Comment by Lauris100
1 day ago
“The only way to go fast, is to go well.” Robert C. Martin
Maybe spaghetti code delivers value as quickly as possible in the short term, but there is a risk that it will catch up in the long term - hard to add features, slow iterations - ultimately losing customers, revenue and growth.
Anecdotally I'm already seeing this on a small scale. People who vibe coded a prototype to 1 mil ARR are realizing that the velocity came at the cost of immense technical debt. The code has reached a point where it is essentially unmaintainable and the interest payments on that technical debt are too expensive. I think there's going to be a lot of money to be made over the next few years un-fucking these sort of things so these new companies can continue to scale.
So basically the new version of the 1990's people's projects that grew to high ARR based on their random Visual Basic codebase? That's how software companies have been starting for 30 years.
Time is a flat circle and what is old is new again.
if i have 1mil ARR, i can hire some devs to remake my product from scratch. and use the Vibecoded Example as a design mockup.
If i manage to vibecode something alone that takes off, even without technical expertise, then you validated the AI usecase...
Before Claude i had to make a paper prototype or a figma, now i can make Slop that looks and somehow functions the way i want. i can make preliminary tests, and even get to some proof of concept. in some cases even 1million $ annual revenue...
Yes, this is exactly where AI shines: PoCs and validating ideas. The problems come when you're ready to scale. And the "I can hire some devs to remake my product from scratch" part is the exact money making scenario some of my consulting friends are starting to see take shape in the market.
5 replies →
I guess that depends on how you get that ARR-figure. If more than all of it goes to paying your AI bills, then you can't really afford that much engineering investment.
> hire some devs
you're making an assumption these devs you hire actually know what they're doing and not just a proxy back to an LLM.
> if i have 1mil ARR, i can hire some devs to remake my product from scratch
This assumes a pool of available devs who haven't already drunk the Koolaid.
To put it another way: the 2nd wave of devs will also vibe code. Or 'focus on the happy path'. Or the 'MVP', whatever it's called these days.
From their point of view, it will be faster and cheaper to get v2 out sooner, and 'polish' it later.
Does anyone in charge actually know what 'building it right' actually means? Is it in their vocabulary to say those words?
You would only be able to hire me to do that job if you gave me every last dollar of that ARR. And I still might turn you down tbh..
By then, the startup will have folded, and the C-levels will have moved on to the next Idée Du Jour.
This is true, but what I've come to realize is companies only prioritize the short term, no matter what, no exceptions. They take everything on as debt.
They don't care about losing customers 10 years later because they're optimizing for next quarter. But they do that every quarter.
Does this eventually blow up? Uh, yeah, big time. Look at GE, Intel, Xerox, IBM, you name it.
But you can get shockingly far only thinking about tomorrow over and over again. Sometimes, like, 100 years far. Well by then we're all dead anyway so who cares.
Or, you can be like many modern CTOs: AI will likely get better and eventually be capable of mostly cleaning up its own mess today. In which case, YOLO - your startup dies, or AI is sufficiently advanced enough by the time it succeeds. The objections about quality only matter if you think it’s going to plateau.
That is, literally, faith-based business management. "We suck, sure - but wait, a miracle will SURELY happen in version 5. Or 6. Or 789. It will happen eventually, have faith and shovel money our way."
If the AI gets that good, what value does your startup add?
I suspect it's going to tank instead of getting better, no matter what they try to do with attention or agents or whatever, especially if it's training on AI-written code of which there will be more and more of as time goes on. I'm not an AI expert by any means, so take that with a grain of salt.