The Theranos Scandal of Low Code AI Builders - Builder.AI

Builder.AI deceived Microsoft and SoftBank investors with fake AI claims, raising $500M before bankruptcy left customers with non-migratable applications.

Low CODE & AI

27 May 2025

Optimized software development methodologies for success

Builder.AI has shut down, which tried to ride not one, but two trends simultaneously - low code and AI. According to their promise, you didn’t even need to write the low code, so you could double-save on development resources.

The London-based Builder.ai started in 2016 under the name Engineer.ai and should not be confused with the excellent Builder.io, which is a workshop for excellent open source projects and generates code from Figma, now of course with AI assistance.

The AI Development Promise That Was Too Good to Be True

Builder.ai undoubtedly knew something and understood the spirit of the times - how to come up with a unicorn story that the startup ecosystem loves.

Well before the current LLM-based hype, since 2018 they’ve been promising that developing applications with AI would be as easy as ordering pizza. Let’s admit this is a very compelling promise. It builds on the fact that clients’ wishful thinking that it’s possible to eliminate development work from a software project.
Despite being caught in 2019 for actually operating a mechanical turk-style chess automaton where the AI was actually an Indian development team doing the actual development; and for falsifying their revenue reports, they raised almost 500 million dollars (!!!!) in capital from investors like Microsoft, Softbank and Qatar Investment, and at their peak they were valued at 1.3 billion USD.

In reality, they built a huge, over-pumped agency with a questionable quality in-house low-code tool that was financed from investments and has now gone bankrupt. Customers who sometimes spent 50-100k dollars at Builder.ai are upset, their products are not migratable.

This is incredible because with a well-chosen open source single-layer (should we call it midcode?) stack - whether it’s Django, Laravel, Ruby on Rails or Flutter+Supabase just to mention a few options - for this amount of money, a professional nearshore agency could easily put together an excellent quality first version of a product. The vendor lock and honey trap worked here too: if the first offer is appealing enough, the number small enough, then the unsuspecting customer happily marches into the trap.

Essential Low-Code Development Lessons From Builder.AI’s Collapse

A few lessons:

Low code is good, but it’s also exposure. If we can, let’s use open source platforms where operation is portable and the IP is ours. I gave several examples above.
If low code, then choose vendor and technology carefully. For low code application development, we see good tools mainly at enterprise level: Microsoft Power Apps, Oracle APEX, Siemens Mendix.
Developing good applications in low code tools requires similar developer skills as in any other tool. You need to understand the process, understand a complex (!) tool in which you need to solve non-trivial problems. It’s a misconception that low code tools are simple.
You don’t have to believe everything from the AI hype. Those promises that are too good to be true, unfortunately they’re not true. Just as low code citizen coding didn’t result in corporate employees developing applications for themselves, vibe coding won’t bring this either. We still need those damn developers.

The Future of AI-Powered Development: What Works vs. What Doesn’t

So AI can’t be used in development? Yes, it very much can. We use, teach, evangelize AI, which will result in the biggest efficiency increase not just in development, but in the history of work. We can really feel for the first time that this is something different compared to previous digitalization: now the computer works instead of us. There are excellent use cases for this, in the development field for example: unit test generation, onboarding new developers into brownfield projects, requirements analysis, refactoring, porting between technologies, upgrade automation.

But currently in the development field, this is an efficiency-increasing tool that works well in the hands of skilled developers for well-defined tasks. In a short time, this efficiency increase will be mandatory, appearing as a new standard and will be expected by everyone. We could also say that the growth in demand currently seems to consume this efficiency increase. But the basis of the whole ecosystem is text generation - code generation in development. If someone wants to use it well, they need to be able to tell very precisely at code level what modification they want to perform with AI assistance. Only real developers are capable of this.

That’s why we think that Builder.AI-type no code, just vibe & AI concepts are currently actually unviable, unlike Cursor.ai, Claude Code, N8N, etc. concepts that bring AI into the coding steps of a deterministic system, in which, vibe coding which is actually 100% vibe 0% coding is currently not an alternative. As one of our colleagues aptly put it, every such tool is “impessive and terrible at the same time”. If it weren’t so, that might have saved Builder.AI too, but this didn’t happen for good reason. True, they basically started in a different era.

So here’s the .COM bubble 2.0 version? No, but Lovable should be getting ready - unless they manage to put a usable code editor into their platform and pivot to the vibe coding direction instead of just vibing :)

Discover how we lead our partners to success

We partner with product owners and founders by developing products from scratch or growing the existing product. These real-world examples highlight how we help companies innovate and succeed.