In this post, I’ll be talking about some AI-focused no-code platforms:
The objective of these platforms is to reduce the use and integration of AI APIs and tools to a drag-and-drop process. Given how much of, for example, a Python implementation of LangChain is boiler-plate code, this isn’t as lofty a goal, as say AppSmith or Bubble, which seek to do the same but for a broader audience.
All of these tools, at least from the AI perspective, are in their infancy. I know N8N of old, whilst the other two contenders are new to the block. As such, they all have weaknesses which will mean they’re probably not production ready – but that doesn’t make them any less useful for MVP (minimum viable product, aka first prototype) development or for general noodling on an experimental basis.
Here’s a totally unobjective review of the strengths and weaknesses of the three candidates:
- Flowise
Strengths:
– Great Youtube videos to explain how to get started.
– Each ‘Flow’ can be embedded, which is absolutely awesome, and allows for a level of composition. (However, the dependencies between flows are then opaque, a maintenance nightmare in the making).
– Open Source and unencumbered.
Cons:
– Buggy. I had to use docker, I couldn’t get it to build under npm on debian 11 or debian 12, our default server builds, I ended up defaulting to a docker install. The UI is very temperamental under Firefox. Some changes to flows don’t ‘stick’, and require a restart to take effect. - Active Pieces
Strengths:
– Based on sequential execution in a way that Flowise isn’t. Supports loops!
Cons:
– Very constrained unless you want to pony up big money. Clearly I’m not the audience (I suspect ‘Digital Agencies’ are their target market). - N8N
Strengths:
– A long standing, mature product, although AI capabilities are new additions and very much a work in progress.
Cons:
– Closed source. You’re at their commercial mercy.
– Buggy. And because its closed source, you can’t fix it.
My final take: I’d use Flowise for a proof-of-concept and then use that as the basis for a ‘living specification’ and subsequent development in a conventional programming language. The no-code environment provides very natural way to experiment with ‘prompt engineering’. Every flow having an authenticated endpoint and being embed-able is a killer feature.
In time, if any of these platforms gain traction, I’d expect them either to be absorbed by the likes of make or zapier or gazumped by one of the Big Boys (if I was a betting man, I’d say Microsoft…).
No-code is, without doubt, here to stay. I don’t think we’ve scratched the surface of what it can do – we should be able to lay out the initial flow in natural language, and then tweak in a gui for example – but be wary of ‘eggs in baskets’.