Mintlify analyzed 790 million requests across its documentation platform. The finding: AI coding agents account for 45.3% of all traffic, nearly tied with traditional browsers at 45.8%.

Two tools are driving almost all of it:

The rest of the field, OpenCode, Trae, ChatGPT, and NotebookLM, is showing up but nowhere close.

One caveat: OpenAI's Codex doesn't send an identifiable user-agent header, so the real agent percentage is likely even higher.

The takeaway for anyone maintaining developer docs: your documentation now serves two audiences. Structure and machine-readability matter as much as clarity for human readers.

Try Mintlify today!

Last year, I decided to build a free CV scanner for job hunters.

The idea was simple: upload your CV, get instant feedback on what's missing. Bullet points, keywords, action verbs, the whole lot.

I started the project on a Monday morning. By Friday evening, I had maybe 40% of it working — basic PDF parsing, a clunky scoring system, zero polish. That 40% took me four weeks of evenings and weekends. I'd find a bug, spend an hour debugging it. I'd want to add a feature, spend two hours learning the library. I'd write 200 lines of code, realise it was fragile, and rewrite it. That's how building used to feel.

Have you had a moment like this — where something that used to take weeks suddenly took hours? Hit reply, I’d love to hear your story.

Last 2 month, I decided to rebuild it from scratch.

I opened Claude Code (an AI paired programming tool), described what I wanted in plain English, and started iterating. The tool understood my architecture. When I said "I want users to see a score out of 100," it didn't just generate code — it understood the full feature, flagged edge cases I'd missed, and suggested improvements.

Four hours later, the new version was live.

It was faster, cleaner, and covered edge cases the old version didn't. I've now got 4,000+ people using it. One of them got a job offer as an AWS Senior Cloud Engineer after using the tool.

The gap between "idea" and "shipped" just collapsed.

Here's what surprised me: AI didn't replace the builder. It removed the friction.

The thinking was still mine. I decided what the tool should score. I chose which CV sections matter most. I built the release strategy (landing page, lead magnet integration, LinkedIn posts). Those decisions are architectural — they're human work.

What AI handled was the tedious stuff. Boilerplate code. Syntax. Debugging. "How do I parse a PDF in Python again?" Library documentation. Refactoring. The friction that used to burn three hours per feature is now gone.

Previously, the skill that mattered was "Can you code?" In 2026, the skill that matters is "Can you direct AI to build what you need?"

That's a different muscle. It's not about typing faster or knowing every syntax rule. It's about:

- Clarity: Can you describe your problem so clearly that an AI understands it?

- Critique: Can you look at what the AI built and spot the gap?

- Direction: Can you steer it toward what actually works?

If you can do those three things, you can build faster than most software engineers who are still writing code line by line.

Here's how to start building with AI this week.

Pick a tiny problem.

Don't rebuild your whole product. Pick something small that's annoyed you. "I want a script that automatically logs my exercise." "I want a web form that calculates how much I'd earn with a 10% salary raise." "I want a Chrome extension that highlights keywords in PDFs."

Pick something that would take 2–4 hours to build the old way. That's your proof of concept.

Choose one AI tool.

If you're building a website or web app: Claude Code (pairs with you in an IDE). If you're writing a script or backend logic: Claude API (via the web interface). If you want a more interactive editor: Cursor (AI-native code editor). For this exercise, start with Claude Code — it's the most beginner-friendly.

Describe what you want in plain English.

Don't write code. Write like you're explaining it to a junior developer.

Good: "I want a form that takes a user's current salary and shows how much they'd earn if they got a 10%, 15%, and 20% raise. Show it in a clean table."

Bad: "Build a React component."

The more specific you are about the experience (what the user sees, what happens when they click), the better the AI will build it.

Iterate — AI gets it 70% right, you do the last 30%.

The AI won't nail it on the first try. That's normal. You'll see something and think "that's not quite right" or "I want this to look different." Instead of saying "Fix this," say what you actually want: "Make the form mobile-friendly so the table stacks on small screens" or "Change the green colour to navy blue."

Each iteration takes minutes, not hours.

Example you can try this week:

Create a "Salary Calculator" tool in Claude Code:

1. User enters current salary

2. Shows 5%, 10%, 15%, 20% raise amounts

3. Shows the yearly and monthly impact

4. Design it to look professional (you can deploy it for free on Vercel in 10 minutes)

That's a real project. It's tiny enough to finish in an afternoon. It's impressive enough to share with your network. And it proves that you can build in the AI age.

If you want to explore this angle deeper, I've built an AI Coding Toolkit I put together — comparison of five AI tools I actually use, plus a breakdown of which tool works best for what (scripts, web apps, mobile, etc.).

Grab it here: sholastechnotes.com/ai-coding-toolkit

It covers Claude Code, Cursor, Kiro, and a few others.

Here's my question for you: What's the first thing you'd build if you could ship it in a weekend?

Hit reply and tell me. I read every reply, and I love seeing what people want to create.

If you're thinking bigger — "I want to rebuild my whole career around AI," or "I want to build a product that solves a real problem" — let's talk. I offer paid 1:1 calls where we map out a plan specific to your goal.

Mentorship: mentorship.sholastechnotes.com