One of the interesting conversations that I am noticing lately is that SaaS is dead and will be replaced by your personal apps built by AI. I have been thinking about SaaS for quite some time, as that is one of the key industries NonBioS should enable and impact. This is how we think about the AI impact on SaaS in the near to mid-term.

But before I share our thesis on SaaS, let me take a step back and try to reason where the "SaaS is dead" movement is coming from. There are a bunch of AI code-generating platforms that encourage you to come and build your webapp. The webapp is in most cases just a frontend. The better ones are able to also connect it to their backend/DB of choice. The result is a seemingly functional webapp, delivering just your use case.

While these AI-generated applications might seem impressive at first glance, they represent only a small fraction of what makes modern SaaS work. As any engineer who has written code for a few years would tell you, modern SaaS is a lot more than a frontend connected to a DB. The frontend-backend is just the tip of the iceberg which powers most modern SaaS. The real body is below the surface and what you don't see.

Modern SaaS can best be described as a compendium of services, interacting together to deliver a product. Some of these services are the core services which engineers would write - including the frontend and backend - but the rest of the services might come from third-party SaaS providers. These services could come as hosted offerings, accessible through an API, or through self-hosted options (in case of open source) or even co-located with the core services.

Consider Slack's architecture, which handles millions of messages in real-time across global teams. Their system relies on a sophisticated message delivery pipeline that ensures sub-second latency across continents. This involves not just a simple frontend-backend setup, but a complex orchestration of message queues, caching layers, and backup systems. They use specialized services for handling file uploads, message search, and user presence detection. Their architecture includes custom-built solutions for handling websocket connections at scale, along with sophisticated rate limiting and load balancing systems.

Now here is the interesting part. Some of these services work independently. But others require careful orchestration with other services. And this is determined by your scaling requirements, architectural choices, and product positioning. All of these interconnected concerns come together to orchestrate the perfect symbiosis of services which power any modern-day SaaS.

While AI-generated applications might eventually replace traditional SaaS in the long term, the short to medium term reality is more nuanced. The complexity of building and maintaining production-grade SaaS applications, with their intricate service orchestration and scaling requirements, remains beyond current AI capabilities. However, this landscape could shift dramatically as AI technologies evolve and mature over the next several years.

What is possible today is that SaaS itself will be built more thoughtfully and more efficiently using AI. This is now possible. What we are doing at NonBioS will help you do exactly that. It is possible now for SaaS to be built by a single CTO working alongside a fleet of NonBioS AI engineers. The NonBioS army has already begun work on more than one such SaaS products, lead by their human CTO Generals.

To further demonstrate this capability, we have also begun work on Kortex, a prompt evaluation framework that we needed internally at NonBioS. We found that existing services weren't robust enough to handle our specific requirements for evaluating prompts in complex workflows. Kortex allows for sophisticated prompt composition and evaluation in a workflow-oriented manner, filling a crucial gap in the current AI tooling landscape.

Kortex is also open source, allowing others to benefit from and build upon our work. Through Kortex, we're not just building another SaaS application - we're demonstrating how AI can be leveraged to create sophisticated, production-grade software that solves real-world problems. All code of Kortex is untouched by humans, and you can see the very first check-ins already deployed to Github.