Divider

So-called low-code and no-code tools are in demand these days – entire companies get built on this technology and people have been raving about its potential on social media. At the same time, some people remain suspicious of this movement, most notably pointing out certain disadvantages of these tools and (we believe wrongly) labeling this movement as a the hype.

We are taking a big bet on it, even if our platform is all but typical for this no-code environment. Here is why.

No-code before (we knew) it was cool

When we started working on our product, it was clear to us that it would be designed for non-technical people. Given the limited options in the Machine-Learning-enabled automation space, we knew that full-time developers would eventually be implementing our tool as well but we purposely set out to serve "users who are comfortable using Slack, Zapier, and Dropbox".

Not so long ago, our view of the world was split into "software" and "developer tools". Only after we had started, we learned that the "no-code movement" was even a thing, let alone such a big and fast-growing one. But upon finding this niche, we were glad that at least one aspect of our product could be explained in less than one sentence – the vibrant community we then discovered around it was a nice bonus.

No-code defined

Skip to the next session if you know what it is.

Most prospect users are entirely unemotional upon hearing "and it works without having to write code" and you might as well be one of them. Just as we initially thought that all end-user software should be productive without having to program them (such as Excel), most people simply take it for granted. But there is a fundamental shift happening in the software space and that's worth a specification.

No-code includes a family of tools that allow people to build web applications and systems without having to program them conventionally. This is possible because of the many direct and indirect integrations between different tools, as well as open interfaces (APIs), that allow data to be exchanged more or less seamlessly. In pre-cloud times, this used to be much harder to set up and could generally only be achieved through programming.

A typical application built on no-code might then look like this:

Using these tools, it is entirely possible to build fully-functional and powerful software, be it a website, mobile app, or web application – entire software businesses exist that are exclusively built on this so-called "technology stack". You could achieve the same result writing code, of course. But many people choose to build software using nothing but the above.

Debunking a common myth: On performance

Before we dive into this topic, a word of comfort in case you are considering testing our software: As long as this article sits on our website, you can be sure that we are going to watch your model's performance very closely. We are at least equally invested in your success with our platforms as you are and will tweak the system where needed, for the good of all users that come after you.

Ben Tossell is one of the most active people on Twitter when it comes to no-code tools as well as the philosophy behind them.

There are some myths around no-code tools, suggesting they were not as performant, flexible, or scalable. For brevity, we will just agree with these counter-arguments and add one additional tweet by one of the most influential people in the no-code space, Makerpad founder Ben Tossell, in case you want to dive into more advanced issues:

But since our application has Machine Learning built-in, we frequently face skepticism about whether our platform can match entirely code-based Machine Learning systems and this deserves "to be destroyed" in a more thoughtful manner.

When discussing the abilities of Machine Learning systems, we are entering dangerous territory. That's because contrary to "normal" software, there are two components in systems using Machine Learning: The software itself (including the Machine Learning stuff) and the data. And even though algorithms keep getting better, actual performance is much more dependent on the input.

In broad terms, we are aiming at "state-of-the-art performance". What this means is that we are using the most recent models and testing whether they generalize across a variety of datasets. Because the latter is what our customers will do.

The "state of the art" may not sound spectacular at first and indeed it can be achieved by any reasonably good engineer. But there is a but: AI is one of the most active fields of research and keeping up can turn into an occupation on its own. While an in-house engineer will get increasingly overwhelmed with keeping all systems up-to-date, we are doing this in the background for all of our customers at once.

Having said all that, the question of performance alone is short-sighted and tends to get over-emphasized in the AI space. The truth is that almost all companies will experience a step-change by implementing productive AI solutions for the first time, not by fighting their way up from 98.9 percent to 99.5 percent accuracy.

To make it more concrete, let's translate what "below 100% accuracy" means in reality: At Vetevo, laboratory employees used to manually inspect 180 samples each day. After implementing our image recognition model, this number went below 20, which is a 900 percent productivity increase. Could this be improved even further with a custom system? Certainly. But this would take anywhere between 2-4 months of developer time compared to one hour from the first login to a working system.

To close the circle on that topic, some Machine Learning engineers might claim that they can build better-performing systems and in many cases, we would have to agree. But to assess the total impact, one should factor in development time as well as keeping the system up-to-date, as these two will ultimately affect the net benefit of such a system.

Why we are going all-in on no-code

We are not the first to claim that no-code is here to stay. After all, it's just another term for something that already existed before. But we will stick with the principle that our users should be able to use our product with ease and phase it into their systems without any coding.

Knowing what we alone are going to build over the next months and years, it is clear that we are merely scratching the surface on the supply side. On the demand side, large companies are just starting to discover the possibilities and while they tend to move much more slowly, the impact of "GE and friends" getting hooked on no-code will be perceptible.

Apart from the favorable commercial environment, we truly believe in increasing the possibilities for everyone. If you are a developer, you may as well fire up Stackoverflow and a cloud instance and get cracking. But for all others, this is simply not an option and those are the ones who will greatly benefit from such performant no-code tools, especially in the AI space.

Now that you're here

Levity is a tool that allows you to train AI models on images, documents, and text data. You can rebuild manual workflows and connect everything to your existing systems without writing a single line of code.‍If you liked this blog post, you'll love Levity.

Sign up

Now that you're here

Levity is a tool that allows you to train AI models on images, documents, and text data. You can rebuild manual workflows and connect everything to your existing systems without writing a single line of code.‍If you liked this blog post, you'll love Levity.

Sign up

Stay inspired

Sign up and get thoughtfully curated content delivered to your inbox.

Thanks!