Look, I'm going to be straight with you. I use AI to code. A lot. And if you're not, you're basically showing up to a gunfight with a knife.
Let's Talk About the Elephant in the Room
I've been building stuff for over 14 years. Started when jQuery was hot shit and "responsive design" was revolutionary. And you know what? In all that time, nothing has changed my workflow as much as AI has in the last year.
I used to spend days building a feature. Now? Hours. Sometimes less.
But here's the thing - there's a massive difference between using AI the right way and just... vibing your way through code.
The "Vibe Coding" Epidemic
You've seen it. Hell, maybe you've done it. I definitely tried it when ChatGPT first came out.
"Hey ChatGPT, build me a payment system."
Copy. Paste. Ship it.
Then your production database gets wiped because you didn't understand the SQL injection vulnerability in line 47. Or your Stripe webhooks process the same payment 5 times because you had no idea what idempotency means.
I watched a junior developer literally copy-paste an entire authentication system from ChatGPT last month. When I asked him what JWT refresh tokens were, he just stared at me. That's vibe coding. That's how you build toys, not products.
How I Actually Use AI (And Ship Real Products)
So I was building Elithr's payment system last month. Real money, real users, can't fuck this up.
Here's what actually happened:
First, I sketched out the architecture. Old school. Pen and paper. What needs to happen when someone subscribes? How do credits work? What about refunds?
Then I opened up Claude (yes, I use Claude for architecture stuff - fight me) and said: "I need a Stripe webhook handler in Node.js. It needs to handle subscription.created, subscription.updated, and subscription.deleted events. TypeScript, with proper error handling."
It spat out 200 lines of code. Pretty good code, actually.
But here's where most people stop. They see it works in dev and ship it.
Not me. I went through every damn line. Found three issues:
- No idempotency key handling (webhooks can fire multiple times)
- No signature verification (anyone could fake a webhook)
- Database writes weren't wrapped in transactions (hello race conditions)
Fixed all of it. Had AI write the tests. But I wrote the edge cases myself because AI always misses the weird shit. Like what happens when Stripe sends a webhook for a customer that doesn't exist in your database yet? (It happens. Trust me.)
What Actually Changed My Workflow
Look, I'm not going to give you the typical "here are my favorite AI tools" bullshit. Everyone has their preferences. Some people swear by ChatGPT, others love Claude. Some use fancy IDEs with AI built in, others just copy-paste into a browser.
The tools don't matter. What matters is the mindset shift.
I used to be precious about my code. Every function had to be hand-crafted. Every component built from scratch. I'd spend hours making sure my code was "elegant."
Now? I let AI write the boring shit. The CRUD operations. The form validations. The API endpoints that all look the same anyway. And you know what? That code works just as well as my hand-crafted artisanal bullshit. Sometimes better.
The real change is I stopped thinking of AI as "cheating" and started thinking of it as what it really is - a ridiculously powerful tool that makes me ship faster.
The Shit AI Can't Do (And Never Will)
But here's what people don't get. AI is terrible at the stuff that actually matters.
Last week, a founder asked me: "Should we use WebSockets or polling for our real-time features?"
ChatGPT would give you a technically correct answer about latency and server load. But it wouldn't know that this specific founder has one junior developer, a $500/month AWS budget, and needs to ship in 2 weeks. So I told him: use polling. It's good enough, easier to debug, and his junior dev won't fuck it up.
That's product thinking. That's experience. That's what AI can't do.
AI doesn't know your users are mostly on shitty Android phones with 3G connections. AI doesn't know your startup runway is 6 months and you need to ship NOW, not build the perfect architecture. AI doesn't know your co-founder hates TypeScript and will revolt if you use it.
Why This Actually Matters (For Real)
Look, I've built a lot of shit over the years. Apps that nobody used. Platforms that died after 3 months. But also some stuff that worked. Elithr. ClassCrafter. That SERHANT app that helped them raise $45 million.
You know what the difference was between the failures and the successes?
Speed.
The failed projects took too long. By the time we shipped, the market had moved. The competitor had launched. The VC money ran out.
With AI, I can build an MVP in a week. A real, working MVP that users can actually use. Not some Figma prototype. Not a "coming soon" page. A real fucking product.
Last month, I built a whole video upload feature for GoViral in 3 days. Three. Days. Pre-AI me would have spent 3 days just setting up the AWS S3 buckets and trying to figure out why the CORS headers weren't working.
The Uncomfortable Truth
Here's what nobody wants to say out loud: If you're not using AI, you're going to lose.
Not tomorrow. Not next year. Right now.
While you're writing your 500th Redux boilerplate file, someone with Cursor and Claude is shipping features. While you're debugging that authentication flow for the 10th time, someone's using a battle-tested AI-generated auth system and focusing on what actually matters - the product.
I know developers who refuse to use AI out of principle. "Real developers write their own code," they say. Cool. Real developers also used to write assembly. How'd that work out?
But Here's the Thing...
I'm not saying become a copy-paste developer. That's not what this is about.
I still architect every system. I still make every important decision. I still review every line of code before it hits production. I still debug the weird edge cases at 2 AM when everything's on fire.
The difference is, I don't waste time on the boring shit anymore. I don't spend hours writing CRUD operations. I don't manually write test files. I don't format JSON responses by hand.
AI does that. I do the thinking.
My Actual Day Now
Let me tell you what my actual day looks like now:
9 AM: Founder messages me. "We need a feature where creators can schedule their content posts."
9:15 AM: I sketch the flow on paper. When can they schedule? How far in advance? What about timezones? What happens if they delete scheduled content?
9:30 AM: Open Cursor. "Create a scheduling system for content posts. Next.js API routes, Supabase for storage, CRON job for publishing."
9:45 AM: Basic system is working in dev.
10 AM - 12 PM: Adding the stuff AI missed. Timezone handling (AI always fucks this up). What happens when a scheduled post fails? Email notifications. Admin overrides. The actual product stuff.
2 PM: Feature is live in staging.
3 PM: Founder is testing it. "Can we add recurring posts?"
4 PM: Recurring posts are live.
That's not an exaggeration. That's literally what happened last Tuesday.
The Real Talk
If you're a founder reading this, here's what you need to know: I can build your product faster than anyone who's not using AI. Period.
But I'm not cheaper. I charge more, actually. Because I deliver more. You're not paying for my time. You're paying for velocity. You're paying to beat your competition to market.
If you're an engineer reading this and you're scared AI is going to take your job - you're right. But only if you refuse to use it. The engineers who embrace AI aren't going anywhere. We're just getting more valuable.
One Last Thing
I still remember my first website. Built it in 2010. Took me 3 weeks to add a contact form. Last week, I added real-time messaging to an app in 2 hours.
That's not because I'm a better developer now (though I hope I am). It's because the tools are insane. And AI is the most insane tool we've ever had.
Use it. Or don't. But don't come crying when someone younger, hungrier, and AI-powered takes your job.
Currently shipping features daily on GoViral that would have taken me weeks just two years ago. Looking for founding engineer roles where speed and quality both matter. Not interested in "we don't use AI here" companies - that's like saying "we don't use Google" in 2024.
Hit me up if you want to build fast: peter@bricksandcanvas.com