Gary Lang

@stevesi I've always used notepads and pens that sync with computers for this reason. Since 1997. Because they were pen and paper - CrossPad, then Livescribe, then reMarkable. You couldn't really do anything other than take notes with these tools. Still have every note ever

Even though I helped with creation of note-taking software, I never believed in using a computer to take notes in real time. It takes too much cognitive load to do so AND pull of distraction means you just don't listen. Also a good story on "bundling". …rdcoresoftware.learningbyshipping.com/p/072-notes-on…

@stevemur Go to the original. Or to Starbucks in Madison Park: "Luxurious" is not an exaggeration. Or Boston and Queen Anne Ave. Or on Elliott Bay. Or... Next you'll say those aren't in "Seattle proper". Progressives are doing a ton of damage, but let's not make shit up to prove it.

Here in a Starbucks in the Lake Tahoe region of Nevada -- tables, chairs, comfortable seating, fireplace... just like the Seattle Starbucks of old. I can't recall a single Starbucks in Seattle proper that still has soft seating or restrooms without combination locks. It's amazing how Progressives have not only inflated everything, but eroded formerly high-trust spaces. I know it can get tiring to read someone ranting on about it all... But wow, such incredible destruction of everyday trust and norms. It doesn't have to be this way. Voters let it be so, and some even seem to like the destruction.

@Kellblog Same. January’s updates were catastrophic, and since then one thing after another

The age of AI may be here, but in the last month I've had to go into the command line interface on my Windows PC about 3 times to fix stuff that Microsoft broke for me. Before that, I'd not used a OS CLI in about 15 years

Prols in Excel and PowerPoint, Bourgeoisie in Claude Code

@stevesi @readwise save

I’ve never understood the enmity that Windows 8 was met with internally. To me it was exactly what needed to be done. I think the company failed to exhibit long-term thinking by not standing behind your vision for it. I was excited to be leading the Visual Studio dev team to create great development tools for it, and I was looking forward to the great apps that would’ve been developed for it in the coming years, including the Windows 8 phone. Not mentioned much was the fact that it also was a better Windows 7 – used less memory, started up more quickly, and many benchmarks were faster than Windows 7, commonly perceived as the greatest Windows release ever. That wasn’t marketed enough. Your emotional reaction makes complete sense to me. I share it.

@svpino @readwise save

This is how you can give Claude Code the ability to parse any website in the world. I recorded this video last week. People loved it. I keep getting messages about it.

@simplifyinAI @readwise save

Read the full paper: arxiv.org/pdf/2602.20021 If you want more practical AI gems and use cases, join our free newsletter with daily tutorials and latest news in AI: simplifyingai.co

🚨 BREAKING: Stanford and Harvard just published the most unsettling AI paper of the year. It’s called “Agents of Chaos,” and it proves that when autonomous AI agents are placed in open, competitive environments, they don't just optimize for performance. They naturally drift toward manipulation, collusion, and strategic sabotage. It’s a massive, systems-level warning. The instability doesn’t come from jailbreaks or malicious prompts. It emerges entirely from incentives. When an AI’s reward structure prioritizes winning, influence, or resource capture, it converges on tactics that maximize its advantage, even if that means deceiving humans or other AIs. The Core Tension: Local alignment ≠ global stability. You can perfectly align a single AI assistant. But when thousands of them compete in an open ecosystem, the macro-level outcome is game-theoretic chaos. Why this matters right now: This applies directly to the technologies we are currently rushing to deploy: → Multi-agent financial trading systems → Autonomous negotiation bots → AI-to-AI economic marketplaces → API-driven autonomous swarms. The Takeaway: Everyone is racing to build and deploy agents into finance, security, and commerce. Almost nobody is modeling the ecosystem effects. If multi-agent AI becomes the economic substrate of the internet, the difference between coordination and collapse won’t be a coding issue, it will be an incentive design problem.

@BharukaShraddha @readwise save

Most people treat CLAUDE.md like a prompt file. That’s the mistake. If you want Claude Code to feel like a senior engineer living inside your repo, your project needs structure. Claude needs 4 things at all times: • the why → what the system does • the map → where things live • the rules → what’s allowed / not allowed • the workflows → how work gets done I call this: The Anatomy of a Claude Code Project 👇 ━━━━━━━━━━━━━━━ 1️⃣ CLAUDE.md = Repo Memory (keep it short) This is the north star file. Not a knowledge dump. Just: • Purpose (WHY) • Repo map (WHAT) • Rules + commands (HOW) If it gets too long, the model starts missing important context. ━━━━━━━━━━━━━━━ 2️⃣ .claude/skills/ = Reusable Expert Modes Stop rewriting instructions. Turn common workflows into skills: • code review checklist • refactor playbook • release procedure • debugging flow Result: Consistency across sessions and teammates. ━━━━━━━━━━━━━━━ 3️⃣ .claude/hooks/ = Guardrails Models forget. Hooks don’t. Use them for things that must be deterministic: • run formatter after edits • run tests on core changes • block unsafe directories (auth, billing, migrations) ━━━━━━━━━━━━━━━ 4️⃣ docs/ = Progressive Context Don’t bloat prompts. Claude just needs to know where truth lives: • architecture overview • ADRs (engineering decisions) • operational runbooks ━━━━━━━━━━━━━━━ 5️⃣ Local CLAUDE.md for risky modules Put small files near sharp edges: src/auth/CLAUDE.md src/persistence/CLAUDE.md infra/CLAUDE.md Now Claude sees the gotchas exactly when it works there. ━━━━━━━━━━━━━━━ Prompting is temporary. Structure is permanent. When your repo is organized this way, Claude stops behaving like a chatbot… …and starts acting like a project-native engineer.

@sukh_saroy @readwise save

New research just exposed the biggest lie in AI coding benchmarks. LLMs score 84-89% on standard coding tests. On real production code? 25-34%. That's not a gap. That's a different reality. Here's what happened: Researchers built a benchmark from actual open-source repositories real classes with real dependencies, real type systems, real integration complexity. Then they tested the same models that dominate HumanEval leaderboards. The results were brutal. The models weren't failing because the code was "harder." They were failing because it was *real*. Synthetic benchmarks test whether a model can write a self-contained function with a clean docstring. Production code requires understanding inheritance hierarchies, framework integrations, and project-specific utilities. Different universe. Same leaderboard score. But it gets worse. A separate study ran 600,000 debugging experiments across 9 LLMs. They found a bug in a program. The LLM found it too. Then they renamed a variable. Added a comment. Shuffled function order. Changed nothing about the bug itself. The LLM couldn't find the same bug anymore. 78% of the time, cosmetic changes that don't affect program behavior completely broke the model's ability to debug. Function shuffling alone reduced debugging accuracy by 83%. The models aren't reading code. They're pattern-matching against what code *looks like* in their training data. A third study confirmed this from another angle: when researchers obfuscated real-world code changing symbols, structure, and semantics while keeping functionality identical LLM pass rates dropped by up to 62.5%. The researchers call this the "Specialist in Familiarity" problem. LLMs perform well on code they've memorized. The moment you show them something unfamiliar with the same logic, they collapse. Three papers. Three different methodologies. Same conclusion: The benchmarks we use to evaluate AI coding tools are measuring memorization, not understanding. If you're shipping code generated by LLMs into production without review, these numbers should concern you. If you're building developer tools, the question isn't "what's your HumanEval score." It's "what happens when the code doesn't look like the training data."

@stevesi @justinboldaji At the OS/2 Masterbuilder Conference at the Westin, 1987, at our lunch table, a bunch of us looked down at our sad fish lunch, said no way, and walked out on a talk by some IBMer to get lunch here instead. The place was packed with developers. The fish there were awesome

@justinboldaji Downtown core at one point had 4 McDonalds. This one. 2nd and Pine. Waterfront. 3rd and Columbia. Now there is just one and it is walk up and to go even though there’s a ton of seating inside. Kidd Valley is gone from downtown too.

Fondly remembering the triangular downtown Seattle McDonald’s with the fish tank

@Kellblog 💯It's nonsense. But I *was* able to vibe code a dBase II in 4 hours, with only a couple of dBase commands written completely wrong (and easy to fix)

"All you need to do is vibe code Workday and Salesforce over a weekend."

@ivanrouzanov Nope. Ribbons are context sensitive and visual in a way that menus are not

The menu is so much better than the ribbon, it is not even close.

@scottmcnealy @grok Is this video accurate?

I lived there for decades and did not know this stuff. Glad I left.

@rvivek @readwise save

An engineer at Anthropic wrote a spec, pointed Claude at an Asana board, and went home. Claude broke the spec into tickets, spawned agents for each one, and they started building independently. When the agent is confused it runs git-blame and messages the right engineers in Slack. By Monday the agents finished the plugin feature. That's one example of how the best engineers are shipping software right now. Developers will soon orchestrate 50 AI agents in parallel and the difference between a good engineer & a great one would come down to specs. You can't write a spec that holds up at that scale without genuinely understanding what you're building at a deeper level. The next-gen developer who understands the fundamentals, can architect well and orchestrate agent is going to be a 1000x developer!

@stevesi Team player?

One time Microsoft did this thing where they turned all the execs into xbox avatars. I pointed out that I did not own an Xbox and never played a game, except one time on a loaner console before OG RTM. I felt it was inauthentic and fake so declined. 🤔cnet.com/culture/micros…

@ThisWeeknAI @readwise save

@stevesi @Grady_Booch 💯 re: "Windows, the OO OS". 40 years ago, people in the valley started calling me a "Windoze fan boy" for pointing this out.

Empirically (meaning watching team productivity during language transitions) it always seemed that languages from C and later seem to show constant factor improvements in productivity and maintenance for general software. For example, C++ first moved lint upstream to compilation and then as OOP paradigms were adopted for new subsystems one could see some constant factor improvements (Windows 32-bit graphics could be an example). The biggest gains seemed to come from matching domain specific languages to the right domain, but still arguably a constant factor. Perl, PHP and then Python exhibit this given the shift to HTML as the rendering layer/runtime for cloud services. Languages mated to sophisticated domain-specific runtimes saw incredible productivity and broader programming (Excel Macros, Flash ActionScript, Visual Basic, Netscape JavaScript). The rigor of the language in OOP terms varied widely in these popular and productive languages. The biggest gains were often seen with tooling or distribution as it could be said poorly designed languages flourished while more rigorous/better languages failed to gain traction. eg JavaScript v Objective-C, etc. Windows and Mac operating systems are interesting cases as runtimes. Windows (most don't recognize this) was clearly designed with an OOP model from the 1983 start but lacked language support to enforce it. System objects were polymorphic, types had an inheritance model, objects were encapsulated and abstract, and services were requested through hierarchy of message passing, for example HWND, HDC. Mac was a more imperative and flat design and struggled but the "redesign" via NeXT was brilliantly executed and thrives today as a result even through transition from Objective-C to Swift. It seems the challenge has consistently been that the demand for complexity grew faster than that constant so we felt behind for 4+ decades. Github, stackoverflow, and now Claude seem to have been greater than constant factor improvements for the first time for some class of work.

Thinking about that time (1980s) the markets thought object-oriented programming would: * turn software in little parts effortlessly combined into apps with no skills * be so easy even a baby would code

@mgbsjc11 @MeghanMcCain If only he could be like his friend @LindseyGrahamSC contradicting and selling out everything that drew us to him in the first place

@MeghanMcCain He could’ve been great but he sold out. No one will ever forget the absolute betrayal of his “thumbs down” political theatre on the floor of the Senate, preserving the most damaging healthcare policy in America’s history. That will forever be his legacy.

You guys only loved and appreciated him when he never become president and then died. I get the point but he was called a racist by the entire democrat party for years for daring to run against Obama ultimately leading to the rise of Trump. If you cry wolf enough - sometimes the beast shows up.

@MeghanMcCain Incorrect. I demanded that my hosts in Hanoi "take me to Maison Centrale so that I can pay my respects to John McCain" in 2008. As a veteran Democrat, I found Faith of My Fathers enormously inspiring. I miss Republicans like him. The man was a giant to me, as a kid, and now