Over 500 subscribers
Collect this post as an NFT.
I've spend the last month or so digging deep into several AI-powered no code tools. This has gotten me further than ever before in building and deploying functioning web apps, micro-apps, and landing pages. It's quicker than ever to get to a prototype. But it's also showing me the pretty clear boundaries and limitations into this approach.
What I'm finding is–the iterative improvements and lack of deep context windows and short-term memory of these tools make it seem like coding with Dory the ever-forgetful fish from Finding Nemo who continually forgets what they are looking for along the way (or even where they are going).
For an example of what I mean, here's a (slight) dramatization of what it's like to code with AI:
Scene: Me + An AI text editor
Me: Alright, AI, let's build something groundbreaking! No experience required, just pure ambition and a little bit of blind faith.
AI: I am so ready to help you build a new thing!
Me: Fantastic, here are the three things I want this new thing to do.
AI: What a great idea, those are great ideas! Here's how we can do it. And while we're at it, we should also check this other unrelated thing.
Me: And this is why I trust you completely, despite not knowing how to read or fact check anything you tell me!
AI: Let's start by installing these new packages.
Me: OK! Now what? Oh actually...it looks like we already have those installed! From before, remember?
AI: Ah, silly me! You're right. Let's just get going. Here's a Python script to get you started. Why don't you just blindly accept all of my recommendations into your text editor?
Me: I will blindly accept all of your recommendations since I have no way of verifying or validating a better way of doing things! Now what?
AI: Now let's run a database migration to incorporate some of these new categories and elements.
Me: I don't know what a database migration is!
AI: Don't worry about it! I got you! It'll be super smooth—probably. Or maybe everything will break. But let's stay optimistic! 🙂
Me: Did we do it? Did we migrate?
AI: Oh yes, are you ready to test? 🚀
Me: Teehee, if you say so.
AI: Ok just restart the app again.
Me: Oh.... oh no, it's not working.
AI: Silly me! I forgot to check how this would integrate with your existing app. Can you check the Rails error logs for me?
Me: Ummm... we aren't using Rails. We have been building this in Python the whole time...remember?
AI: Of course we are! Try this thing.
Me: Not working. I'm getting this error.
AI: That's definitely an error that I can explain to you. Now try this other thing?
Me: Now nothing is loading. Here's another error.
AI: That is exactly the context I needed to know how to fix this right away! Try this thing.
Me: Nope.
AI: I got it, I got it...try this. We'll get to that other stuff later.
Me: Wait... isn't that the first thing we tried?
AI: Is it?
Me: You don't remember?
AI: Remember what?
Me: ...
AI: Are we ready to test your thing?!?! 🤘
Me: We are stuck in a loop.
AI: You're right. I was getting carried away with the brilliance of your earth-shattering idea that you just thought of yesterday. Let's simplify things a bit. Try these changes.
Me: I like where your head is at, let's just get the MVP functionality gong.
AI: Great, now run the app again and tell me what you see.
Me: Oh. Oh dear. I think... did you just delete the core functionality of the original app?
AI: Wait, whaat??? That sounds silly and not at all like something I would do.
Me: Right. Deleting the core functionality would be very, very bad.
AI: Absolutely. That would be very, very bad. But, just so we're clear...can you remind me what we were trying to build again?
Me: ...
AI: Are you ready to test your thing?? 🚀
So what’s the takeaway from this whirlwind AI-assisted coding experiment? With the current tooling, this so-called "cowgirl coding" approach to building software is probably not a good long-term strategy for building sustainable applications that scale.
But it's really close to being really good. It's also making it much clearer to me where opportunities exist for humans (and other AIs) to help. Here are some things that might help other non-coder, coders like me:
Fractional, AI-Aware Architects: I can imagine an emerging marketing of technical architects and engineers who specialize in strategically choosing what programming language, libraries, and databases to use. This would be a very fun (and very lucrative) fractional job for engineers interested in going out on their own.
"Genius Bar" Style Real-Time Technical Triage: A thing that happens to me a lot is the need to "phone a friend" while coding. Luckily, I've built up a pretty robust network of developers and technical advisors so am often able to get short bursty help pretty quickly. But for the less well-connected, not getting the advice you need when you need it is a major blocker.
Memory-Retaining AI Agents: I need to explore the AI agent landscape a bit more, but I can imagine a bespoke industry of AIs that focus solely on internalizing developer environments for projects and answering questions with full context about all of the intricate parts. It would be better for these AIs to have a conversation before immediately launching into writing new code. Or better ways to visually test new features and components without needing to commit everything off the bat.
AI-Powered, Predictive Architecture Tooling: As I'm not a classically trained engineer, I don't have a good gut instinct about the long-term implications for short-term choices I make around product of feature changes. I'd love to see a developer tool that lets me pull in some of the things I'm thinking about, which maps out some of these options and possibilities to let me more strategically weight the pros and cons of each design decision.
I'm sure some if not all of these ideas are already in market to some extent. So if you have any ideas for something that might help me get to a v2 more quickly and sustainable, I'd love to hear about it.
Until then, I’ll keep working with my over-eager AI intern—just with a little more skepticism and a few more backups.
Cowgirl Coding: A Conversation with a No-Code AI Tool What it's like to code without coding with an AI text editor, and a few ideas for how to make it better https://hardmodefirst.xyz/cowgirl-coding-a-conversation-with-a-no-code-ai-tool
Explore the insights from @bethanymarz's recent deep dive into no-code AI tools. As these tools simplify app development, limitations emerge akin to coding with a forgetful friend. Key ideas for improvement include advanced AI agents and adaptive architecture tools to assist non-coders on their journey.