Thoughts on Systems

You’re rolling out LLM-powered tools like Claude Code and OpenAI Codex.

Nono is a new lightweight sandboxing tool that takes advantages of native Linux and macOS kernel-level isolation controls to wrap Claude Code and other agentic coding harnesses (or any process) and prevent them from deleting your home directory or dropping

OpenAI engineer Ryan Lopopolo writes in Harness engineering: leveraging Codex in an agent-first world on how to effectively use coding tools by managing context.

Thariq Shihpar, an engineer at Anthropic, breaks down how Prompt caching is everything in building LLM apps such as Claude Code.

Chris Lattner, who created projects like LLVM that back modern open-source compilers, reflects on What the Claude C Compiler Reveals About AI Coding.

Michael Bolin provides a detailed explainer Unrolling the Codex agent loop | OpenAI on the OpenAI blog. this is the first post in an ongoing series where we’ll explore various aspects of how Codex works, as well as hard-earned lessons. […] To kick off, we…

Zoë Hitzig writes that OpenAI Is Making the Mistakes Facebook Made.

Cedric Chin recently started a series on The Study of Asian Tycoons.

Harper Reed documents his current LLM codegen workflow: effectively using LLMs in an agentic way via different prompts.

Charity Majors takes on the recent rollbacks of DEI policies at major companies, writing that Corporate “DEI” is an imperfect vehicle for deeply meaningful ideals.

Declan Chidlow expresses concern that AI is stifling tech adoption.

How did DeepSeek achieve such remarkable capabilites in their recent models?

Jon Keegan notes in Sherwood News that A free, powerful Chinese AI model just dropped — but don’t ask it about Tiananmen Square, exploring what chain-of-thought models reveal about the perspectives encoded in their training.

Molly White educates us again on how to think about cryptocurrency valuations, writing in her latest Citation Needed newsletter that, No, Trump didn’t make $50 billion from his memecoin When someone tells you that a pile of crypto tokens are “worth” X dol…

Jack Danger explains the concepts of Infrastructure Gravity & Domain Engineering in an excerpt from his book, Executive Engineering.

Lena Cohen writes for the Electronic Frontier Foundation that Online Behavioral Ads Fuel the Surveillance Industry — Here’s How.

Derek Thompson writes in The Atlantic about The Anti-Social Century.

Benji Webber provides a list of tips for overcoming resistance to Extreme Programming.

Simon Willison summarizes key learnings about LLMs in 2024.

You don’t have to take on technical debt as a startup.

At rapidly growing companies, teams move quickly and can easily accrue a little more debt than might be ideal.

Have you ever had to burn down a lot of bugs to hit a release date?

How can you dramatically increase the quality of your product?

As a hiring manager at a growing company, I’m often reviewing resumes.

Every day, we come upon problems where something isn’t working.

Sooner or later, every technology organization faces a necessary evil: migrations.

Years ago, a mentor of mine talked to me about the distinction between leadership (coping with change) and management (coping with complexity).

My mother recently forwarded me this interview of the pianist Glenn Gould: I encourage you to watch this (and all 6 parts), even if you know nothing about classical music.

Daniel Lemire (@lemire) recently posted on the Mythical Reproducibility of Science, noting that sharing code also makes it easier to spread your ideas: The reproducibility that matters is getting people to use your ideas.

The Chord protocol dynamically constructs robust and scalable overlay networks that map a given key to an active node.

When a graduate student completes their PhD, the software they wrote for that degree begins an almost inexorable decline into obscurity.

Many corporate firewalls prevent git from using its efficient binary protocol by blocking outbound network connections.

Tim Daly, one of the developers of Axiom, has a vision for solving the following problem: Computational science seems to be in a state where people work independently.

In the course of my research, I tend to do a fair amount of data analysis and reduction.