TNS
VOXPOP
Favorite Social Media Timesink
When you take a break from work, where are you going?
Instagram/Facebook
0%
Discord/Slack
0%
LinkedIn
0%
Video clips on TikTok/YouTube
0%
X, Bluesky, Mastodon et al...
0%
Web surfing
0%
I do not get distracted by petty amusements
0%
AI / Frontend Development

The Promise of Riches from AI Wrappers

David Eastman shows how to create an AI wrapper app based on the OpenAI API, after being inspired by YouTube videos of AI side hustles.
Dec 18th, 2023 6:19am by
Featued image for: The Promise of Riches from AI Wrappers
Image via Wes Roth on YouTube

Whatever you decide to watch, YouTube will eventually place on your feed one of those videos that feature outsize claims in the title. Like a carnival barker at a circus, they announce the amazing things the video covers, only for those to be downplayed as the video goes on. One that took my attention screamed “AI Apps Making $20,000+ per month with 1 person teams.” Ignoring the uncorroborated incomes, the video was fairly sane, pointing out that most of the ideas are just AI wrappers, and didn’t really require a great deal of work to get started. I’ll look at some of the featured sites below.

While this post has a completely different aim to the barker, I do agree that Artificial Intelligence provides new opportunities for apps that satisfy niche requirements. If you are a single dev or a small dev team, creating a small MVP with AI is a good idea — whether for a skunkworks challenge or for an internal issue within your organization. I’m partly feeling the tech startup vibe of 15 years ago in the air, but perhaps AI is poised to shift from “excitement” to “deployment” in 2024. While cashing out for the big jackpot is far from impossible, I’d encourage you to just settle with scratching an itch and practice getting a product out. You can start it as a “side hustle” if you suddenly get the ambition to work on something.

While the entrepreneurial spirit is about spotting a hole in the market and exploiting it before the big guys notice and release a free version, the virtuous developer should focus on the customer journey. But let’s start at the beginning.

So What Is an AI Wrapper?

What AI apps build on, or wrap, is the basic API call to an LLM middleman. To get the basic API experience, we don’t need any code:

    1. Join OpenAI. This is now the McDonald’s of AI, so let’s start here.
    2. Create yourself an API key. This identifies you and your requests. I have used the clearly false key xx-xxxxXX in my example query below.
    3. Pay as little as $5 to buy some compute. Let’s face it, Sam Altman has provided at least that much entertainment recently.
    4. Open a terminal (i.e. a command line). I have used Warp as my terminal, so it is easy to separate command from response. Use the example curl command that OpenAI gives to call their basic GPT models. Here is my query:

      Think of this as a standard REST call with a JSON payload. The important bit is to put your long API key in the place of xx-xxxxXX. You can use a different model if you like, but that might cost more. Note that our content is the question “What is TheNewStack?”. The temperature controls the creativity of the response. We want some life in the description, but no hallucinations.
    5. Here is the response:

      The prompt tokens are not literally the number of words in the question, but they are related to how the query is broken down. As you can see, the answer is a reasonably good summary. A quick look at the usage reveals that this has cost less than a cent. If you get the query wrong, read the error code carefully.

The Carnival Barker Examples

Now that we have seen the minimum for what an AI backend must do, let’s look at some working sites.

First up is Formula bot, which helps users with Excel formulas — or “AI-powered formula generation”. You might well argue that if spreadsheets are still hard to use, then surely the problem is with that tool, but this ignores reality. Spreadsheets are all over the place, often in the hands of people who have no interest in numbers or formulae whatsoever.

After a short sign-up, it offers you some test market table data to help you see how things work. After looking at the test data and asking “Which beverage product makes most money?”, a green line creeps very slowly from left to right and eventually the right answer is arrived at. The steps are revealed, including Python code that failed due to not having the “necessary pandas”. Why Python needs pandas is not clear to me.

While far from smooth, the time from signing up to getting a solution was quick. As usual, there are various upgrade options to the paid product. There is even a regex finder on the page — a sign the page has sprawled out to offer more services.

Next up is PhotoAI, used to “Create beautiful AI photos without using a camera.” The continuous scrolling of beautiful people — all created from selfies — is a perfect advert. This is followed by a simple diagram to show how 4 average selfies can be used to generate a full head and body shot in a cool backdrop. Again, it mentions video generation — once an audience comes, you can sell more. It also shows a remarkably honest advert; and no doubt this is exactly the case.

Finally, we have PDF AI. Apparently the domain “pdf.ai” cost $10k, so this does underline the need to ensure the site matches the user’s expectations, as well as reminding us that successful efforts are far from casual. All the examples match the website domain closely with the task.

When you are trying to buy a house, a lot of documentation will come your way; and the ability to extract information hidden in them is certainly valuable. There are many times in life when you might need a service intensively for a short period of time, and a small cost is less of a problem than hoping you can.

The Approach to Developing AI Apps

If you are a seasoned developer, you can take your good habits straight to making the Minimum Viable Product (MVP). This will likely be less of a technical challenge and more of a routine discipline challenge. Start by testing the steps out by hand; from a user query, putting the query into the AI, and isolating the response. Spot the types of questions and answers you see.

There are now plenty of AI platform helpers, like Fixie, which offer alternatives to using OpenAI’s API directly, like we did above.

Certainly collecting money from eager users is now much simpler with solutions like Stripe. Always try to start a limited company (depending on the law in your region) to isolate your product financially.

Make sure you can use code to convert the required API responses into user HTML pages. Use whichever web solution you are most comfortable with. For most developers, this will be JavaScript-driven, but you use development tools like Blazer if want to avoid JavaScript directly. The website can live on your laptop for now.

Once you get a local build and the code into the simplest source control, you just need to automate as much as possible. Most of these topics are covered in my “How to start a project” post.

Getting a public site up as soon as possible should be the aim of your first MVP. From then on, it is just continuous iteration. You are now working outwards to inwards — not focusing on what you put out, but what others see and respond with, thus paving the customer journey.

If you are scratching an itch, you will find places like Reddit where people are asking “How can I..?” about the very thing you have answered with your MVP. From there it is a matter of marketing effort. If you are hunting riches, I wish you good luck.

Group Created with Sketch.
THE NEW STACK UPDATE A newsletter digest of the week’s most important stories & analyses.