Forget softwares or software-as-a-service. The future belongs to Service-as-Software (SaS), where customers don’t buy tools; they hire agents. Imagining ditching travel booking website and instead use a personal travel agent AI to curate, plan, and manage your vacation. That is the power of Service-as-Software, and it’s powered by Large Language Models(LLM) and Multimodel Models (MMM).

Instead of selling tools, we’ll be selling outcomes. These AI “agents” won’t just complete tasks, they’ll understand customers’ needs and adapt. Think specialist consultants, not robots. This doesn’t happen with off-the-shelf LLMs, stuck in their generic, NPC ways. They need domain-specific knowledge, trained on real data, not just scraped text.

That’s where domain adapted LLMs come in. These are LLMs post-trained and fine-tuned on specific domains, imbued with the nuances and expertises of a particular service persona. Current RAG approach patches the symptom but does not treat the root cause of NPC LLMs. Domain adaptation is key. Like master chef seasoning a dish, we will be tailoring LLMs to specific domains and personas. Data is the fuel, analytics the spice, and fine-tuned LLMs the outcome.

In the AI future, The model is the product. You will no longer buy softwares; you subscribe to agents.

@article{
    leehanchung,
    author = {Lee, Hanchung},
    title = {Subject: Service as Software},
    year = {2024},
    month = {02},
    howpublished = {\url{https://leehanchung.github.io}},
    url = {https://leehanchung.github.io/blogs/2024/02/01/service-as-software/}
}