Forem: IntelligentTools

The AI Bubble Looks Exactly Like January 2000

IntelligentTools — Thu, 19 Feb 2026 10:56:07 +0000

Super Bowl LX just happened. Nearly a quarter of the ads were for AI companies. OpenAI, Anthropic, Google, Amazon, Meta—all spending $8-10 million per 30-second spot.

Companies that burn billions in losses pay millions to tell you their products are revolutionary.

I've seen this before. So have you, if you were coding in 2000.

The Dot-Com Parallel Nobody's Talking About

January 2000: A leaked essay called "bubble.com" warned that internet companies were catastrophically overvalued. Nobody listened. Two months later, the market peaked and then lost 75% of its value over 18 months.

Super Bowl XXXIV (January 30, 2000) was packed with dot-com ads. Pets.com with the sock puppet. E*Trade. Seventeen different internet startups are burning venture capital on advertising.

Pets.com had its IPO eleven days after the Super Bowl. The company shut down nine months later.

The Numbers Don't Make Sense

OpenAI is reportedly discussing a funding round that would value it at $830 billion—more than Coca-Cola, Chevron, or Costco.

The company projects:

$14 billion in losses for 2026
$115 billion in cumulative losses through 2029
$20 billion in revenue against $1.4 trillion in compute commitments

That's 1.4% of commitments covered by revenue.

HSBC analysis: Even if OpenAI hits $200 billion in revenue by 2030, they'll still need an additional $207 billion in funding to stay in business.

Microsoft lost $440 billion in market capitalization in a single day after investors questioned its AI spending plans.

The Debt Bubble Under the AI Bubble

Companies supplying infrastructure to OpenAI have taken on $96 billion in debt to fund operations. The big five tech companies (Amazon, Google, Meta, Microsoft, Oracle) added $121 billion in new debt in 2025 alone—four times their historical average.

Oracle is being sued by bondholders for concealing the amount of debt they'd need for an AI buildout. Their credit rating is approaching junk status.

This is exactly how bubbles work: companies with unproven business models borrow unsustainable amounts, then spend desperately on advertising to create the appearance of legitimacy.

The Tools Aren't Going Away

I use Claude Code daily. AI coding assistants are genuinely useful when used correctly. But they're tools that require human judgment, not replacements for it.

The valuable tools will survive the crash. The rest won't. Companies that burn billions while losing money won't make it. The ones with sustainable unit economics might.

What This Means for Developers

Short term: Keep using AI tools while they're heavily subsidized. Learn what works and what doesn't. Build things you couldn't build before because the tedium was too high.

Long term: The opportunities after the crash are probably more interesting than what we're working on now. The dot-com crash gave us Google, Amazon, and modern web development. The AI crash will leave us with whatever actually works.

The code still needs to be written. The bugs still need to be found. The architecture still needs human judgment.

That's not changing, regardless of how many AI companies burn through their funding. That said, I am writing this article in a blog pair-programmed with Claude. Grammarly checked it and offered suggestions. AI is everywhere, and it can't be stopped, no matter the economy.

And yeah, the image is also AI 😄

Originally published at intelligenttools.co

I Let Claude Code Run Unsupervised for 8 Hours - Here's What It Built

IntelligentTools — Fri, 16 Jan 2026 22:19:27 +0000

Last week, I discovered the Ralph Wiggum technique for Claude Code. Named after the perpetually confused Simpsons character, it lets Claude iterate on code autonomously until tests pass.

I gave it a real production task: refactor our healthcare app's authentication module (8 files, 1,200 lines, JWT handling, session management, password reset).

Started: 11 PM Friday

Woke up: 7 AM Saturday

Result: 47 commits, auth module fully refactored, test coverage improved from 62% to 87%

Cost: $23.14 in API credits

Time saved: 6-8 hours of my weekend

The Setup

claude
> /plugin install ralph-loop@claude-plugins-official
> /ralph-loop "Refactor auth module. Extract business logic from controllers. 
   Max 20 lines per function. Coverage 80%+. All tests must pass."
   --max-iterations 100

Then I went to sleep.

What Actually Happened

Claude ran 47 iterations, each one:

Making changes
Running tests
Reading failures
Adjusting approach
Trying again

When tests finally passed at iteration 47, it stopped with <promise>COMPLETE</promise>.

The Real Results

Worked:

Business logic cleanly separated from controllers
All 8 files refactored systematically
Test coverage jumped to 87%
Zero TypeScript errors
Clean, consistent commit messages

Didn't work:

Still had to review everything (found 2 edge cases it missed)
Cost analysis wasn't perfect ($23 vs my estimate of $15)
Some function names were... creative

The Prompt Makes Everything

My first attempt with a vague prompt ran for 30 iterations and produced garbage.

What worked:

Clear success criteria (tests pass, coverage %, linting)
Explicit constraints (max line length, style rules)
Step-by-step process
Definitive exit condition

What didn't:

"Make it better"
"Improve code quality"
Subjective goals with no metrics

When to Use This (and When Not To)

Perfect for:

Refactoring with good test coverage
Adding tests to untested code
Bug fixes with failing tests
Code cleanup tasks

Don't use for:

Anything without tests (Ralph has no feedback)
Security-critical code (needs human review at each step)
Subjective work (UI design, naming, documentation)
Production data (run in isolated dev environments only)

The Bottom Line

This isn't AGI. It's an extremely capable pattern matcher with good self-correction when tests provide clear feedback.

For healthcare software with HIPAA requirements, I'm keeping autonomous loops in dev environments only. But for the right tasks? It's like having a junior developer who works 24/7 for $20/day.

Full writeup with detailed breakdown, failures, lessons learned, and exact prompts:

👉 Read the complete article on intelligenttools.co

Questions:

Have you tried autonomous coding loops?
What's your experience with Claude Code or similar tools?
What guardrails would you add to this approach?

Drop your thoughts below - genuinely curious how others approach this.

AI Coding in 2026: 10 Predictions

IntelligentTools — Fri, 26 Dec 2025 11:38:46 +0000

I spent 18 months testing AI coding tools in production and analyzed Stack Overflow's 2025 AI Developer Survey (71,000+ developers, 76% now using AI for code generation).

After testing Claude Code, Cursor, GitHub Copilot, and 17 other tools on genuine healthcare software, here are my predictions for what actually happens in 2026.

Some of these will make you uncomfortable. Good.

🚨 Junior devs stop writing boilerplate entirely

By late 2026, companies will test for "AI collaboration skills" instead of syntax memorization. Interview questions shift from "Write a function to sort an array" to "How would you prompt Claude to build this feature?"

The more established your tech stack, the faster this hits. I'm already seeing it with Node.js/TypeScript.

💰 The "AI-First Developer" becomes a job category

New role: 80% directing AI agents, 20% writing critical logic. Salary range: $120k-180k. One AI-first developer replaces 3-4 traditional juniors on greenfield projects.

This isn't about replacing developers. It's about creating force multipliers who know how to orchestrate AI tools while maintaining code quality.

🔐 First major security breach from AI-generated code

A high-profile incident stems from AI-generated code with a subtle vulnerability. Trust in AI tools drops from 44% to sub-30% for 2-3 months before recovering with better tooling.

This is inevitable. AI tools generate millions of lines of code daily. Eventually, one will cause a significant incident.

💸 Pricing flips from subscription to usage-based

Cursor's $20/month unlimited becomes unsustainable as compute costs surge. By mid-2026, most tools switch to $0.02-0.05 per request. Power users generating entire backends pay $300+/month.

The economics don't work for flat-rate unlimited access when some users generate 100x more tokens than others.

Read All 10 Predictions

I've shared 4 of 10 predictions here. The complete analysis includes:

Why Claude Code captures 30% of the CLI-based market
MCP Protocol becoming the connector standard (5,000+ servers by Q4)
Tool fragmentation by developer specialty (frontend vs backend vs DevOps)
Open-source AI tools reaching feature parity
First AI coding tool IPO (Cursor at $5B+ or Copilot spins out)
Prompt engineering is becoming a 2-3 hour/week skill

Each prediction includes my reasoning, timeline, and what developers should do to prepare.

👉 Read the full article with all 10 predictions and analysis →

Methodology

Based on:

Stack Overflow 2025 AI Developer Survey (71,000+ devs)
18 months testing 20+ AI coding tools
100+ conversations with developers in production
My own production use at ICANotes (healthcare software)

Not based on: YouTube demos, marketing materials, or Twitter hype.

About Me

I'm Boki, a senior full-stack developer in Serbia working on healthcare systems. I run intelligenttools.co where I share honest, technical AI tool reviews from a developer's perspective.

What's your boldest prediction for AI coding in 2026? Drop it below 👇

--- title: 10 Bold Predictions for AI Coding Tools in 2026 published: true description: 10 predictions from 18 months of production testing (some will make you uncomfortable) tags: ai, webdev, predictions, coding cover_image:

IntelligentTools — Fri, 26 Dec 2025 11:34:17 +0000

ChatGPT vs Claude Code: Over a Year of Building Production Software

IntelligentTools — Sat, 20 Dec 2025 12:04:21 +0000

Quick intro

I've spent over a year using both ChatGPT and Claude Code CLI daily while building healthcare software. Not for side projects or demos - for actual production systems where bugs matter.

This isn't a benchmark comparison. Those tell you nothing about real work. This is what it's actually like to use both tools when you're shipping code that matters.

Spoiler: They're not competing. They solve different problems.

ChatGPT vs Claude Code: Over a Year of Building Production Software

I've been using AI coding tools for about a year now, and for the last 6 months, Claude Code CLI has been running in my terminal every single day. But I'm not here to tell you it's perfect or that you should switch. I'm here to tell you what actually matters when you're building real software.

Because here's the thing: ChatGPT and Claude Code aren't really competing. They're solving different problems.

The Setup

I'm a senior full-stack developer at a healthcare software company. I build clinical systems with Node.js, MySQL, and AWS. My typical day involves fixing bugs in production, adding features to existing systems, and reviewing code from our team.

For the past year, I've used Claude Code CLI as my primary coding assistant. I also use ChatGPT (with the new GPT-5 models) for specific tasks. This isn't a benchmark comparison – those tell you nothing about real work. This is what it's actually like to use both tools when you're shipping code that matters.

What They Actually Are

ChatGPT is a web interface for chatting with an AI. You can describe problems, paste code, and get suggestions. With the new GPT-5, GPT-5.1, and GPT-5.2 models, it's legitimately smart at coding. OpenAI also has Codex (their cloud coding agent), but I'm comparing ChatGPT's web chat interface here.

Claude Code CLI is a terminal tool. You run it in your project directory, and it can see your entire codebase, modify files, run git commands, execute bash scripts, and use your CLI tools. It's powered by Claude Sonnet 4.5 (I can switch to Opus 4.5 for complex stuff). It's not a chat interface – it's integrated into your actual development workflow.

That difference is everything.

[Continue with the rest of my blog post content here: https://intelligenttools.co/blog/chatgpt-vs-claude-code-1-year ]