Intro I started to use the Hermes Agent and OpenClaw. They burn a lot of tokens for what they do, but they do get the job done. When I’m out of home, I can just message them or email them (cuz they manage their own email address, because i’ve created it for them), but they are good only based on the underlying model they are being used on. For example: torrent search and download, extensive research on a topic, using built-in search and integrations (since I already have skills for those), and even coding minor features with some guidance.

(following post has been updated 5th of july, 2026) Intro A shift in LLM/AI usage: GitHub Copilot changed its user pricing. Microsoft changed the pricing. Nvidia & AMD are coming up with new laptops with 128 or more GB of VRAM to run agents and LLMs locally! But the prices of the chips are not going down; at the time of writing, the Nvidia RTX 5070 16GB is 65k INR.

Intro I wanted to test the performance and accuracy of the ai models at the time of writing this (May 2026, so I can leverage them in my day-to-day tasks). Currently, I am testing 3 models: 1. Deepseek (v4 just got released) 2. GLM 3. Qwen. Identifying a common task At the time of writing this, there are Tamil Nadu elections. I’ve made up my mind to ask the AI to visualize the TN elections and would rate them based on their accuracy and how beautifully they present the data.

Intro Today I want to talk about, Google Gemma 4; this sets a new standard for open source AI models. I am astonished by the quality of this, though it has a small model size compared to the other big ones out there.. like Kimi K2.5 from Moonshot AI, that thing is a 1 trillion parameter model.. 1 trillion! and Gemma 4’s biggest is only 31 billion.. but here is the thing, Kimi K2.

Intro Anthropic came and published their skills documentation on what they are roughly 2 months ago, but I suspect they and other AI companies were using them internally even before they published. The idea of people seeing them as a new thing and being astonished is abysmal..because we were already using it..but without the naming convention and the optimised context limiting. The concept is simple: you create a folder named skill you want, then inside it, the main file is called skill.

Intro In the previous post, I was talking about claude skills. This is about automation, no code/ minimal solutions. At this point, it’s been over 1.5 to 2 years since n8n and automation; nowadays it’s easy to just build an app/ ask AI to build the app instead of creating the automation, it’s easier that way, but regardless, let’s see how the future goes; I’m already seeing the trend of creating custom apps for each one’s personal preference.

Welcome to 2026, I once thought what it’s gonna look like when it’s 2024..2026.. when I was in 2017..2018. A lot has changed, I wouldn’t have even predicted the things that happened. Sideloading Before I used to say you can’t sideload in iOS/Apple phones without jailbreaking. In a way that was dumb, because when you think about it.. iPhone apps have to be developed..obviously they can be tested on a physical device.

Intro I’ve been mentioning these things in my previous posts. I don’t use AI to write these, but I use them as a way to fix spelling and grammar mistakes. There is also another way that I use these tools for, and that is for local workflows. Local workflows help writing automation scripts in natural language; you don’t have to write code and handle edge cases to cover it. Give an example and it will take it over.