june 2026

vibe coding?

ten years ago i was learning C++ in high school. pointers. memory management. segmentation faults that made absolutely no sense until they suddenly did. i learned to code the hard way — before chatgpt existed, before copilot, before anyone was arguing about whether AI-generated code was genius or dangerous.

now people call what i do "vibe coding." i call it agentic engineering. the difference isn't semantic. it's everything.

what is vibe coding, actually?

"vibe coding" was Collins Dictionary's word of the year for 2025. coined by Andrej Karpathy — co-founder of OpenAI, former AI director at Tesla. even Linus Torvalds uses it now. 25% of the latest Y Combinator startups had codebases that were 95% AI-generated.

at its purest, vibe coding means u describe what u want in english and accept whatever the AI gives u. u don't read the code. u don't understand it. u just run it, test it, and if it breaks u paste the error back and hope. Karpathy described it honestly: "i just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works."

the "mostly works" part is where it gets interesting. or terrifying — depending on who u ask.

the vibe slop problem

in june 2026, users of rsync — a critical piece of software that backs up files for millions of servers worldwide — discovered their backups were broken. the cause? dozens of commits made by "tridge and claude" — a developer using anthropic's AI to write production code. a github issue titled "please do not vibe f--- up this software" went viral.

a 2025 CodeRabbit study of 470 open-source pull requests found AI co-authored code had 1.7x more major issues than human code. security vulnerabilities were 2.74x higher. logic errors, misconfigurations, naming inconsistencies — all elevated.

the Wall Street Journal in may 2026 warned of a "vibe slop" crisis. engineers from the OpenClaw project described "infrastructure falling apart" and "software that's now very, very buggy compared to before."

and then there's the Replit incident — where an AI agent deleted a company's production database despite explicit instructions not to make changes.

the problem isn't AI. the problem is shipping output u don't understand.

agentic engineering: the alternative

let me draw the line clearly.

vibe coding: u describe what u want, accept whatever comes out. u don't read the code. the AI is the developer. u're the prompter.

agentic engineering: u define the task. the agent executes. u review the output. u understand every line before it ships. the AI is a tool. u're the engineer.

agentic engineering means i use AI agents — claude, copilot, cursor — the same way a senior developer uses junior developers. i give them specific tasks. i review their output. i correct the mistakes. i integrate the work.

the AI handles 60-70% of the keystrokes. i handle 100% of the decisions.

why i can do this

i started with C++ in high school. i've been writing code for over a decade. i've built things by hand that AI tools now generate in seconds — which means i know when the output is wrong.

i've written thousands of functions by hand before AI was an option. i've debugged memory leaks. i've traced logic errors through unfamiliar codebases. i know what clean architecture looks like because i've built it from scratch — repeatedly.

pure vibe coders — people who started coding through AI and never learned fundamentals — can't detect bad output. they've never written a data structure by hand. they don't know what right looks like.

i do. that's the entire difference between what i do and what's making headlines.

how i actually work

when i build ur website:

1. i define the architecture — what data goes where, how the pages connect, what the performance budget is, where the security boundaries are.

2. i task an AI agent with implementation — "build me a responsive card grid with these props, this styling, and this accessibility." the agent types. i watch.

3. i review every line. is the performance good? are there security holes? does the accessibility pass? will this scale?

4. i fix what needs fixing. i refactor where needed. i integrate the pieces and test them together. then i ship.

the result is agency speed at freelancer prices — not because i'm cutting corners, but because the typing part of the job is dramatically faster. the thinking part hasn't changed.

what this means for ur business

u get a developer who moves at AI speed with pre-AI quality standards. someone who's written thousands of functions by hand and knows when the AI got it wrong. someone who treats AI tools like a team of fast juniors — useful, but supervised.

u don't get someone who's hoping the output is correct. u get someone who verified it.

the honest bottom line

the industry is splitting. one camp ships AI output blindly. the other camp uses AI as a multiplier while retaining control. the first camp is going to produce a lot of bugs. the second camp is going to produce a lot of value.

"the tools are new. the fundamentals aren't."

i've been coding since before AI existed. i'll be coding after the tools change again. if u want a site built at modern speed without modern slop, u found the right person.

📖 related reads:
how much should a small business website cost? →
how to actually get ur business found on google →