Search the archive

for $200 OFF. GET YOUR TICKETS

THINGS I'VE DONE WITH AI

A programmer with a deep passion for coding, initially skeptical of AI, has shifted to exclusively prompting and reviewing LLM output, allowing them to complete a large number of projects by just problem-solving.

TWO OF MY FAVORITE THINGS TOGETHER AT LAST: PIES AND SUBDOMAINS

This dev built pies.jim-nielsen.com to archive his pie photos by programmatically extracting them from Instagram exports, aiming to own his content independently from the platform.

PRODUCTION QUERY PLANS WITHOUT PRODUCTION DATA

PostgreSQL 18 introduces new functions and `pg_dump --statistics-only` to export and inject production optimizer statistics into non-production databases, enabling accurate reproduction and testing of production query plans without needing the actual data.

TWO YEARS OF EMACS SOLO: 35 MODULES, ZERO EXTERNAL PACKAGES, AND A FULL REFACTOR

This dev has a unique Emacs configuration built entirely without external packages, instead relying on built-in features and 35 self-written Elisp modules.

Deploy the template →

weeks of configuration. This Infrastructure-as-Code template

deploys a complete serverless foundation quickly—no manual setup required.

MOST CODING AGENTS BREAK 75%+ OF THEIR OWN FIXES OVER TIME

Traditional coding agent benchmarks, like SWE-bench, only test one-shot bug fixes, failing to assess how agents maintain codebases over time. The new SWE-CI benchmark addresses this by evaluating models on continuous integration loops across months of codebase evolution. Results show that most models introduce regressions in over 75% of tasks, with only Claude Opus exceeding a 50% zero-regression rate.

NOTES ON WRITING WASM

This post provides practical patterns for improving the experience of writing Rust-based WebAssembly (Wasm) using `wasm-bindgen`. The main challenge is in effectively managing the distinct memory models of JavaScript and Rust, which can lead to issues like broken handles and re-entrancy.

YOUR LLM DOESN'T WRITE CORRECT CODE. IT WRITES PLAUSIBLE CODE

LLMs often generate code that is plausible and compiles but is fundamentally incorrect or vastly inefficient. A prime example is an LLM-generated Rust rewrite of SQLite, which was found to be over 20,000 times slower for basic primary key lookups due to critical architectural flaws like missing B-tree optimizations and excessive disk synchronization. This sycophancy is a structural problem in their training.

I DON'T KNOW IF MY JOB WILL STILL EXIST IN TEN YEARS

A pessimistic view is that SWE jobs may not survive another decade due to the rapid advancement of AI agents. On the optimistic side, Jevon's Paradox shows that with software production becoming much cheaper, there will be more demand for software, and thus software engineers, than before.

TELL HN: I'M 60 YEARS OLD. CLAUDE CODE HAS RE-IGNITED A PASSION

Numerous experienced, older developers praise Claude Code for re-igniting their passion for coding, bringing back the excitement of earlier tech while cutting through the frustration of constantly changing modern web development frameworks.

Run the upgrade CLI and read the full release

ELECTROBUN

Electrobun is an open-source framework designed for building ultra-fast, tiny, and cross-platform desktop applications using TypeScript. It uses Bun for execution and Zig for native bindings, with features like isolation between processes and very small app bundles and updates.

REACT NATIVE GRAB

React Native Grab is a tool that allows devs to "touch-to-grab" exact native UI elements in a React Native app to capture their precise source and runtime context. It bridges the context gap to quickly pinpoint UI issues and hand relevant information directly to coding agents without guesswork.

PAGE AGENT

Page Agent is an in-page JavaScript GUI agent for controlling web interfaces using natural language commands. It has easy integration without requiring browser extensions or external tools, relying on text-based DOM manipulation and allowing users to bring their own LLMs.

GNU AND THE AI REIMPLEMENTATIONS

Reimplementing software is a lawful and historically validated practice, with GNU and Linux as precedents. Copyright law protects specific code implementations, not ideas or behaviors, making clean-room reimplementations legal even with exposure to original source material. AI now accelerates and cheapens this process.

FILESYSTEMS ARE HAVING A MOMENT

AI agents are using traditional filesystems for persistent context and memory, moving beyond the limitations of LLM context windows. This approach allows agents to "remember" project details, user preferences, and skills by storing them in files. Standardized practices like AI Skills are acting as de facto APIs.

THE SURPRISING WHIMSY OF THE TIME ZONE DATABASE

Time zones are complex, leading experts to advise relying on established resources rather than attempting to code solutions from scratch. The IANA Time Zone Database is an open-source repository for global time zone information, with historical information. However, beyond its technical function, the database also contains an amount of whimsy, like a Canadian intellectual's diatribe against daylight saving.

GROW FAST AND OVERLOAD THINGS

AI companies like OpenAI and Anthropic are struggling with poor LLM reliability because rapid, unpredictable user growth is leading to system overload and saturation.

I'M NOT CONSULTING AN LLM

Relying on LLMs for information is intellectually corrosive because it bypasses the challenging, critical research experience necessary for developing true understanding and intellect.

WILL CLAUDE CODE RUIN OUR TEAM?

The introduction of AI tools like Claude Code is disrupting traditional software team roles by blurring responsibilities and causing a standoff among specialists, but the author hopes this turbulence will eventually lead to new, more collaborative ways of working.

THE SYNTHETIC DATA PLAYBOOK: GENERATING TRILLIONS OF THE FINEST TOKEN

Optimal synthetic data for LLM pretraining allows for efficient generation with smaller models and quality mix-ins.

BUILDING RELIABLE AI CHAT ON MOBILE

Doctolib found that building AI chat on mobile was much harder than expected. Streaming responses broke React Native's standard scroll patterns, so they open-sourced their own list component to fix it. The main insight was switching snapshot polling for event polling (“what changed since X?”), which eliminated ~80% of wasteful requests and made error states explicit.

YOU NEED TO REWRITE YOUR CLI FOR AI AGENTS

CLIs need to be redesigned specifically for AI agents, as their requirements for predictability and defense-in-depth differ a lot from human user needs. Agent-first CLIs should prioritize raw JSON payloads for inputs and offer runtime schema introspection for documentation, and practice context window discipline through field masks and pagination. Good input hardening is necessary to defend against agent hallucinations, alongside explicit documentation to guide agent behavior.

GOOD SOFTWARE KNOWS WHEN TO STOP

The article sarcastically introduces a scenario where the simple `ls` bash command is replaced by an "AI-Powered Directory Intelligence" tool called `als`, showing the author's point about software overreach. Good software understands its core purpose, knows when to stop adding features, and focuses on continuous improvement within its defined scope.

THE ILLUSION OF BUILDING

AI lowers the cost of producing software that appears to work, leading to confusion between simply building an app and the extensive process of "engineering a system." These superficial creations are like "clay Bugattis," which look real but don't have the underlying engineering for actual functionality, reliability, and scale.

INTRODUCING GPT-5.4

OpenAI has released GPT-5.4, its most capable and efficient frontier model designed for professional work, now available in ChatGPT, the API, and Codex. This new version advances reasoning, coding, and agentic workflows, excelling in tasks involving spreadsheets, presentations, and documents with improved accuracy and reduced errors. GPT-5.4 also introduces native computer-use capabilities, allowing agents to operate across applications.

EXPO ROUTER V55: MORE NATIVE NAVIGATION, MORE POWERFUL WEB

Expo Router v55 advances its goal of delivering a more native feel to apps while supporting complex web scenarios. The release brings major native UI improvements, including a new Stack API for declarative headers, native tabs, dynamic colors, toolbars, and gesture-driven zoom transitions.

WHERE THINGS STAND WITH THE DEPARTMENT OF WAR

Anthropic announced it has been designated a "supply chain risk" to national security by the Department of War (in the US), a decision the company believes is legally unsound and plans to challenge in court. They clarified that this designation has a narrow scope, applying only to the direct use of their AI models in Department of War contracts, not all customers or broader business relationships. Despite the dispute, Anthropic reiterated its commitment to US national security and offered continued transitional support for its models to the Department.

THE BRAND AGE

The Swiss watch industry faced a "quartz crisis" in the early 1970s, decimated by superior Japanese competition, unfavorable exchange rates, and the advent of highly accurate and thin quartz movements that made traditional mechanical watches obsolete. To survive, a handful of prestigious Swiss watchmakers pivoted from being precision instrument makers to luxury brand marketers, transforming their products into expensive status symbols. This might happen in tech too, due to AI, where brand becomes the primary driver of value.

LABOR MARKET IMPACTS OF AI: A NEW MEASURE AND EARLY EVIDENCE

This research by Anthropic assessed AI's labor market impact, finding no systematic increase in unemployment but suggestive evidence of slower hiring for young workers in highly exposed occupations.

MEET KARL: A FASTER AGENT FOR ENTERPRISE KNOWLEDGE, POWERED BY CUSTOM RL

Databricks' AI Research Team has developed KARL, a custom Reinforcement Learning-powered agent that reduces inference costs and latency while improving the quality of enterprise knowledge tasks.

REME

ReMe is an AI agent memory management framework that provides both file-based and vector-based memory systems to overcome limited context windows and stateless sessions.

THE REJECTION OF ARTIFICIALLY GENERATED SLOP (RAGS)

RFC 406i, "The Rejection of Artificially Generated Slop (RAGS)," is a sarcastic imaginary standard that outlines a protocol for project maintainers to immediately detect, reject, and punish contributors submitting low-effort, AI-generated content.

Schedule a demo to learn more

SECURELY INDEXING LARGE CODEBASES

Semantic search improves agent performance, but indexing large codebases traditionally takes hours. Cursor addresses this by reusing existing codebase indexes from teammates, using the high similarity often found within an organization's repositories. This secure reuse is enabled by similarity hashes to identify the best index to copy, and Merkle tree-based "content proofs" which make sure users only access code they already possess.

MAKING FIREFOX'S RIGHT-CLICK NOT SUCK WITH ABOUT:CONFIG

Firefox's default right-click context menu has 26 options, which can feel cluttered. However, editing the `about:config` settings allows one to disable a large number of these unwanted buttons, such as "Ask an AI Chatbot" or "Copy Text From Image." While these changes reduce the menu to 15 items, several buttons are not removable, like "Bookmark Link" and "Email Image".

LIFE'S TOO SHORT TO HAND-WRITE API TYPES: OPENAPI-DRIVEN REACT

The common problem of React frontend types drifting out of sync with backend APIs can be fixed by a "contract-first" approach. This means making the OpenAPI specification the single source of truth for generating TypeScript types, API clients, and validation schemas automatically using tools like Hey API.

SIX RULES FOR DESIGNING COMPANY GOALS THAT ACTUALLY CHANGE BEHAVIOR

Company goals should primarily serve as a communication tool to create clarity. There are six rules for designing effective goals, including limiting the number of goals to three, ensuring one "wins in a fight" for prioritization, making them easily understandable, and explicitly defining non-goals.

HUMANS AND AGENTS IN SOFTWARE ENGINEERING LOOPS

attendees. Save a spot.

GOOGLE WORKSPACE CLI

The Google Workspace CLI provides a unified tool for interacting with all Google Workspace APIs, including Drive, Gmail, and Calendar. Designed for both human users and AI agents, it dynamically builds its command surface by reading Google's Discovery Service, automatically incorporating new API endpoints and methods. For humans, it has features like tab-completion and easy API calls, while for AI agents, it provides structured JSON output and over 100 pre-built agent skills.