Is Lovable getting monetization wrong?

I'm a non-tech worker in a non-tech industry, let me state two things:

- Software today is written to cover as many use cases with as many features to target as many users a possible.

- End users very often only use a tiny slice of the program's capabilities, but still pay for the entire program.

This creates a situation where the people writing software see it as a monumental undertaking to get good functional programs (it is), and end users see programs as having annoying learning curves with lots of bloat and "unnecessary" features.

LLMs do an excellent job of fixing this for end users because it allows them to easily create a program that does the handful of tasks that they normally need to use MegaSoftware for. And it's tailor made exactly for the use case. And the LLM can tell you exactly how to use it.

I can give a brief example where I used gemini to create a CAD file transposition tool that utilized a simple GUI tailor made for the files my company works with. This allowed us to forgo a (very) expensive CAD software package to work through converting our archive of files. A probably 2M LOC program could be skipped because we only needed 3k LOC functionality.

I really cannot stress enough how often this is the case, and why SWEs see LLMs as weak tools while end users see them as gods.

There will still be a need for huge software packages in the future, but I know I never again have to pay for a huge class of "here is a large solution space that covers your small scope problem" software.

To bring it home, loveable understands this, an sees that the futures has lots of non-tech people "writing" software. Standard IDEs are not the tools your mom will use to make a "Friends and family birthday reminder" app.

End users rarely pay for the program. Someone in their management chain OKs the purchase, or there's a larger purchase with a cross-charge to the department for the license cost. the problem comes when software needs to meet every whim of the decision maker, when really the users only will ever use 20% at best.

I agree with you, and I think the Jevons paradox will eventually manifest itself once again. How many smaller companies are stuck with outdated workflows and tools because they can't afford to pay ten engineers for months on end for something better?

Now those companies may very well be able to afford one engineer and some AI subscription to do the equivalent work.

I think you may be assuming a certain enterprise size and accompanying workflow, probably would need stats to actually know how much of software is bought in this way however, and if the other way described would open up possible purchases by smaller companies, as was claimed.

Good point. "Build vs buy" is a perennial controversy

https://www.thoughtworks.com/content/dam/thoughtworks/docume...

I don't disagree with the thrust, but I've recently cleaned some of those up.

One example: LLMs aren't smart enough to do things like properly manage zip codes with leading zeros. It was round tripping strings through an integer representation and corrupting them. The users did notice, but did not have the vocabulary/concepts to explain. To them, sometimes zipcodes get corrupted because inscrutable reasons (tm).

chatgpt also authored a bash script that would have blown away a chunk of my drive if any paths had a space in them. :shrug:

So true. Some of my favorite enterprise software I use often I could never afford or would never purchase for my own use. I've taken jobs because of this.

IDA?

> I'm a non-tech worker in a non-tech industry…

Certainly you must have enough detailed knowledge of CAD files to validate the output of the transposition tool you had AI create for you. This might not be enough for you to think of yourself as ”technical” but I’d argue that it’s far above the level of “entry level employee using CAD”.

This does also seem to fit the paradigm of “AI is a productivity booster for people who already know how to do x”

I think there was a consensus over the past decade, but we are now having to adjust our priors. The answer is changing month by month. It also means people are delaying their decision because they are afraid of the wrong solution right now except in the most obvious cases.

I don't deny that there's utility in what you are describing: if you can make it work for you that's fantastic. However:

- if you can ship software like that given the current state of the technology, you are probably not the average non-tech worker in a non-tech industry. There are people paying exorbitant consulting rates for dashboards in PowerBI. LLMs in mid 2025 are orders of magnitude more operationally complex than anything most people have seen.

- "citizen developers" doing something to scratch their own itches sounds very much like how a professional software project starts. Suddenly the scope grows and you need a nerd to handle it. Then two. Then four. You get the idea. Maybe that won't be the case for your specific needs, but that's how it generally goes.

Weak or strong is a matter of framing, but that's why I see them as tools and not gods.

Fun thing I noticed is converting a CSV to XLSX in Excel also drops leading 0s from zip codes...

Build vs buy vs vibe?

I think users generally have the expertise to evaluate the quality of output for their own problems

You're absolutely right and one has to judge the HN consensus on AI tools through a lens of bias of the very demographic being challenged by AI tools.

Most spreadsheet software does that because the zips are parsed as integers, not strings . Prefix numerical fields that may contain leading zeros with a non-numerical character/word and save yourself the pain of late discovery! You can just find and replace after it goes in successfully.

Spoken as someone that experienced the same pain years ago in local SEO data that made it way too far before discovery...

You need to select the correct column data format in the Text Import Wizard: https://filestore.community.support.microsoft.com/api/images...