The Numbers That Changed Everything

Last week, OpenRouter dropped their State of AI report. One hundred trillion tokens processed. The number is so big it almost stops making sense. But buried in that report was something more interesting than the headline: programming queries grew from 11% to over 50% of all serious AI usage in 2025.

Strip out the roleplay (yes, that’s 52% of open-source model traffic), and what’s left? Developers. Millions of them. Writing code. Building products. Shipping faster than ever before.

I did some back-of-napkin math, and the implications hit me like a truck.

Let Me Show You The Math

According to the report, programming now dominates non-entertainment AI usage. Claude handles 60%+ of programming traffic. OpenAI and Google split most of the rest.

Here’s what that looks like:

• 100 trillion tokens processed through OpenRouter alone
• Programming went from 11% to 50%+ of total tokens
• That’s roughly 50 trillion tokens spent on code generation in one year
• On one platform. That doesn’t count enterprise Azure, direct API contracts, self-hosted models, or any of the big players routing through private channels.

Now let’s get conservative. Really conservative.

The average coding session might use 10,000 tokens (a few prompts, some back-and-forth, context). Let’s say developers use AI for 20% of their work. That means:

50 trillion tokens ÷ 10,000 tokens per session = 5 billion coding sessions

If each developer does 50 AI-assisted sessions per month, that’s 100 million active developers using AI coding tools. Just on OpenRouter’s visible slice of the market.

The real number across all platforms? Probably 3-5x higher.

What This Actually Means

We’re watching the largest productivity shift in software development history. And it’s not theoretical. It’s happening right now.

• Solo founders ship in weeks instead of months
• Non-technical people build functional MVPs
• Small teams compete with enterprise engineering departments
• The “can you code?” barrier just collapsed

But here’s the thing everyone’s missing: this isn’t about replacing developers. It’s about creating millions of new ones.

Every product manager who can now prototype their idea. Every designer who can bring their vision to life. Every founder who doesn’t need a technical co-founder anymore. They’re all writing code now. With AI.

The 1849 Moment

You know what this reminds me of? The California Gold Rush.

In 1849, everyone rushed to California to mine gold. Most miners went broke. You know who got rich? The people selling picks, shovels, jeans, and supplies to the miners.

Levi Strauss didn’t mine gold. He sold pants to miners and built an empire.

Right now, we’re in the AI coding gold rush. Everyone’s building apps, launching SaaS products, shipping MVPs. And yeah, some will strike it rich.

But the guaranteed opportunity? Building the tools that AI developers need.

The Tool Gap is Massive

Think about what just happened in the last 18 months:

• Cursor exploded in popularity
• Lovable lets you build full apps from prompts
• Bolt.new, Replit Agent, dozens more
• Every week there’s a new “GPT-4 but for X” coding assistant

And what happened to the tooling ecosystem? It’s scrambling to catch up.

The old tools don’t work:

• Traditional SAST tools miss 94% of AI-generated bugs
• Manual code review can’t keep up with AI shipping speeds
• Security testing wasn’t built for prompt-to-production workflows
• There’s no “standard stack” for AI-assisted development yet

Where The Money Is

Based on what I’m seeing, here are the obvious opportunities:

1. Security & Testing Tools

AI generates code fast. Really fast. But it also generates bugs, security holes, and auth bypasses just as fast. Solo founders shipping Cursor-built apps need automated security that matches their velocity.

That’s literally why we built Vibe-Eval. The market is screaming for it.

2. Integration & Workflow Tools

Developers are using 3-5 different AI coding tools. They need:

• Unified interfaces
• Context sharing between tools
• CI/CD pipelines that understand AI-generated code
• Version control that tracks prompt history alongside commits

3. Specialized AI Agents

Generic coding assistants are table stakes. The next wave is vertical-specific:

• “Claude but for fintech compliance”
• “GPT-4 but it knows Shopify’s API by heart”
• “Cursor but for game development in Unity”

4. Training & Education Platforms

Millions of new developers who learned to code in the last 12 months. They need:

• Courses on prompt engineering for code
• Best practices for AI-assisted architecture
• How to review and debug AI-generated code
• Communities and support networks

5. Meta-Tools & Infrastructure

The picks and shovels for people building AI coding tools:

• Datasets for fine-tuning code models
• Evaluation frameworks
• Benchmark suites
• Prompt template libraries

The Vibe-Eval Story

We started Vibe-Eval because we kept seeing the same pattern: brilliant founders building in days what used to take months, then getting absolutely wrecked by security issues they didn’t know existed.

Auth bypasses in 47 seconds. Prompt injections leaking PII. Rate limit bugs that could drain Stripe accounts. The AI generated it fast, shipped it fast, and the vulnerabilities were invisible until production.

Traditional pen testing takes weeks and costs $15k+. By the time you get results, you’ve already shipped three more features with three more bugs.

So we built AI agents that red-team your AI-generated apps. Automatically. In minutes. Before you deploy.

The response has been insane. Solo founders, YC startups, indie hackers—they get it immediately. They’re shipping too fast for old-school security workflows.

Why This is Just The Beginning

Here’s what keeps me up at night (in a good way):

We’re at maybe 1% penetration of the developer market with AI tools. The OpenRouter report shows programming usage doubling every few months. Enterprise adoption hasn’t even really started yet.

And every single one of those developers needs:

• Tools to ship faster
• Tools to ship safer
• Tools to ship smarter
• Tools to learn, collaborate, and scale

The AI coding market isn’t just big. It’s “build a billion-dollar company selling to a subset of a subset” big.

What You Should Do

If you’re building:

• Look for the friction points in AI-assisted development
• Talk to developers who use Cursor, Lovable, Bolt, Replit
• Find the workflow gaps, security holes, integration nightmares
• Build the tool you wish existed

If you’re investing:

• Developer tools for AI coding is the category
• Look for teams who understand both AI and real development workflows
• Bonus points if they’re already using AI to build their own tools
• The winners will be the ones who ship as fast as their customers do

If you’re just watching:

• You’re watching the pickaxe sellers get rich while the miners fight over gold
• The opportunity is right there
• The market is validated
• The demand is screaming

The Bottom Line

100 trillion tokens. 50% to programming. Millions of developers. Growing exponentially.

This isn’t a trend. It’s a tectonic shift.

The gold rush is here. The miners are flooding in. And there aren’t nearly enough pickaxes to go around.

So the question isn’t whether to build tools for AI developers.

The question is: which tool are you going to build?

Ready to build in the AI coding gold rush? Start by securing what you ship. Try Vibe-Eval and see what your AI-generated code is hiding.