AI, ChatGPT and the death of thin SaaS

· 4 mins read min read

Is general, basic SaaS dying?

One of the most interest economic quirks of the rise of generative AI and coding tools in particular is the direct and arguably hidden impact it's having on a whole range of low-priced SaaS utilities.

There's a general subscription fatigue arguably kicking-in, and I'd argue that this is felt in no greater place than basic SaaS services. I'm talking about the types of sites that offer a service for $10 p/month but only offer a thin wrapper around a highly specific, but basic, requirement. Things like keyword suggestions or image manipulation. Without the breadth of services that a site like Canva provides or the additional and proprietary data sources something like Ahrefs provides, it's difficult to see how the value these offerings provides isn't totally eroded by the rise of coding assistance provided by generative AI.

What do I mean?

Take this example seen today:

Screaming Frog provides a free version and a €239 annual version for a range of SEO functions. The trigger here is the customer experiences either a hitch or a re-subscription event and turns to ChatGPT. This wasn't possible before. You could turn to Google for competitors, but not for a generalist tool that could help you. It resulted in him creating in 30 minutes something that resolved his requirement, for no additional cost (beyond arguably his $20 p/month). More than that he then made the code available on Github.

My similar experience

I ended up with a similar experience not so long ago. I needed access to Midjourney's capabilities, Dalle not being up to scratch thought available by API (Dalle-3 no doubt in due course via API will solve world famine, but it wasn't available then). In short, I sought an API that could generate high quality Midjourney images.

I stumbled upon an Unofficial Midjourney API, it was excellent. Really well put together. And was about $40 or so a month over and above the Midjourney price. This is clearly most likely against MJ's T's and C's (sorry MJ!), but their model is amazing and Discord a massive pain if you just don't get Discord (I don't!) and need non-spammy programmatic access.

Anyway, I used it, it was great. Then one day it stopped working. I thought it was a ban, it wasn't. It was something to do with their API, it was probably temporary, but it was a pain. Who knows, it was probably my fault.

So, like the person in the screenshot, I turned to ChatGPT. I'm no superstar coder, and I use Laravel for most things. But within less than 20 minutes I had a Node script (I can just about spell Node) working from my Laravel scripts on-demand, perfectly tweaked to my requirements.

This was insane.

It was insane, because it was exactly what I wanted. Yes, i'd saved myself $480 p/year, but more than that it shifted my view on a lot of the other 3rd party APIs I was using for similar tasks.

Other APIs and micro-SaaS

There was a whole bunch of 3rd party APIs I'd used, especially at POC stage, when the cost didn't justify full dev investment (either in-house or otherwise). Things like Text Classifiers, URL extraction, summarisers etc. - these were all single feature SaaS tools - and all very good, at the time most of them were cutting edge. Yes, they could now in part be replaced by smart calls to OpenAI, but more than that ChatGPT can actually help you create pretty much the whole service.

Now no doubt, there'll be many who'll argue that yes, for early stages this works, but then the maintenance etc. requisite to keep these services alive mean they're worth the cost. But i'd argue in counter that actually this is far outweighed by the fact that the internal APIs are so custom for requirements and so cheap, that in-fact there's no competition.

The paradigm has shifted. A smart tech CEO recently explained an interesting anecdotal insight that he felt maintainability of legacy systems has now been made easier. A separate, but related discussion.

Anyway, who cares?

So, sure, this is an interesting micro-point (to me anyway!). But the broader point is much more significant. There's a hidden economic impact in a generalised tool. It's akin to a Swiss Army Knife that must, at it's point of commercial launch, have made a whole bunch of other smaller tools redundant. This is just much broader. It's really a strong kick in the teeth to a lot of the indie hacker movement. It competes with them exactly where they had such strength.

On the web, I sense over time this will be felt very keenly. Not only is micro-SaaS really up against it from a customer acquisition perspective, but now they have a direct competitor that not only likely performs what they do cheaper, but likely does it better and with very low friction and in a highly customised way.

No major points beyond that, it's just quite dramatic. And crucially not easy to sense happening, which raises the question what the hell else is the development of multi-modal generative AI going to render obsolete. Are sections of the economy just going to be wiped away - that seems unlikely, but at the same time plausible.

That for some will be an existential risk (and for others an opportunity), arguably more than the risk of a rogue AI hitting the nuke button.

More to Read

How much does prompting matter with LLMs?

The short answer: it does and it doesn't.

LLMs, algo trading and the promise of an algo-trading co-pilot

The concept I had in my head, wasn't the finGPT stuff, where you use a simple wrapper to get the cha...