The Existential Threat to Developers Conference
I recently attended Creatio's No-Code Days seminar in Chicago with the expectation of gathering an industry view of the No-Code phenomenon. I expected it to be marketing-heavy (and I wasn't disappointed), but I also wanted to see how No-Code tools bore out, even when multiplied by a hype machine, which is usually when I throw my hands up and walk away.
I haven't chatted too much about No-Code yet, because I haven't explored the tools too much, but I've certainly been interested.
Through a combination of my LinkedIn feed and a handful of articles, I've decided to look more into the commercial offerings from Notion and Bubble. I'll bypass Creatio because their offering is more B2B-centric and, in the short term, I'm more apt to focus on consumer-facing tools when thinking about No-Code offerings.
I spent an hour going through Notion's 101 series and was impressed enough to appreciate its offering. Bubble's capabilities seem to be much greater, and it bills itself as "The full-stack, no-code app builder for everyone," so I want to spend more time with it.
All three of the No-Code vendors mention being - at the very least - AI-assisted, which is unfortunate, because I see No-Code as a reasonable refuge from the Generative AI hype. I assume the three companies all padded their marketing content with AI phraseology, because it's what one must do these days, even if consumers react with disgust to everything AI.
And, at least in the case of Creatio, much of what was discussed or hyped as "AI" wasn't necessarily Gen AI, but more traditional (circa 2022) techniques involving machine learning and computer vision. They likely didn't make that distinction because everyone, for some reason, still wants to hop on the Gen AI train, but I'm glad it wasn't necessarily focused solely on how Gen AI is going to be such a powerful disruptor (although there was one dubious claim about the "$3 trillion AI market by 2030").
I'm far less cynical about the No-Code movement because the products take some forethought to produce and aren't promising to know what I'm thinking and be my personal assistant, best friend, and lover as the LLM-adjacent claim does.
No-Code works because it's opinionated and bounded - the very opposite of most LLM offerings. It's generally expected to be point-and-click or drag-and-drop and allow users to generate web pages or web page adjacent functionality without needing to get into the guts of the code, even if the code, in this case, is HTML or CSS.
As I watched the Notion videos, I kept coming back to view it as a wiki on steroids. As someone with development experience, I'm not certain that it would save me any time over using something like Astro with a CRM, but I can see that, for non-developers, it could provide productivity gains.
I'll have to defer my opinion on Bubble, because its documentation is so dense, and I've only taken a cursory glance. Its video series looks to be 10 hours or more worth of content, so I'll probably just peruse the manual.
I can imagine I'll probably come away with a guardedly optimistic view, but suspect that it can't be the full-stack builder for "everyone," because it has to wall off functionality to make it usable.
But, again, I find this to be more of a feature than a bug. It allows people to do more development than in previous years. Because it isn't a probabilistic text machine, "citizen developers" have a much better chance of reasoning through the apps they create.
[Update: Since my first draft, I've discovered a couple of Intro to Bubble series. One is 10 2-minute vignettes and the other one is about an hour long. I went through the series and have made my way through the first 15 minutes of the longer piece, but think I have enough information to form my opinion.
I do think there's a lot of capability there, even, possibly, enough to be able to create websites that don't solely fit in a static page or e-commerce solution. However, the thing that picked at the back of my brain in the beginning and continued the further I got into both the Notion and Bubble series is - when you eventually need access to the code (and you will eventually need access to the code), it's not yours. You will likely need to engage professional services to make changes. Even if you did have access to the code, there's a good chance that, because it's machine-generated, it won't be easily readable or maintainable. Maybe this is a point where you can engage a Gen AI code agent to clean up your newly minted legacy codebase.
Finally, if you are a developer or have access to a developer, the tools may not provide as much of an advantage as you're anticipating. The capabilities are powerful, especially for those with limited coding skills, but developers can likely get things done faster and with more control using familiar tooling - Astro, tailwindcss, and Python in my case.]
Contrast this with new Gen AI agents everyone has been hot and bothered about recently. I had a chance to test Replit's agent because I'm a paying customer. I believe my prompt was something like "Create an online travel agency similar to Expedia or Booking for me." And, lo and behold, it asked me a series of further prompts (it even knows about the concept of reprice for searches!) and generated scaffolding in Flask and Javascript for the site.
It then asked me to quality check the site, so it could fix any bugs it made during its initial generation phase. I noticed that the calendar functionality on the search page didn't work. After telling it 5 times that the functionality didn't work, it still didn't work and still didn't work in the original manner that it didn't work.
This isn't to piss on what the agent created. The initial scaffolding code is impressive. But it's also so dense that you still need a software engineer to verify it, debug it, and augment it. I'll take this opportunity to preemptively head off a couple of arguments I've encountered recently:
- The models are getting better and it will just be a matter of time. They may be getting better, but it won't just be a matter of time. The models as they exist are fundamentally flawed due to their probabilistic nature and the fact no one understands how neural nets really work, and are reaching the limits of the energy the planet can produce to create them and cost way more than they bring in. I'm not holding my breath for the singularity.
- They still reduce the number of software engineers needed. Nope. The agent generated a lot of code, but it wasn't so much that I would've hired a second engineer to assist me with hand-coding. Granted, it would've taken me about 2 hours to do what it did in 5 minutes, but I didn't spend much time debugging it either, which adds to its generation time. It's a great productivity tool, not a replacement. Plus, any company interested in making more money will note that the productivity gains might be enough to shorten project life cycles, so that can pack more things into a planning cycle with a more realistic chance of achieving them rather than opting for deep cuts in their workforce that could undermine their plans (then again, never bet against short-sighted corporate greed).
Comments
Post a Comment