Happy New Year
What 2024 Taught Me
This post is coming in quite late (but Happy Thai New Year!). I’m taking this opportunity to reflect on various aspects of my life in 2024. Overall, it was a great year—I got a lot done, tried many new things, and learned a ton.
I’m breaking this post down into three parts: finance, business, and technology (specifically AI). I’m intentionally leaving out personal relationships—not because they’re not important, but because I don’t want to drag others into a very personal reflection.
1. On Finance 📈🚀🤑
2024 was another great year for my finances. I saw a 13% increase in net worth (after expenses) and a 5% increase in total assets.
To be honest, this wasn’t due to anything clever I did—it’s more a result of a healthy post-COVID economy. Not much else to say here except: I hope the trend continues into 2025 (spoiler alert: it doesn’t).
2. On Business 😃💼
2024 marked the first full year of SaiJaiAI. Honestly, I’m still figuring out what kind of company I want it to be. Right now, it’s a mix between a consultancy and a small software house. I’ve worked with all kinds of clients—banks, SMEs, government agencies, non-profits—and I knew they’d all operate differently.
What I didn’t expect? You’re always trading one set of problems for another.
2.1 Private companies care about data privacy—but not for the reasons you think 🤬
Yes, they’ll ask if you’re PDPA (and GDPR) compliant. But no matter what you tell them, using their usage data to improve your product is a hard no.
Why? Because they’d rather not use your product at all than risk their competitors benefiting from your improvements. On-prem is non-negotiable. If you want them to consider a shared model, you’d better offer freemium pricing—which is completely unrealistic for AI-powered systems.
In AI, usage data is essential. Without it, how do you benchmark real-world use cases? Public benchmarks only work for generic tasks. Real products are messy systems with multiple components—you need custom benchmarks to understand what to optimize.
2.2 Non-profits don’t mind data sharing—but they have zero budget 😵
Non-profits are often thrilled to adopt your tool. They don’t mind you using the data to improve things. Great, right?
Except… they expect you to offer everything for free and handle the integration effort—usually into some archaic, barely functioning system.
Of course, this varies case by case, but you really have to ask yourself: Will the time and effort ever be worth it for you?
2.3 Banks are full of ex-consultants—and they’re not your friends 💀
There aren’t many big fish in the Thai tech pond, but somehow, all the banks ended up hiring the consultants they used to work with.
Now those consultants are in charge—and running things into the ground. I don’t think you need to be a Big Tech company to build great products, but if you’ve never shipped a single piece of software… I’m skeptical you know what you’re doing.
3. On AI 🤖🔥
2024 was an exciting year for AI—wild, messy, but incredibly interesting.
3.1 LLMs made huge leaps—and that trend will continue through 2025
The big players are innovating and shipping at breakneck speed. But new contenders are popping up all over the world. DeepSeek R1, for example, sent NVDA stock on a rollercoaster.
How? Probably by taking advantage of being a Chinese company—a gray-zone, lightly regulated, data-harvesting operation. My guess is they mass-collected output from ChatGPT or Claude and reverse-learned the weights. It’s like distributed, malware-embedded, involuntary transfer learning (just my theory, don’t sue me).
If this becomes common, then what’s the moat? Companies spending fortunes training foundational models might just get “cheaply copied.” Sure, clones won’t out-innovate the original, but even the fastest man alive can’t outrun a train forever—especially when investor money starts running out.
OpenAI needs to figure out how to moat their products in ways other than just aiming for AGI first.
3.2 RAG became the new hotness—then hit its ceiling
At first, it was all about chatbots. Then RAG (Retrieval-Augmented Generation) took center stage. It’s a real use case, no doubt. But people are missing something:
The key to RAG isn’t the G. It’s the R.
Even with the best LLM in the world, if your retrieval layer is trash, you’ll get trash results. It’s like having the world’s best orator reading from poorly written notes.
Sure, LLMs can help with embedding, but that costs tokens and time. Tons of papers on Arxiv propose alternatives, but no clear winner has emerged. And now that attention is shifting elsewhere, I’m not sure RAG will stay relevant in 2025.
3.3 Say hello to Agentic AI 🧠🔧
People finally realized LLMs can do more than chat. Enter the agentic framework—where LLMs perform actual tasks autonomously: data aggregation, cron jobs, system management, etc.
Interoperability is the magic ingredient here. Model Context Protocol (MCP) is shaping up to be the leading standard—but it brings back all the old security headaches of RPC.
Looking ahead, I hope Anthropic and others put more effort into ironing out the kinks. If it works, here’s what 2025 might look like:
3.3.1 Content aggregation and digestion
Raw web pages with ads will die off. AI agents will read everything, summarize it, and serve it to you cleanly. Think of tools like Tivoli that just read the whole internet for you.
3.3.2 Personal data indexing
Your PDFs, bills, contracts—even screenshots of receipts—will be embedded and searchable via a friendly assistant. “Hey, show me my last three rent payments” becomes trivial.
3.3.3 Zero-friction discovery
Bots will constantly surface interesting products, trends, and media tailored to your taste—and give you instant paths to explore or buy. Like Tastemade, but 10x.
This post was supposed to be a 2024 recap—but clearly, I’m already speculating about 2025. Looking back, 2024 wasn’t just about progress—it was about pattern recognition. I saw how different players in the business world operate, how hard it is to build truly useful AI products, and how much of my financial growth is just riding bigger waves. I don’t have all the answers yet, but I’m starting to see the right questions. Whether SaiJaiAI becomes something big or stays scrappy and sharp, whether the AI hype fades or transforms—I’m here for it. 2025 might not be smooth sailing, but at least I know which direction I’m rowing.