Hives

I've been heads-down working with AI code generation tools for a solid month now.

May 14, 2025

Here's the proof, courtesy of the VS Code plug-in for GitHub Copilot Chat.

Copilot usage: 11823 chat requests in last 30 days

Too much? Or not enough?

Yep. That says 11,823 chat requests in the last 30 days.

That's about 400 per day, or 1 request/minute over an eight-hour day.

Let's make a baseline estimate: Each of those chat requests, whether for generating ideas, writing code, or compiling documentation, would have taken at least an hour to handle a similar request by hand.

If we follow the 10,000-hours rule, that makes me a virtual expert on AI. Or it makes AI a virtual expert on Ivan.

And what did all that experimentation cost me?

$10 USD plus tax.

But the party's over. Starting June 4, 2025, billing for premium requests will start the meter running at $0.04 each.

Comparing Copilot plans: Copilot Pro: $10 USD per month for 300 Premium requests per month

Too much? Or not enough?

On the Copilot Enterprise plan, my experiments would have cost $471.92.

What did I gain through all my experiments? Was it worth it?

I'll tell you later.

But I'll tell you this right now: Unless you're as obsessed as I am with the capabilities of AI code generation, you may be in trouble.

Especially if you're a project manager, product owner, technical lead/architect, UX/UI designer, information architect, backend developer, frontend developer, database engineer, DevOps engineer, integration specialist, API developer, QA engineer, test automation engineer, performance engineer, security engineer, compliance specialist, or specialty specialist in whatever field.

Generally available AI code development tools do all of the above and more, and it's accelerating quickly.

For my early Python projects, I fed prompts into Copilot in the browser and got code results back, which I could paste back into VS Code for testing and refinement. That was enough of a boost to get my QR Code generator working as an Azure Function App.

In GitHub Copilot, Ask mode gives similar results with the added benefit that the AI can see your codebase and documentation to give better answers.

With GitHub Copilot in Edit mode, AI updates your code for you, wherever it is in your repository.

And with GitHub Copilot in Agent mode (a premium feature), AI not only updates your code, but independently comes up with a plan to execute multistep commands. You watch it bounce along from one part of your project to the next. You can jump in to redirect or provide clarification, but it handles the rest independently.

You're probably skeptical right now.

Can AI really do all that? How do you know it works? Can you trust what you build? Where's the proof?

Is Ivan bluffing/deluded/overconfident or does he have a monster hand?

Well, I'm not showing my cards just yet.

This game is just getting started.