The Promise and Pitfalls of No-Code AI
But is no-code all it's hyped up to be?
When I first started using no-code AI tools like Chat GPT Builder, to build apps, I was filled with skepticism. As an engineer, I doubted that AI could truly replicate what I do. But I was amazed by what I built in a short time just by describing my idea in text.
> I would like to build a hero + villain image generator
for kids. The app must know the user's age. The app then
tailors age-appropriate questions, one at a time, to
get information about the superhero or villain. For
each question, give 3-4 choices and leave it open-ended.
After the 5th question, suggest three cool names to the
user. After they select the hero name, generate an
image of the hero in a dynamic scene with an
age-appropriate art style.The Initial Excitement
I typed a simple prompt into ChatGPT, asking it to design a hero and villain creator for kids. What it generated was remarkable - an age-tailored questionnaire to gather details, name suggestions, and customized images. The first version felt like magic. But my excitement soon gave way to frustration.
What it made was amazing. Holy crap... It did kind of what I wanted in the first version. I was happy yet somehow I didn't feel accomplished. Did I really do anything? I just typed a bunch of text. This morning I read "A Coder Considers the Waning Days of the Craft," and was happy to see I was not the only one who felt this way. It took away the joy of coding...
Kid Create: Hero vs Villain
Allow children to turn their imagination of cool characters into amazing illustrations.
DALL-E
Not specific to builder, but... Everything with DALL-E was an issue. Don't get me wrong, it produced some remarkable images. But the first issue started with something as simple as a logo. Initially it produced some horrible logos, but with time and prompting I got something close to what I wanted. However, trying to get the AI engine to make minor tweaks was painful. Every time, the AI produced something vastly different and its comprehension seemed to fail a lot with normal language.
The AI's comprehension of detailed language requesting small changes was lacking. I found it best to just keep generating new images from different prompts rather than trying to tweak a particular image. Overall DALL-E created impressive images, but guiding it toward a very specific end result was difficult. ( All of this, while ignoring that all text it produces is more or less garbage. )
This spills over to images produces. They each vastly differ, and all the kids were very frustrated with when they got something they were 80% happy with, but only pain and agony happened after that.
Frustrations of using Ai to make app
Next was the height of frustrations, but a learning experience. You give unclear commands, you get unclear results. You need to be as specific as you can. Having the builder do minor tweaks is not good - fix one thing, it breaks another. House of cards.
The other frustration was the "app" it built was not consistent. Each time you launch the AI, it would interpret things differently. Rinse and repeat over and over and over... argh!
I was about to throw in the towel. It was so close, but I could not get it to modify in the precise ways I wanted. Then I asked a magical question, and what it produced was golden. It finally understood the core concept of my idea.
A path forward
Here was the response. It kinda knows what I want.
Based on the updated instructions, I understand that
the Hero Villain Creator app now has a more specific
process for engaging children in creating their
superhero or villain characters. Here's a
breakdown of the updated process:
Age Inquiry: The app will start by determining the user's
age. This is crucial for ensuring that the content
is age-appropriate.
Tailored Questions: The app will present age-appropriate
questions, one at a time, to gather information about
the superhero or villain the child wants to create.
Each question will have 3-4 choices, but they
will also be open-ended to encourage creativity.
Question Sequence: There will be a total of 5 questions.
The nature of these questions will be designed to
incrementally build up the character's traits and backstory.
Naming Suggestions: After the fifth question, the app will
suggest three unique and cool names for the character,
allowing the child to select one.
Image Generation: Once a name is chosen, the app will
generate an image of the hero or villain. This image will
be in a dynamic scene and created in an art style that
is appropriate for the child's age.
Focus on Age-Appropriateness: Throughout the process,
the emphasis is on ensuring that the content,
questions, choices, and images are suitable for the
child’s age group.
The goal is to provide a fun, engaging, and educational
experience, encouraging children to use their imagination
in creating their unique superhero or villain. The app
is designed to foster creativity while being
sensitive to the user's age.
I suggest you take the base prompt, go over to the Configure tab, and paste it what the GPT says it understands. From there, just modify it to do what you want. Do not go back to Create unless you have a brand new idea you want to start from scratch. Otherwise, sticking with Configure avoids frustration from the inconsistency of regenerating from scratch.
Everything looks the same
The UIs from AI builders feel generic - every app looks and behaves the same. It's like the early days of WYSIWYG editors where flexibility suffered. The ease could flood app stores with low-quality, cookie-cutter apps again. Innovation and customization are limited for now. Finding the right balance between design control and development speed will be key as the technology improves.
Is Coding Still Relevant?
What is interesting (as the New Yorker article mentions) is that you have to instruct AI in a way that is detailed yet unambiguous. It's like coding...kinda. So I think the analytical problem solving and engineering skills are still irreplaceable for now, but for how long? Algorithms, writing out specs, strawmen...all these techniques help in instructing the machine.
Basically, as engineers always do, we adapt! Do we treat prompting AI as a new language to master? The skills of breaking down problems, outlining solutions, and giving clear instructions are still vital - we just need to apply them in this new medium of "coding" the AI through prompts. Coding may look different, but many of the core engineering and problem-solving skills will still be essential. We need to evolve our skills to this new AI paradigm.
The future...
I am optimistic and excited about the future. This technology opens up app creation to more people, letting anyone bring their ideas to reality. Now, anyone can use AI to make things that solve real-world problems for users.
Up until now, only a very few could take their thoughts to the next level. This democratizes app development. However, I don't think the tech is quite there yet to make a fully functional, polished app in a repeatable way. For now, you still need to hire professionals to get your idea production-ready.
But I'm thrilled to see how this technology evolves, and what crazy creative things people will make to improve the world. While it needs more maturation, it unlocks app development for all, letting anyone's ideas come to life. I'm excited to see what people build when bringing their imagination to reality gets easier. This has potential to make the world better by empowering novel solutions to real problems.
