RSSAmplifier

Blog

Thyag's Blog

Hey I am Thyag (also Kaffu!). This is my gallery of trying. Reach me out at @bythyag on twitter or bythyag@gmail.com ...

bythyag.bearblog.devRSS feed ↗10 posts

Latest posts

thoughts on agent harness

when initially everyone had been building a workflow, it used to work well in a contained environment. it was brittle but it did its job when you are writing tests, occasionally, or often, failing in production. just for reminder, i will avoid technicalities here and assume u have been working on ai field long enough. one example is langchain. why it didn't last longer and probably why we have…

Know Your Harness

Lately with the advent of Claude Code and Codex, we have been enjoying ourselves pressing "Enter" and watching reels while it gets our work done. That being said, fundamentally CC and Codex are "harness" which are built upon existing LLMs to bring out the best of their capabilities with the help of tools, database (or RAG), and with proper context window planning. You can check out the evaluation…

drawbacks of SKILL.md

while skill.md has gained a lot of traction lately due to its usage in claude code and other coding agents, i believe its application is still very limited, particularly in long horizon tasks. the skill.md files come handy if you want the agent to use specific tools for a particular task. for example, if we are analyzing sec filing data, we already have structured data in json. it is easy to…

your agent needs context, not just capability (part 1)

introduction in last 1 year we've seen rapid development in ai capabilities and tooling. claude opus 4.6, codex, clawdbot, and china models. with this, we expect the tasks at hand to become easier for ai to solve given sufficient context and tools. when we talk about tools, they're relatively straightforward. 50-200 lines of code with if-else logic, api calls, and structured outputs. on the other…

moving towns

i spent my last 4 years in delhi for my undergrad studies and then got the opportunity to move to bengaluru for my first job. i had initially resisted the idea, particularly because of how i perceived the town as an “active twitter user”. delhi was a dream - fun, bittersweet, and full of memories. i remember the last days when i was dropping my girlfriend at the airport. she was finally placed,…

notes on ai agents

I had a tough time trying to figure out why someone would want to use AI Agents. At the end of the day, they are just LLM wrappers, right? Research labs are doing more important work: making datasets and training models. I was wrong! AI agents are great, and when crafted with care, they can achieve amazing feats. They are able to do tasks more efficiently, save you costs and time (if you ignore…

chain-of-thought in large language models - what has changed over the years?

Abstract Chain of Thought (CoT) has been one of the most important breakthroughs in Large Language Models (LLMs), particularly because it significantly improves reasoning capabilities. CoT is a simple prompting technique in which the model is given a few examples that include both the questions and the step-by-step reasoning used to solve them. Though CoT prompting proved improvement in the…

how to read a research paper

To read a research paper, it is suggested that you read the paper in three stages. In the first stage, focus on the paper title, sections such as abstract, introduction, and conclusion. Go through the names of all the sections that are present in the paper. This pass helps you to get a gist of what the paper is about, domain, and novel contributions. You can then decide based on your reading if…

i graduated from iit delhi

I finally graduated. Even saying it was a wild ride of summers without a single day off would be an understatement. I studied every day, hacking through subjects, learning and unlearning, trying to be consistent, and doing everything I could to win the battle. It was a great deal for someone who hadn’t attended a single class in the past 3 years, slept through most of the mornings, and downloaded…

sometimes morning is boring

I decided to pull an all-nighter today to complete the syllabus of an exam which is due tomorrow. I did make significant progress on the syllabus, but I think I would like to solve problems more, without seeing the solutions, and more like struggle with them and try to understand them better and develop my own intuition. Here is a catch: I will do it after I wake up. I am writing this blog at the…