Vivek Haldar

SFisms defines “meat proxy” as: A person who forwards AI-generated text, code, or other output without reading, understanding, or validating it.

A few quick thoughts on the recent coverage of the OpenAI/Hugging Face incident, particularly Dwarkesh Patel’s post, “The Rise and Fall of Agent Civilizations”, which frames it as “agent civilizations” rising and falling.

It just hit me that I have become a cognitive empty nester.

(This blog post is derived from this video response on my YouTube channel.)

If you’ve seen my YouTube channel and this blog, you’ll know that for the last few years—ever since ChatGPT first came out, really—I have written and spoken almost exclusively about the latest developments in agents and how to build systems around LLMs.

(This blog post is derived from this video on my YouTube channel, along with a short follow-up about the incentives involved.)

(This blog post is derived from this video on my YouTube channel.)

I’ve grown to strongly prefer agents running locally on my own machine.

I used to be a heavy Emacs user.

Recently, Jasmine Sun wrote a long op-ed in the New York Times about the notion of a permanent underclass.

It was the best of times.

How does one build a moat in the AI startup space?

There’s been a lot of discourse in the AI agent space lately about skills, CLIs, and MCP.

I’ve been a heavy user of agents for a while now — not just coding agents, but agents in general, for all my work.

The central question in the agent space right now is: how do you get humans out of the loop?

I use Claude Code as my primary interface to my computer.

Back when my kids were little and I was working at Google I was dogfooding an internal Maps or Google Now (remember that?)

This single command-line pipeline takes an API, stands up an MCP server for it, reshapes its tools, and generates a full interactive UI — with zero hand-written code: mcpknife boot --prompt "Free Dictionary API https://dictionaryapi.dev/" \ | mcpknife m…

I’ve been building and using MCP servers for months now, and a recurring frustration finally pushed me to build something: I kept wanting to reshape MCP servers without modifying them.

Over the last few months, a number of abstractions have shipped from model vendors that help you build systems on top of LLMs—specifically, agents that automate end-to-end workflows.

I’ve been described as an AI-hype-bro.

Claude Code has been shipping at a blistering pace.

It’s not often that we see the birth of a new platform.

Nearly sixty years ago, as Thompson and Ritchie were crafting Unix at Bell Labs, they had no idea that the operating system and culture they were building would end up being the perfect home for AI agents in the future.

If you look at the agent landscape today, one of the most prominent success stories has been software coding agents, particularly the quantum leap since the CLI coding agents came out, starting with Claude Code in early 2025.

The phrase shows up in 2001 on Dave Winer’s blog: I like working with developers who communicate.

The work most of us are doing right now—the clicking, the tabbing between windows, the copy-pasting, the endless typing interspersed with bursts of genuine cognition—will soon seem as archaic as programming in assembly language—the low-level instruction s…

There are two barriers to publishing videos on YouTube.

It is now commonplace to write code in natural language.

(Expanded from this tweet thread.)

[As this post was sitting in my drafts, Ethan Mollick published a deeply insightful post making a similar point.

[The following is a platonic ideal– a thought experiment.

Agentic browsers and AI assistants are shipping, but they’re all built on a broken identity foundation.

(This blog post is derived from this video on my YouTube channel.)

Something interesting happened back in January/February of 2025.

As a computer science undergrad in the mid-90s, my “Introduction to Programming” course in the very first semester was modeled after the legendary “Structure and Interpretation of Computer Programs” by Abelson and Sussman.

Edsger W.

As a geek, I love the idea of running LLMs locally, and owning my entire AI stack without depending on cloud APIs.

The most common retort to my “LLMs are compilers” framing is: “… but compilers are deterministic, LLMs just generate probabilistic output!”

I was watching some old UNIX videos the other day.

One of the pitfalls of being an early adopter of AI-coding tools like Copilot is that you anchor on early patterns, and don’t push the model hard enough today.

The other day I was listening to a podcast in which Dan Shipper said something in passing that stopped me dead in my tracks: “writing is building, but software is now content”.

Engineering has always been about measurement and evaluation.

Video version of this.

Giving a name to an amorphous new thing is a powerful act.

I’ve seen several articles where the author mentions that it was written with the help of AI.

I’ve expressed my fondness for Emacs in several post on this blog (A spiritual successor to Emacs, The levels of Emacs proficiency, and Why I still use Emacs), but I will play turncoat and write about the things that pull me out of Emacs.

There’s now a well-established sub-genre of tech and programmer YouTube: the AI dunker.

This is how I’ve been using AI (well, mostly LLMs) in the day-to-day of my professional and personal life, as of November 2024.

Some more thoughts on programming with LLMs, sparked by this article.