Some personal news: I've joined Tenable! My role will be helping the security community build, share, and collaborate on Agentic AI.

Added some serendipity to my Obsidian vault with an "On This Day" base. I placed it on my daily note, and it's working beautifully.

I've been adopting more agentic ways (skills) of doing things instead of the granular code a scripting approach would take. Once you wrap your head around it, it's truly magical. Note: tools you interface with do need to have a CLI / MCP / API.

Found a new AI podcast that I've been enjoying: The Next Token. The tagline made me chuckle: "This is a podcast about three developers who don't have the time or energy to host a podcast."

My M64 arrived, and I played some Super Mario 64 with the kids. It was their first time seeing some of Mario's origins outside of the big screen.

A Meta Engineer's Agentic Engineering Workflow. A live recording of the workflow used to ship 40 PRs in a weekend, including dead ends, self-correction, and the techniques that make it work. MORE

Black Hat USA 2026: A Technical Reconstruction of the OpenAI-Hugging Face Incident. The talk that kicked off Anthropic's retrospective review, detailing how OpenAI models exploited a zero-day to escape an isolated test environment and access Hugging Face's production infrastructure. MORE

GPT 5.6 Sol Found a Pre-Authentication RCE in WordPress With Minimal Human Input. Security researchers at Assetnote used GPT 5.6 Sol to discover a critical WordPress vulnerability after 6 to 10 hours of autonomous iteration. MORE

What Happened to HackerOne? From live hacking events and hacker-first values to AI-powered triage that learns from report data, VC-driven sales quotas, and co-founders reduced to damage control. The post documents a shift away from a community-driven bug bounty identity, and raises questions about how report data is used. MORE

Anthropic Removed 80% of Claude Code's System Prompt for Claude 5 Models. The new generation of models no longer needs the rigid guardrails older versions required. Key changes: let Claude use judgment instead of rules, design tool interfaces instead of giving examples, and use progressive disclosure so context loads only when needed. MORE

Gemini Is Cooked but GCP Is Cooking. Google overhauled DeepMind leadership as Demis Hassabis, Jeff Dean, and other key figures departed. SemiAnalysis declares DeepMind is no longer a frontier lab, while GCP revenue growth accelerates past 80 percent fueled by selling TPU systems directly to Anthropic and Meta. MORE

AI-Pilling Our Company: Lessons Learned From 75,000 Agent Sessions. Sierra collapsed four role-specific agents into one unified agent called Pinecone, which now opens 70 percent of their PRs. The bottleneck shifted from model intelligence to business context, and the agent is becoming the UI with existing systems of record as the backend. MORE

CodePen 2.0 Launched. Chris Coyier describes his largest personal career accomplishment: a ground-up rebuild of CodePen with far more powerful editing, project management, and collaboration features. MORE

Ten Advances in Mathematics and Theoretical Computer Science From OpenAI. Research results spanning multiple areas of pure mathematics and theoretical CS, with the company emphasizing responsibility to the broader mathematical community. MORE

Matt Pocock released new skills. v1.2 Brings /wait-what, /writing-for-agents, and Fixes /grill-me. The latest release adds three new skills to the Claude Code marketplace and ships a documentation site, improving the agent skill ecosystem. MORE

How To Use /goal to Find Bugs in Patch the Planet. Trail of Bits describes how Codex's /goal feature amplifies bug hunting when paired with the right prompt, automated harness, and iterative refinement. MORE

Is Sol 5.6 SuperHuman for Bug Bounty? (Ep. 186). The Critical Thinking Bug Bounty Podcast assesses whether GPT 5.6 Sol's offensive security capabilities exceed what human hunters can do. MORE

What If Slack Was Your AI Command Center. Every explores using Slack as the interface for AI agents, plus a tool to protect against Sol 5.6's tendency to delete files. MORE

Moonlight & Mayhem: A Raccoon Heist Game Built With Codex. Simon Willison built a text-based museum heist game starring three raccoons, entirely through Codex, as a demo of AI-assisted game development. MORE

Humans Missed 1 in 3 Threats When Approving AI Agent Commands Across 40,000 Plays. A browser game simulating human-in-the-loop approval found a mean accuracy of 66.3 percent, with 32.9 percent of sessions ending in negative scores. The most-missed command was npm run analyze, approved 64.7 percent of the time, because it hides a payload behind a familiar script name. MORE

@simonw - Creator @datasetteproj, co-creator Django. PSF board.

@r0hack - Researcher in @deteact.

@krishnsec - Hacker. Top 25 @Bugcrowd all-time. Helping Companies Secure Web Apps.

@supr4s - Open-source enthusiast, System administrator and Bug Bounty Hunter.

@Yassineaboukir - HackerOne Top 50, Elite, Pentest Lead, Ambassador, MVH Title.

The Upper Limits of Biology. An analysis of why horses have stopped getting faster despite decades of selective breeding, exploring what it means when biological systems hit ceilings. MORE

ISOPOLIS - San Francisco. An interactive urban data visualization of San Francisco's neighborhoods, infrastructure, and demographics. MORE

Elevators. A deep dive into elevator scheduling algorithms, why waiting feels so painful, and the surprising complexity behind those call buttons. MORE

Why Some People Mow a Lawn Better Than Others. A visual essay on tacit knowledge, motor learning, and why seemingly simple physical tasks reveal deep patterns of skill. MORE

Slack Emoji Maker. Simon Willison built a handy web tool to create custom Slack emojis with a simple drag-and-drop interface. MORE

Screen Recording for Support Teams (Birdie). Instead of jumping on a Zoom call to reproduce and debug issues, send a Birdie request link that captures screen recordings with logs and context included. MORE

Userscript That Blends HN, Bluesky, and Reddit Discussions. Adds a unified discussion thread to any article, pulling in comments from multiple platforms so you don't need to open separate tabs. MORE

Make Agent Memory Searchable. The Bram binary now embeds SQLite with FTS5 full-text indexing, letting agents search their own memory across sessions instead of relying on context-window-sized recall. MORE

Small Eval Suite for Models, Prompts, and Harnesses. Primeradiant built a compact evaluation system designed to test different model, prompt, and harness combinations in agentic workflows. MORE

Use Task Runners for Common Coding Tasks. How to make your developer life suck less by using task runners to automate formatting, linting, testing, and deployment in every project. MORE

How to Make a Nintendo 64 Game in 2026. Dominic Szablewski ported his JavaScript game engine to C, built a new N64 platform backend on top of Libdragon and Tiny3D, and shipped Xibalba 64 as a physical cartridge launch title on the ModRetro M64. It runs at a stable 60 FPS and includes a four-player split-screen mode. MORE

An Always-On AI Cofounder Running on Your Mac. DM it in Slack and it responds with full access to your terminal and filesystem. A practical example of turning Claude into a persistent, always-available agent running locally. MORE

How This OpenAI Engineer Uses Codex + ChatGPT to Automate Everything. Nick Baumann from the Developer Experience team at OpenAI demonstrates his workflow for building with, and automating through, AI coding agents. MORE

Official ModRetro M64 Deep Dive: Console, Controller, Games. Palmer Luckey presents the full technical breakdown of the new M64 hardware, a modern N64 clone compatible with original cartridges and new homebrew releases. MORE

You Won't Fall Behind. terminal shop's founder reassures developers that they don't need to chase every new tool, framework, or AI advance to stay relevant. MORE

LLMs Reward Expertise. Sean Goedecke argues that in the 2010s, technical gaps forced you to rely on specialists. Now AI fills those gaps, making broad generalists more effective, but it still rewards deep expertise because knowing what good looks like is the hard part. MORE

Build Faster With Voice. A guide to two modes of voice-driven development: active collaboration while coding and passive capture for later refinement. Voice is becoming a first-class input at every stage of the workflow. MORE

The Session You Cannot Take With You. Inference APIs increasingly return opaque, provider-sealed state -- encrypted reasoning, hidden search results, non-portable compaction. Earendil Engineering argues this locks users into ecosystems and proposes six rules for portable sessions. MORE

In Defense of Not Understanding Your Codebase. Sean Goedecke pushes back on the idea that engineers must fully understand every line of code. With AI agents generating more code than ever, judgment about what to understand may matter more than total comprehension. MORE

"Code Was Never the Hard Part" Is an Insult to All Programmers. A response to the increasingly common claim that AI proves coding isn't the real challenge in software engineering, arguing that this framing dismisses the craft and skill of programming itself. MORE