If you happen to’ve ever used a 3D printer, it’s possible you’ll recall the wondrous feeling once you first printed one thing you might have by no means sculpted or constructed your self. Obtain a mannequin file, load some plastic filament, push a button, and virtually like magic, a three-dimensional object seems. However the end result isn’t polished and prepared for mass manufacturing, and making a novel form requires extra expertise than simply pushing a button. Apparently, at this time’s AI coding agents really feel a lot the identical method.
Since November, I’ve used Claude Code and Claude Opus 4.5 by way of a private Claude Max account to extensively experiment with AI-assisted software program growth (I’ve additionally used OpenAI’s Codex in an identical method, although not as continuously). Fifty initiatives later, I’ll be frank: I’ve not had this a lot enjoyable with a pc since I realized BASIC on my Apple II Plus once I was 9 years outdated. This opinion comes not as an endorsement however as private expertise: I voluntarily undertook this mission, and I paid out of pocket for each OpenAI and Anthropic’s premium AI plans.
All through my life, I’ve dabbled in programming as a utilitarian coder, writing small instruments or scripts when wanted. In my net growth profession, I wrote some small instruments from scratch, however I primarily modified different folks’s code for my wants. Since 1990, I’ve programmed in BASIC, C, Visible Primary, PHP, ASP, Perl, Python, Ruby, MUSHcode, and a few others. I’m not an skilled in any of those languages—I realized simply sufficient to get the job carried out. I’ve developed my very own passion video games through the years utilizing BASIC, Torque Sport Engine, and Godot, so I’ve some concept of what makes a great structure for a modular program that may be expanded over time.
Claude Code, Codex, and Google’s Gemini CLI, can seemingly carry out software program miracles on a small scale. They will spit out flashy prototypes of straightforward purposes, consumer interfaces, and even video games, however solely so long as they borrow patterns from their coaching information. Very similar to a 3D printer, doing production-level work takes way more effort. Creating sturdy manufacturing code, managing a posh mission, or crafting one thing actually novel nonetheless requires expertise, endurance, and talent past what at this time’s AI brokers can present on their very own.
