Like most of people nowadays, I am playing around with any AI tool I can get my hands on.
I try, as often as possible, to at least give most tools a rudimentary look.
A few that have stuck out lately:
Cursor powers my day-to-day.
ChatGPT is my processing buddy.
Sora is wild (honestly can’t believe this is a real product).
Napkin.ai is incredibly useful.
Stickerbox might be the best hardware to come out of the AI boom.
These tools truly have been fun to learn, use, and play with.
However, there is another category of tools, that are more frustrating than fun.
I sense that it’s due to where we are in this cycle.
These are the “vibe-coding” tools.
The tools that promise to turn a solopreneur into a 50 person development team for $19/month.
Don’t get me wrong, the fact that we can create a recipe tracker app in 15 minutes from natural language is wild.
Just setting up the plumbing for that in 2018 would have taken a week.
But are 1k “credits” going to get any meaningful work done?
In fact, what is a “credit”?
Is it a token?
Is it an API call?
Is it a prompt?
Is it a GPU clock cycle?
And here lies the issue.
When a pricing model is built around “credits”, the customer needs to know what a credit corresponds to, and trust that the seller counts these credits correctly.
I’ve seen statements like this on a lot of these tools’ pricing pages:
“A typical prompt can be anywhere from 100-500 credits.”
Okay, so, I’ll try to make my input smaller, and more targeted; that should lower the credit consumption.
Well, that might require the model to “think” more, so it uses more credits.
Okay, well, what if I describe clearly what I need, and also define the scope of changes, to ensure the model doesn’t go off on a tangent.
Then your prompt has too many tokens and more credits are consumed.
Who’s even to say these tools don’t inflate credit usage by doing a creditsCharged = usedCredits * 1.5?
There is no transparency. There is no way to know.
Ahhh!
Don’t even get me started on being charged when the model messes up.
The other day, while playing with one of these tools (which I will not name), I asked it to change up the order it made two network calls on the home page.
The model went rogue, deleting two unrelated features, all while ignoring my original request.
This rebellion pushed my usage over the monthly credit quota, and I was unable to make any more changes without swiping my Visa.
Talk about vendor lock-in.
Imagine that I hire an electrician to fix a faulty outlet.
But they instead add a new light to my kitchen ceiling, somewhere that I did not want a light.
Then they charge me for the light, and say they cannot fix the outlet because their schedule is full.
Insanity.
We need to get to a point where these tools align with the outcomes they are enabling.
Currently, their incentive is to ensure users consume more and more credits.
Any improvements to their credit usage efficiency theoretically lowers their revenue.
A pricing model that aligns successful app publishes and maintenance to overall cost is a sustainable and trustworthy model.
This is what we should strive for.
That is, if any of these tools even make it to their next funding round.