I was not an AI skeptic. I've been using AI-assistance in my dev work since 2024.

Still, I held the strong opinion that "vibe coding" could only produce throw-away prototypes.

I know the software development process inside and out. For 20+ years in roles of CTO, head of engineering, and senior software engineer, I've led teams that shipped software deployed in Google, Netflix, TED, and other fortune 500 companies.

My experience with AI coding up had me convinced that using AI to crank out code without careful human review could never produce software that could actually be deployed.

In December 2025, the CEO of the SaaS company I was working for spent a couple weeks vibe coding an entirely new product platform from scratch. He had the product vision and strong design sense, so he just started flowing with Cursor and Claude Opus.

In January he demo'd the product to the team. It was a sprawling platform with tons of features and lots of moving parts. He wanted to launch it live to get market feedback.

Everyone on the engineering team, myself included, pushed back. "This will be an absolute mess in production. There's no way 250k lines of vibe coded slop is going to be anything less than a nightmare."

The CEO agreed the code was probably horrible underneath, and so he asked us how much time it would take to rebuild the platform the right way. Even using heavy AI assistance, the estimate was many months. The timeline wasn't acceptable.

We were directed to deploy the platform for market testing immediately, and start cleaning up the code in place rather than doing a rebuild. Oh yeah, and the CEO was going to continue vibe-coding in production at a breakneck pace, iterating and extending the product as market feedback started to come in.

Everyone on the dev team predicted catastrophes: endless incidents in production; in-place clean-up would be a quagmire; etc.

How it actually played out dramatically shifted my mental models.

Yes, the code had serious flaws. We did a deep security, architecture, and infrastructure audit, and found many critical issues (including a security flaw that would have allowed a clever actor to exfiltrate all the platform secrets and credentials).

However, we were able to fix those issues in place surprisingly easily and quickly. We did some sweeping refactors to add better guardrails to keep the AI code generation on track (separating business logic, enforcing test coverage, CI guardrails, etc).

And the catastrophes never came. The system didn't blow up. The CEO's velocity wasn't impeded. The product kept evolving and pivoting in tune with the market feedback. There were no major incidents.

I don't want to mislead: there was a lot of engineering work cleaning up and solidifying the platform. But it wasn't the nightmare we all predicted. The changes needed were performed surgically (using AI, btw), all while product development kept moving forward at an incredible pace.

That experience finally made me see the truth: software development has changed forever. The process has changed. Code doesn't have to be written or even reviewed by hand any more.

Is engineering expertise still necessary? Yes, absolutely. AI gets things wrong. A lot. And sometimes those things are devastating timebombs waiting to go off.

Taking AI written code from working prototype to successfully launched product is still a highly technical challenge. There are a lot of gotchas in that last mile. Engineering expertise is still invaluable in navigating that part of the journey. An experienced engineer will know the questions to ask and the threads to pull to sniff out the gaps hidden in the code, the scaling and performance problems waiting to show up when the marketing kicks in and the traffic gets real.

So yeah, software development has changed. But engineers aren't obsolete.

Personally, I'm excited by this. Now anyone with vision and domain-expertise can speak software solutions into existence in a way never before possible. Will the solutions be perfect and ready to launch out of the box? No. But with a little expert guidance, they can be solidified and successfully launched. I've seen and proven it first hand.

There are so many people with good ideas that can now bring them to life. As a software engineering veteran, that's where I'm pivoting my career: from being the builder to being the expert pair of eyes helping builders go from working prototype to successful launch. To me, that feels exciting.

(from reddit)