RandomWits

life is too short for a diary




Daily Logs for Aug 14, 2026

Tags: ai java letters news

Author
Written by: Tushar Sharma
Featured image for Daily Logs for Aug 14, 2026

Dear Vishi, dear logs for today.

Skills vs Prompt with LLM

For LLM all input text is broken down into tokens. LLM is token in and token out. LLM doesn't have special neural pathway to differentiate skills or prompt.

Prompt example :

System prompt: You are super engineer

User prompt: Explain this code.

A simple skill would look like

Skills
---
you are super engineer 

First you should understand code 

you have additonaltools to read system docs. 

So skills is prompt + workflow + references/scripts/assets + any additional tools it can have access to. But the skill is not always injected into the model context. Agent/runtime may decide to load a skill when user explicitly asks for it or when the task matches it.

Are newer LLMs trained to follow skills differently?

Maybe models can be trained to follow structured instructions and tool-use patterns better, but skills are mostly runtime-provided context. The model follows the skill after it is loaded into context; the runtime is what loads the skill.

Lazy annoation with Webclient

Les say you have a code

public class DocumentRepository(@Lazy WebClient webClient) {

}

How's this differnet than

public class DocumentRepository(WebClient webClient) {

}

When you add lazy to something, Spring boot adds a lazy proxy object inplace of real object. And Dependency Injection is managed by Spring. but for lazy, it waits till the app actually needs it like when a method is invoked on the dependency.

But why use lazy at all? Expensive bean creation defer it or something that's rarely used. This changes when the bean creation cost is paid; it doesn't make the bean faster after it is created. It can also make first use slower and can delay startup failures into runtime. Also if some other bean eagerly needs the same WebClient, then WebClient can still be created at startup.

News around the world

Nancy Mace is a member of the U.S. Congress. She tweeted:

Little quiz on US Congress

                 U.S. CONGRESS
                /             \
       House of Representatives     Senate
          435 members              100 members
          2-year terms             6-year terms
               \                    /
                \                  /
                 Bill passes BOTH
                       ↓
                    President
                       ↓
                Signs or vetoes


comments powered by Disqus