AI Coding: From Toy to Essential

One Million Users in Five Days

On November 30, 2022, OpenAI released ChatGPT. It hit one million users in five days. Instagram took 2.5 months to reach that number. Netflix took 3.5 years.

Reactions split into exactly two camps. "Coding is over" and "It's a toy." Both were wrong. Coding didn't end, and it was far from a toy.

Over the next three years, public perception shifted four times. This post maps that shift.

Phase 1: "It's a Toy" (2022--2023)

Ask ChatGPT to write code and something came out. Simple functions, HTML pages, Python scripts. People got to "Oh, that's cool" but never to "I'd use this for work."

The reason was obvious. Too much of the code was wrong. It produced confident-looking code that threw errors on execution. It invented variable names, called nonexistent APIs, and got logic subtly wrong. Developers gave it a nickname: "the overconfident intern."

GitHub Copilot was already out in 2022. It suggested the next line in gray as you typed. Developers called it "slightly better autocomplete." More of a convenience feature than a real tool.

The defining perception of this phase: curiosity plus dismissal. Interesting, but useless for real work.

Phase 2: "Is It Coming for Our Jobs?" (2023--2024)

In March 2023, GPT-4 arrived. Its coding ability jumped dramatically. It passed the US bar exam in the top 10% and breezed through coding interview problems. The people who had called it a toy went quiet.

Fear set in. "Junior developers won't be needed anymore." "Half of all programmers will disappear within five years." Every headline was about AI taking jobs.

Counterarguments existed. "AI can't replace senior engineers." "Design and decision-making are human domains." These were valid points, but they weren't enough to calm the anxiety.

Here's the interesting part: during this same period, the number of companies actually adopting AI coding tools surged. People said they were scared, then started using it anyway.

The defining perception of this phase: anxiety plus denial. People felt the threat but refused to accept reality.

Phase 3: "Fall Behind if You Don't Use It" (2024--2025)

The turning point was Cursor. In 2024, AI moved beyond suggesting code to directly editing files. Say "Add dark mode to this component" and it modified multiple files at once. Cursor hit $500 million in annual revenue within two years.

The hiring market moved too. LinkedIn and Indeed job postings started listing "proficiency with AI tools" as a preferred qualification. AI coding was no longer a hobby. It was a hiring criterion.

On February 2, 2025, Andrej Karpathy -- former AI director at Tesla -- posted a tweet: "I call this vibe coding." It crossed 4.5 million views, gave AI coding a name, and landed on the Collins Dictionary Word of the Year shortlist.

Period	Key Event	Public Reaction
2022--2023	ChatGPT, early Copilot	"Cool, but no real use for it"
2023--2024	GPT-4, coding ability skyrockets	"Is it coming for our jobs?"
2024--2025	Cursor explodes, Karpathy tweet	"Fall behind if you don't use it"
2025--2026	85% developer adoption, Apple integration	"It's just the default now"

The defining perception of this phase: acceptance plus adaptation. No way to resist, so people started learning.

Phase 4: "It's Just the Default Now" (2025--2026)

The numbers say everything.

Metric	Figure
Developers using AI coding tools	85%
GitHub Copilot cumulative users	20 million
Cursor annual revenue	$500M (in 2 years)
Non-developer share of Lovable users	63%
YC startups with 95%+ AI-generated code	25%

In February 2026, Apple integrated Anthropic's Claude Agent and OpenAI's Codex natively into Xcode 26.3. Apple makes its own chips. It makes its own OS. But for AI coding tools, it went with someone else's. The reasoning: shipping proven tools delivers value to users faster than building from scratch.

This is not a technology story. It is a perception story. AI coding has passed the experimental phase and become a standard. Apple putting it in Xcode is a declaration -- the question is no longer "should I use this?" but "how should I use this?"

The non-developer market opened up too. Lovable hit $100 million in annual revenue eight months after launch, and 63% of its users have zero coding experience. "Build apps without knowing how to code" stopped being a marketing slogan and became a fact.

The defining perception of this phase: not optional, but essential. Not using it is what requires an explanation now.

Right Now Is the Easiest Time to Start

"Am I too late?" is a question that comes up constantly. The answer is the opposite. Right now is the easiest time to start.

In 2022, getting into AI coding meant studying prompt engineering separately and fixing errors manually. In 2024, Cursor existed but setup was complicated. In 2026, you type one line in the terminal and a blog gets built. You say one sentence in a browser and an app comes out.

Tools get better every year. The barrier to entry drops every year. Starting a year from now might be even easier, but by then everyone around you will already be using it.

What people called a toy became essential in three years. The next three years will move faster. The best time to start is now.

In the next post, I compare the tools you should actually use: Cursor, Claude Code, and Copilot.

Sources: