Bubble
A serious choice when the owner has to change the product
We have shipped four platforms on it, including one doing AI document work for insurance carriers. It is not a toy. It is also not free of consequences.
The verdict
Bubble is the right choice when a non-engineer has to keep changing the product after launch, and the wrong one when your cost per user is decided by heavy background processing, because Bubble bills the server work rather than the seat.
The question is never whether Bubble can build it. Four of our production platforms answer that. The question is who maintains it in year two, and what the workload looks like when usage grows.
Choose it for the handover. Leave it when processing volume, not user count, starts driving your bill.
In short
5 things that decide this
- 01Bubble measures server resources in workload units, describing workload as the server resources needed to host, run and scale apps built on it.
- 02That billing model means cost tracks processing rather than users, which inverts most founders' mental model.
- 03The owner can change the product without an engineer, and on Golancer that was the reason for the choice.
- 04PremiumAudit runs on Bubble with the Claude API doing document parsing, validation and report generation, so external AI work is genuinely viable.
- 05Complex background processing and heavy data manipulation are where teams outgrow it.
For someone who has not built on it
Bubble is a visual application builder. You lay out pages, define data types and draw workflows that fire on events, and Bubble runs the result on its own hosting.
It is not a website builder. You get a real database, user accounts with permissions, scheduled jobs and API calls out to anything else. You do not get a codebase you can take somewhere.
What we have built on it
Four of our platforms run on Bubble, and none of them are simple. PremiumAudit.io digitises the whole premium audit lifecycle for insurance carriers, auditors and policyholders, built on Bubble with the Claude API doing document parsing, data validation and automated report generation. Broollie is a multi-tenant AI meeting platform used across more than 38 countries, wired to Zoom, Webex, Twilio, Stripe and SendGrid.
TomoDomo manages Swiss coliving properties, integrating Stripe, Calendly, Skribble for e-signature and Abacus ERP. Golancer gives freelancers projects, a CRM, Stripe invoicing and booking, and we built it on Bubble specifically so the client could iterate on it directly.
That last reason is the honest case for Bubble. When the person with the product ideas can open the editor and change the thing, the product improves between engagements instead of during them.
- Who editsThe owner, or only engineers?
- WorkloadServer work, not user count.
- IntegrationsAPIs out are fine.
- ProcessingHeavy jobs are the limit.
- ExitA rewrite, not a migration.
The last station is the one to price at the start, because nothing about it gets cheaper later.
Bubble, dimension by dimension
Scored from four products running in production, not from a landing page.
| Dimension | Verdict | Why |
|---|---|---|
| Time to first version | Strong | Weeks of build become days of assembly. |
| Owner maintainability | Best in class | A non-engineer can change the product. Rare and valuable. |
| Data model | Adequate | Real, and it punishes a bad early design. |
| External APIs | Strong | PremiumAudit calls Claude from it in production. |
| Heavy processing | Weak | Where workload costs and limits bite. |
| Cost predictability | Mixed | Workload units track processing, not seats. |
| Version control | Weak | Nothing like a git history for review. |
| Exit | Poor | Leaving means rebuilding, not exporting. |
Where it stands
Good at
- The owner can change the product after launch, which on Golancer was the whole reason for the choice.
- Serious integrations work. TomoDomo runs Stripe, Skribble e-signature and Abacus ERP through it.
- External AI calls are viable in production, as PremiumAudit does with the Claude API for document parsing and report generation.
- Multi-tenant products are achievable, which Broollie proves across more than 38 countries.
Weak at
- Cost is driven by workload units rather than users, so a processing-heavy feature can cost more than the customers using it.
- Heavy background processing and large data operations are where every Bubble build eventually strains.
- There is no git history, so reviewing what changed and rolling back a bad edit are nothing like an engineering workflow.
- A poor early data model is expensive to correct, because restructuring touches every workflow that reads it.
- You cannot export the application and host it elsewhere. Leaving is a rebuild, and that should be priced on day one.
“Their attention to detail, quality of employees, and work ethic were outstanding.”
Nicolas de Quesada · Lexpair
Common questions
01Can a Bubble app handle real scale?
Load shape decides this, not user count. Broollie serves teams across more than 38 countries, so headcount alone is not the ceiling. Processing weight is what strains it. Workload units track server work, and one heavy scheduled job can cost more than a thousand people reading pages.
02When should we move off Bubble?
When processing cost or a hard performance limit starts shaping product decisions. The signal is a roadmap conversation where somebody says a feature is too expensive to run rather than too hard to build. Move before that becomes a habit, and treat it as a rebuild rather than a migration.
03Can Bubble do AI work, or does that need custom code?
It can call any AI API, and we run that in production. PremiumAudit is built on Bubble with the Claude API handling document parsing, data validation and report generation for insurance premium audits. The model work happens outside Bubble, which is exactly the right division.
04Is a Bubble build cheaper than custom software?
Cheaper to start, and the total depends on what happens next. A product that stops changing is cheap to own. One that grows into heavy processing pays for that growth in workload. The eventual rebuild is a real cost, so plan for it rather than discover it.
