Llm

I write and review a lot of Agent Skills, and find myself frequently pushing back in reviews, as many authors assume they’re writing “just another markdown prompt” without considering that they’re actually working with one component of an agentic system. The TLDR of my feedback is usually along the lines of: The description’s only job is to tell the agent when the skill should be triggered. Keep the description as terse as possible while still triggering. The SKILL.md should be lean with detail pushed to reference files. Put repeatable work in scripts rather than relying on model inference. Skill Creator Primer I recommend teams use or create their own version of my skill-creator-primer skill. ...

KL divergence against a known-good reference answers “how much did this quant change the model’s behaviour?” rather than “how good is this model overall?”. What KLD measures KL divergence measures how much one probability distribution disagrees with another. At each token position, both the reference and the quantised model emit a distribution over the full vocabulary (~248k tokens for Qwen-class models). The reference might say "~80% likely the, ~5% likely a, …"; the quant says something slightly different. KLD compares the two per position and averages. ...

NVIDIA artificially restricts peer-to-peer (P2P) GPU communication to their enterprise cards. Turns out this is a software limitation, not a hardware one. I patched my drivers to remove it, hacked vLLM to take advantage of it, and got a 15-50% throughput improvement running Qwen 3.5 35b on dual RTX 3090s.

An observation on functional correctness without domain quality.

If you’re building MCP servers, you should be adding the ability to disable individual tools.

MCP DevTools - The one tool that replaced the 10-15 odd NodeJS/Python/Rust MCP servers I had running at any given to for agentic coding tools with a single server that provides tools I consider useful for agents when coding. The Problem The MCP ecosystem has grown rapidly, but I found myself managing many separate servers, each often running multiple times for every MCP client I had running, not to mention the ever growing memory and CPU consumption of the many NodeJS or Python processes. ...

Square Peg hosted event on June 20, 2025 where I demonstrated a basic version of my daily Agentic Coding workflow using Cline and MCP tools. What does it take to write enterprise-grade code in the AI-native era? Join Square Peg investors James Tynan and Grace Dalla-Bona for a live demo and Q&A session with three leading AI-native developers - Grant Gurvis, Listiarso Wastuargo, and Sam McLeod - and get a behind-the-curtain look at the workflows that enable them to ship faster, smarter, and cleaner code using tools like Cursor, Cline, and smolagents. ...

Picture this: A business leader overhears their engineering team discussing “vibe coding” and immediately imagines developers throwing prompts at ChatGPT until something works, shipping whatever emerges to production. The term alone-“vibe coding”-conjures images of seat-of-the-pants development that would make any CTO break out in a cold sweat. This misunderstanding is creating a real problem. Whilst vibe coding represents genuine creative exploration that has its place, the unfortunate terminology is causing some business leaders to conflate all AI-assisted / accelerated development with haphazard experimentation. I fear that engineers using sophisticated AI coding agents be it with advanced agentic coding tools like Cline to deliver production-quality solutions are finding their approaches questioned or dismissed entirely. ...

I follow a simple, yet effective flow for agentic coding that helps me to efficiently develop software using AI coding agents while keeping them on track, focused on the task at hand and ensuring they have access to the right tools and information. The flow is simple: Setup -> Plan -> Act -> Review & Iterate. It’s a cycle rather than a straight line - what you learn in each pass feeds back into your setup and plans for the next. ...