They’re not a client if they’re asking you to move off Bubble. I’m not sure what they are, but that’s not a client/customer. A true client wouldn’t know what Bubble is.
Either they want a specific feature that you’re not providing for them, or they think they can save money by dropping their external team, but neither of those are related to Bubble per se.
Interesting article, but I think it misses a lot of points.
Most, almost all, of the apps being built with the AI builders will never need an experienced developer to fix anything, because most of them (99.99% plus) will never have enough users to worry about anything.
This is the software landfill era. Software landfills overflowing with useless stuff nobody cares about… except maybe a few friends or family members.
It’s like the website-building era when WordPress came along…
there were suddenly billions of blogs nobody read.
I think, and I’ve said before, the thing that is most important these days is the marketing. Can you get users? Can your app be profitable? Can your agency build me an app and guarantee the first 100 customers? Can you build me an app and give me percentage milestones? And on and on.
AI didn’t kill the freelance developer or agency; it killed the low-end order taker. The developers losing work to AI prompts are the ones who were just translating basic feature requests into code without understanding the client’s actual business model or acquisition funnel.
A serious client won’t ask “Did you use Cursor or Bubble or code this in C++?” They will ask: “How fast can we launch, how easy is it to iterate when early users give feedback, and what’s the plan to get users through the front door?”
Vibe coding did not do away with marketing… it made it even more important than ever.
Bubble is still the quickest, low-cost place to put together an app for your customers that functions properly.
If you’re losing customers to AI builders, then you need to reevaluate your marketing and figure out what you’re doing wrong and why you’re missing the points that the truly serious customers are looking for.
All those other ‘weekend vibe coders’ weren’t real customers anyway 
I think you are missing some points that this article contain, and are related to marketing (because I know that you like to talk about this). Also, for OP this is not a question of bringing new potential users (but it could…) but keep existing customers that already have production apps (so this is not a question of understanding their business model… this is already done).
You have mostly three options from my point of view: Giving up 
, convince them to stay on Bubble or help them in the transition. Theses two last options will give a different marketing strategy and I think this article can help on choosing which strategy to use.
The article seems to say that Bubble as a true MVP builder is done, but as a first production version builder Bubble still fits. It also implies that Bubble learning curve is probably less steep than AI assisted code for non technical users since a lot of the needs to work with AI assisted code well take years of experience.
overall, I think OP was getting to a point that Bubble place in the world has shifted. No longer will experienced entrepreneurs use bubble for an MVP in a weekend, AI can do it, and no longer will enterprise clients get started on Bubble as AI might be cheaper, more flexible and not locked in. Bubble needs to market the platform and lean into where they fit in now.
I was wondering when we would have another one of these threads pop up
Sure, but in OP’s defense, if you are a dev shop then prospects do feel better going with someone that is certified, whether it is a meaningful credential or not. Besides, how many people on here have Bubble Certified Developer creds? Same thing.
did the same few months ago. the speed of development using cc its unseen. i have been able to achieve very very cool stuff. there are some issues though, first is that i think less, much less and cc just fills the thinking necessary to create a feature AZ. Its just amazing but it feels less and less my product. i kind of miss the days when i needed to think by myself the entire business logic. i somehow managed to create a workflow where i just discuss with cc about 2-3 hours then voila: feature is fully tested perfectly running 
We basically no longer check the code at all and deploy to production, we rarely have issues. from my perspective coding has been solved
I have no doubt it looks that way from your perspective
Dunning Kruger is so powerful
It goes back to the points brought up. Building MVPs is easier with coding harnesses but maintaining and growing production apps with active paying users still requires more reliability. Bubble is the better platform to do so.
This has been my experience too. I have been building tools using Codex for my own personal use. It gets the job done but I cannot trust it enough to manage an app for paying users in the long run.
There’s just too much code that gets stitched together as I prompt for changes. So much so I don’t have the patience to review the end result. Even documentation that gets generated is often not complete or detailed enough to get a good understanding of what’s happening under the hood.
Context is always going to be a stumbling block for LLMs.
I do like to talk about marketing (you should read my newsletter sometime )
The app is the tool that solves the problem, but the marketing is what makes the money.
Someone mentioned above that they use Claude Code to sit back and build an app…
ok, so also, anyone else can look at their app, use Claude Code, and sit back and build the same app.
Only marketing will make one successful, and the other one just another app among the millions getting built this year that nobody cares about.
Once the app is built, marketing becomes the most important thing.
If you’re an agency building internal tools, marketing is your most important tool. With the right marketing, you’ll never be out looking for clients; they’ll come to you.
And no, AI will not do marketing for you that helps you win. It can pump out 100s of blog posts that nobody reads, etc., but it won’t bring in users or customers.
There are effective ways to market an app or agency that let you make a lot of money and beat the competition every single time…
that’s why I like to talk about marketing 
I believe a lot of others will start to feel the same way and they should. I find it funny when people say I built an app, when reality is AI did. I’ve experienced this issue with thinking as well and lucky I don’t trust AI on areas I know more than it does. So often I get the wrong answers or just not the best implementation for Bubble or even Stripe. I can’t imagine anybody getting great results on anything AI does unless they know how to think about it themselves first.
I watched a video of a designer recently talk about this issue. He basically does all the thinking and planning as he would have 5 years ago, then feeds it to AI to implement, then manually fixes all issues. If people can’t fix all issues in code, then they likely are not going to have great products, and if they can’t fix all issues in bubble, same thing.
AI is making even quasi intelligent people dumber… our brains need exercise and we get that through thought. Don’t let AI do the thinking on topics you already know about, otherwise you’ll soon forget it, or at least be very out of practice.
I’m just happy that you like the Bubble community so much that you keep coming back 
This is what gives me hope that once the Bubble AI Agent gets on par with everyone else, the other benefits will shine through on their own.
Probably a new “job class” will appear for people that are just good at prompting from A to B. I’m not even talking about “orchestrators”. I’m thinking more about industrial line workers building cars. Most is robots, some parts are just basic human function. It takes a certain amount of training and experience makes you more reliable. But ultimately the AI is the builder and someone else is designing the actual car.
Let me share a bit of my experience: over the last six months, I developed a really interesting Bubble app that handles both hygiene checks and tableside ordering for restaurants.
It’s a multi-user app, where each restaurant is simply another registered user.
It was great—I was ready to start selling it!
That is, until I looked at how much this nice little project was going to cost me in WUs. I saw that the cost was higher than the monthly subscription fee I planned to charge.
Fortunately, there’s BuildPrint—created by our collective savior, George!
I set up GPT 5.6 Sol as an agent, and—admittedly, it took some effort—it developed all the necessary tools in JavaScript (a language I don’t speak). We built a D1/Cloudflare platform, and went from 200,000 WUs/month per restaurant to almost zero for that specific function!
It took barely a week, and it runs perfectly!
The takeaway: Bubble is fine for quickly developing a UI, but for complex, action-heavy (WU-intensive) database operations? It’s a no-go. Yet, even with my limited expertise, I can succeed today thanks to BuildPrint and GPT.
Anyway, that’s my little thought for the day!
Insane, how was the experience integrating it? Did it set up the relevant API calls / workflows all for you?
You used Buildprint to vibe code a non-Bubble app? Have you heard about Codex?
Buildprint fully hosts and integrates it for him with no setup required
Yes it did !!!
I would say more than 80% of the ideas, objectives, tools uses were mine, but hard dvpt is more than 80% of GPTwork, I act as a director of dvpt, but as a programmer actually, very few time ! It prepare all the API ( and all the way NOT to use API which burn WU, by direct calls to Cloudflare, Stripe, Backblaze, openrouter and so… It’s really interesting…
The only thing is that now, when I look at my app, there are a plenty of WF that I don’t understand. i know they work… but how ???
That I think is problem with most vibe coded stuff, no matter which platform hosts it. If it breaks for some reason could be hard to diagnose if AI isn’t able to spot the issue.
What was it that the restaurant feature did that cost 200,000 WUs a month. Usually numbers like that indicate an unoptimized build and not necessarily high costs to build in Bubble.
A complex restaurant order-taking system—handling products, availability, allergens, various prep stations (kitchen, bar, pizza station, seafood, cocktails, etc.), different dining areas, and integrations with services like Uber Eats, all in real-time—consumed a huge amount of WUs when built on Bubble! It took about 200 WUs for a table of four (starters, mains, desserts, drinks)—covering everything from sending orders to the kitchen and notifying servers to order confirmation and bill preparation. So, an average restaurant serving 1,000 tables a month would rack up 200,000 WUs! (and now almost 0 !)
