhook.

2. Retool - Admin Panels in Minutes

What it does: Builds internal tools and dashboards by connecting to your database/APIs with a visual editor.

Why devs use it: Because building admin panels is the most soul-crushing part of development, and nobody ever thanks you for the beautiful CRUD interface you spent two weeks perfecting.

Real example: Needed a customer support dashboard to:

View user data from Postgres
Run common queries (refunds, account issues)
Make simple updates
Export reports

Built it in Retool in an afternoon. Would've taken a week in React, plus ongoing maintenance.

Pro tip: Use it for internal tools only. Your actual product should still be properly coded. But that admin panel your team needs? Retool it.

3. Supabase - PostgreSQL Without the DevOps

What it does: Managed Postgres with instant APIs, auth, and real-time subscriptions.

Why devs use it: Because I know how to set up PostgreSQL, configure pgBouncer, handle migrations, and manage backups. I just don't want to anymore.

Real example: Spinning up a new side project:

✅ Database: Supabase (5 minutes)
✅ Auth: Supabase (10 minutes)
✅ API: Auto-generated (0 minutes)
❌ Deploy: Render/Railway (actually have to write some code here)

Result: Started building features on day one instead of day three.

Is it cheating? No. It's called not reinventing the wheel. You're still writing your application logic, just not babysitting infrastructure.

4. Airtable - When Excel Meets a Database

What it does: Spreadsheet-database hybrid with API access.

Why devs use it: Because sometimes your non-technical team needs to manage data, and teaching them SQL is not the answer.

Real example: Content calendar for a blog:

Writers add ideas directly (no tickets needed)
Auto-assigns to editors based on topic
Generates a public RSS feed via Airtable API
Posts to social media via Zapier

Writers love it because it's familiar (looks like a spreadsheet). I love it because there's an API and I don't have to build another content management interface.

Developer move: Use Airtable as your database for MVPs and prototypes. If you scale to where it's too slow, migrate to Postgres. Most projects never reach that point.

5. Tally/Typeform - Forms That Don't Make Me Cry

What it does: Beautiful forms with logic, payments, and integrations.

Why devs use it: Because building a multi-step form with conditional logic and file uploads from scratch makes me want to quit programming and become a goat farmer.

Real example: User onboarding survey:

Conditional questions based on previous answers
File upload for profile picture
Payment integration
Saves to Airtable automatically

Built in 20 minutes. Coding this properly with validation, file handling, and conditional rendering? Half a day minimum.

Bonus: These tools have better UX than whatever form you were planning to build. Sorry, but it's true.

The Real Question: When to Code vs When to No-Code

Here's my framework:

Use no-code when:

It's not your core product
Speed matters more than customization
Non-developers need to manage it
The problem is already solved a million times
You want to test an idea quickly

Code it yourself when:

It's your main product/feature
You need complete control
Performance is critical
You're learning something new
Privacy/security requires it
The no-code tool costs more than your time

What This Actually Looks Like

My current SaaS stack:

Frontend: Next.js (coded)
Backend API: Node.js (coded)
Database: Supabase (no-code)
Auth: Supabase (no-code)
Admin panel: Retool (no-code)
Email automations: Zapier (no-code)
User feedback: Tally (no-code)
Analytics: PostHog (no-code)

Am I less of a developer because I didn't build my own auth system from scratch? No. I'm just a developer who ships products instead of infrastructure.

The Uncomfortable Truth

Every hour you spend building something that already exists as a service is an hour you're not spending on your actual product.

Yes, you could build your own authentication system. You could also build your own database engine and operating system while you're at it.

Or you could use the good tools, ship your product, and get users.

Nobody cares if you used Supabase or rolled your own auth. They care if your product works and solves their problem.

Your Homework

Pick one thing you're currently building (or planning to build) and ask yourself: "Does this exist as a no-code tool?"

If yes, try it. Just for a week. See if it actually sucks or if your ego was just getting in the way.

My bet? You'll ship faster, sleep better, and have more time to work on the problems that actually matter.

And if you still prefer coding everything from scratch, cool! At least now you're making an informed choice instead of an automatic one.

Want more honest takes on AI tools, developer productivity, and building things that matter? I share the useful stuff (and call out the BS) in my weekly newsletter. Sign up for LearnAI Weekly and get the good intel without the hype.

Hot take time: What's your most controversial opinion on no-code tools? Tell me in the comments. I promise I can handle it.

How I Use AI Coding Assistants Without Turning My Brain to Mush

Learn AI Resource — Wed, 29 Apr 2026 15:01:22 +0000

How I Use AI Coding Assistants Without Turning My Brain to Mush

Here's the thing: AI coding tools are incredible. They're also kinda scary if you use them wrong.

I've been coding with AI assistants for months now, and I've learned the hard way that there's a right way and a wrong way to do this. The wrong way? Copy-paste everything, never think, wake up one day and realize you can't write a for-loop without prompting ChatGPT.

Here's what actually works.

The 30-Second Rule

Before I ask AI for help, I give myself 30 seconds to think about the problem. That's it. Just 30 seconds.

Why? Because most of the time, I already know the answer. My brain is just being lazy. Those 30 seconds force me to actually engage with the problem instead of outsourcing my thinking immediately.

Example: Need to reverse a string in Python? You probably know this. Take 30 seconds. Oh right, string[::-1]. Done.

The AI is there for when you're genuinely stuck, not when you're being lazy.

Use AI for Boilerplate, Not Logic

Here's my rule: AI writes the boring stuff, I write the interesting stuff.

Setting up Express routes? Sure, let AI generate that. The actual business logic that makes my app unique? That's mine.

What I let AI do:

Config files (tsconfig, webpack, whatever)
Basic CRUD operations
Test boilerplate
Type definitions

What I do myself:

Core algorithms
Business logic
Anything that requires understanding my specific domain
Bug fixes (more on this below)

The Debug-It-Yourself Rule

This one's non-negotiable: when something breaks, I debug it myself first.

AI is terrible at debugging because it doesn't see your full context. It'll suggest random fixes that might work but don't actually solve the root problem.

More importantly: debugging is how you learn. Every bug is a lesson. Skip that and you're just collecting code you don't understand.

My process:

Read the error message (shocking, I know)
Check the obvious stuff (typos, wrong variable names)
Console.log / debugger my way through it
Only then, if I'm truly stuck, explain the problem to AI

Usually by step 3, I've found it.

Ask "Why" Even When You Get the Answer

AI gives you code. Cool. Now ask it to explain what it just wrote.

Then read that explanation. Actually read it. If something doesn't make sense, ask follow-up questions.

Bad: copies code "Thanks!"

Good: "Why did you use Promise.all here instead of Promise.allSettled?"

Treat AI like a senior dev who's explaining code to you. You wouldn't just copy their code without understanding it, right? Right?

Keep a "Learned Today" Note

Every time AI teaches me something genuinely new, I write it down. Just a quick note in Obsidian or whatever.

Examples from my notes:

"Optional chaining works in object destructuring"
"Array.at(-1) is cleaner than array[array.length-1]"
"CSS has container queries now"

Writing it down makes it stick. Plus, I review these notes every month. Stuff that used to be "AI knowledge" becomes my knowledge.

The Rewrite Test

Once a week, I pick something AI helped me build and rewrite it from scratch without AI.

Can you do it? If yes, congrats, you learned something. If no, you just borrowed code you don't understand.

This is uncomfortable but necessary. It's like checking if you can actually speak Spanish or if you just know how to use Google Translate.

Real Talk: The Balance

Look, I'm not saying never use AI. I use it constantly. It's saved me hundreds of hours.

But here's the question: are you using AI to augment your skills or replace them?

The difference is huge:

Augment: "I know how to do this, but AI can do it faster"
Replace: "I have no idea how this works, but AI gave me code that runs"

One makes you a better developer. The other makes you dependent.

What This Looks Like In Practice

Yesterday I built a feature that parses CSV files and generates reports. Here's how I used AI:

AI generated: The CSV parsing boilerplate (30 seconds vs 5 minutes writing it myself)
I wrote: The actual report logic, validation rules, error handling
AI helped: Suggested edge cases I hadn't considered
I debugged: When it broke (because of course it broke)

Total time: ~2 hours. Without AI? Maybe 3 hours. With AI but no thinking? Either 30 minutes of copying code I don't understand, or infinite time debugging mysterious issues.

The Bottom Line

AI coding assistants are tools, not shortcuts. Use them like you'd use Stack Overflow: helpful for learning, dangerous if you just copy-paste blindly.

Stay sharp. Keep thinking. Let AI handle the boring stuff so you can focus on the interesting problems.

And for the love of all that is holy, please debug your own code.

Want more practical dev tips and AI tool reviews? I send out a weekly newsletter with the good stuff, no fluff. Sign up for LearnAI Weekly and get smarter about using AI without losing your edge.

Your turn: How do you balance AI assistance with actual learning? Drop a comment, I'm curious what's working for other devs.

Top 3 AI Tools for Developers This Week

Learn AI Resource — Sat, 25 Apr 2026 01:08:46 +0000

Issue #1 | April 28, 2026

Welcome to LearnAI Weekly

This newsletter is for developers who want to stay ahead without drowning in AI hype.

What you'll get every week:

3 AI tools that solve real problems in your workflow
Real-world use cases - not marketing demos
Practical insights - what's working, what's noise

No tutorials you won't finish. No tools you'll never use. Just the stuff that makes you more productive.

Let's dive in.

Hey developers 👋

Every week, I'm cutting through the noise to bring you 3 AI tools that actually matter for your work.

No fluff. No hype. Just tools you can use today.

🔥 #1: Cursor IDE - Your AI Pair Programmer

What it is: An AI-native code editor built on VS Code that understands your entire codebase.

Why it matters:

Context-aware completions that understand your project structure, not just the current file
Natural language edits - describe what you want, and it writes the code
Built-in chat with access to your codebase docs

Real-world use case:
You're refactoring a legacy .NET API. Instead of manually updating dozens of endpoints, you tell Cursor: "Update all controller methods to use async/await pattern and add proper error handling." It does it in seconds, maintaining your coding style.

Get started: cursor.sh | Free tier available

🛠️ #2: v0.dev by Vercel - UI to Code in Seconds

What it is: Describe a UI component, get production-ready React code instantly.

Why it matters:

Skip boilerplate hell - no more "how do I center this div?"
Generates Tailwind CSS + Shadcn components that match modern design standards
Iterate in natural language - "make it responsive" or "add dark mode"

Real-world use case:
You need a dashboard with charts for your side project. Instead of spending 3 hours on styling, you describe it: "Admin dashboard with revenue chart, user stats cards, and a data table." Get working code in 30 seconds. Customize from there.

Get started: v0.dev | Free credits available

🧪 #3: GitHub Copilot Workspace - Debug Like a Pro

What it is: AI-powered debugging and testing environment that runs in your browser.

Why it matters:

Explains stack traces in plain English - no more Googling cryptic error messages
Suggests fixes with context from your entire repo
Generates test cases automatically based on your code changes

Real-world use case:
You push a commit and CI fails with a weird TypeScript error. Instead of context-switching to dig through logs, Copilot Workspace analyzes the failure, identifies the type mismatch, and suggests 2 ways to fix it—complete with test cases.

Get started: GitHub Copilot Workspace | Included with Copilot subscription

💡 Quick Take

This week's pattern: AI tools are moving from "helpful suggestions" to "actually doing the work."

The winners? Tools that understand context (your full project, not just a snippet) and integrate into your existing workflow (VS Code, GitHub, etc.).

If you're still using ChatGPT in a separate tab for coding, you're leaving 10x productivity on the table.

What's Next?

Next week: AI tools for API testing and documentation - because nobody likes writing OpenAPI specs by hand.

Got a tool I should cover? Hit reply and let me know.

— Kevin

learnairesource.com

P.S. If this was helpful, forward it to a dev friend. They'll thank you.

Forem: Learn AI Resource

Stop Getting Mediocre Answers from AI Coding Assistants

Stop Getting Mediocre Answers from AI Coding Assistants

1. Give Context Like You're Onboarding a Junior Dev

2. Show Your Failed Attempt First

3. Ask for Explanations Like You're Five

4. Iterate in the Same Conversation

5. Use "Rubber Duck Mode" for Complex Problems

The Meta-Skill: Treat AI Like a Pair Programming Partner

Bonus: The Copy-Paste Test

The Compound Effect

Level Up Your AI + Dev Game

5 Terminal Tools That Actually Changed How I Code

1. fzf - Fuzzy Finder That Feels Like Magic

2. bat - Cat on Steroids

3. ripgrep (rg) - Grep But Insanely Fast

4. tldr - Man Pages for Humans

5. jq - JSON Swiss Army Knife

The Compound Effect

Want More Dev Productivity Tips?

5 No-Code Tools That Devs Actually Use (And Why You Should Too)

5 No-Code Tools That Devs Actually Use (And Why You Should Too)

1. Zapier/Make - Because Writing Glue Code Sucks

2. Retool - Admin Panels in Minutes

3. Supabase - PostgreSQL Without the DevOps

4. Airtable - When Excel Meets a Database

5. Tally/Typeform - Forms That Don't Make Me Cry

The Real Question: When to Code vs When to No-Code

What This Actually Looks Like

The Uncomfortable Truth

Your Homework

How I Use AI Coding Assistants Without Turning My Brain to Mush

How I Use AI Coding Assistants Without Turning My Brain to Mush

The 30-Second Rule

Use AI for Boilerplate, Not Logic

The Debug-It-Yourself Rule

Ask "Why" Even When You Get the Answer

Keep a "Learned Today" Note

The Rewrite Test

Real Talk: The Balance

What This Looks Like In Practice

The Bottom Line

Top 3 AI Tools for Developers This Week

Welcome to LearnAI Weekly

🔥 #1: Cursor IDE - Your AI Pair Programmer

🛠️ #2: v0.dev by Vercel - UI to Code in Seconds

🧪 #3: GitHub Copilot Workspace - Debug Like a Pro

💡 Quick Take

What's Next?

1. `fzf` - Fuzzy Finder That Feels Like Magic

2. `bat` - Cat on Steroids

3. `ripgrep` (rg) - Grep But Insanely Fast

4. `tldr` - Man Pages for Humans

5. `jq` - JSON Swiss Army Knife