Nikhil Singhal

@eng_khairallah1 Honest truth? Stop generating AI slop. Give it your voice

@ihtesham2005 Dude. Stop using Claude to write your posts. Just a cursory check shows this is pure AI slop. Do a little effort

A Hungarian psychologist raised three daughters to prove that any child could become a chess grandmaster through early specialization. He succeeded. Two of them became grandmasters. One became the greatest female chess player who ever lived. Then a sports scientist looked at the data and found something nobody wanted to hear. His name is David Epstein. The book is called "Range." The Polgar experiment is one of the most famous case studies in the history of deliberate practice. Laszlo Polgar wrote a book before his daughters were even born arguing that geniuses are made, not born. He homeschooled all three girls in chess from age four. By their teens, Susan, Sofia, and Judit were dominating tournaments against grown men. Judit became the youngest grandmaster in history at the time, breaking Bobby Fischer's record. The story became the gospel of early specialization. Pick a domain young, drill it hard, and you can manufacture excellence. Epstein opens his book by telling that story honestly and then quietly demolishing the conclusion most people drew from it. Chess works that way. Most things do not. Here is the distinction that took him four years of research to articulate, and that almost nobody who quotes the 10,000 hour rule has ever read. There are two kinds of environments in which humans develop expertise. Psychologists call them kind and wicked. A kind environment has clear rules, immediate feedback, and patterns that repeat reliably. Chess is the cleanest example. Every game ends with a winner and a loser. Every move is recorded. The board never changes shape. The pieces never invent new ways to move. A child who plays ten thousand games will see most of the patterns that exist in the game, and pattern recognition is exactly what chess mastery is built on. A wicked environment is the opposite. Feedback is delayed or misleading. Rules shift. The patterns that worked yesterday may be exactly the wrong patterns to apply tomorrow. Most of the real world looks like this. Medicine is wicked. Investing is wicked. Building a company is wicked. Scientific research is wicked. Almost every job that involves a complex changing system with humans in it is wicked. The Polgar sisters trained in the kindest environment any human can train in. Their success was real and the method was correct. The mistake was generalizing the method to fields where the underlying structure of the environment is completely different. Epstein's research is what made the implication impossible to ignore. He looked at the careers of elite athletes outside of chess and golf and found that the pattern was almost the inverse of what people assumed. The athletes who reached the very top of their sports were overwhelmingly people who had played multiple sports as children, specialized late, and often switched disciplines well into their teens. Roger Federer played squash, badminton, basketball, handball, tennis, table tennis, and soccer before tennis became his focus. The kids who specialized in tennis at age six and trained year-round for a decade mostly burned out, got injured, or topped out at lower levels of the sport. The same pattern showed up everywhere he looked outside of kind environments. Inventors with the most patents had worked in multiple unrelated fields before their breakthrough work. Comic book creators with the longest careers had drawn for the most different genres before settling. Scientists who won Nobel Prizes were dramatically more likely than their peers to be serious amateur musicians, painters, sculptors, or writers. The skill that mattered in wicked environments was not depth in one pattern. It was the ability to recognize when a pattern from one domain applied unexpectedly in another. That kind of thinking cannot be built by drilling a single subject. It can only be built by accumulating mental models from many subjects and learning to move between them. The deeper finding is the one that should change how you think about your own career. Specialists in wicked environments often get worse with experience, not better. Epstein cites studies of doctors, financial analysts, intelligence officers, and forecasters showing that years of experience in a narrow domain frequently produce more confident judgments without producing more accurate ones. The expert builds elaborate mental models that feel comprehensive and turn out to be increasingly disconnected from the actual structure of the problem. They stop noticing what does not fit their framework. They mistake fluency for understanding. Generalists do better in wicked domains for a reason that sounds almost mystical until you understand the mechanism. They have less invested in any single mental model, so they abandon broken models faster. They are used to being a beginner, so they are not threatened by the discomfort of not knowing. They have seen enough different domains that they can usually find an analogy from one field that unlocks a problem in another. The technical name for this is analogical thinking, and the research on it is one of the most underrated bodies of work in cognitive science. The single most useful sentence in the entire book is the one Epstein puts almost as a throwaway. Match quality matters more than head start. A person who tries six different fields in their twenties and finds the one that genuinely fits them will outperform a person who picked one field at fourteen and stuck to it on willpower alone. The lost years were not lost. They were the search process that produced the match. Every field they walked away from taught them something they later imported into the field they finally chose. The reason this is so hard to accept is cultural, not empirical. We tell children to pick a path early. We reward the prodigy who knew at six. We treat the late bloomer as someone who failed to launch on time, when the data suggests they were running an entirely different and often more effective optimization process underneath. The Polgar sisters were not wrong. The conclusion the world drew from them was. If your environment is genuinely kind, specialize early and drill hard. If it is wicked, and almost every interesting human problem is, then the people who win are the ones who refused to specialize until they had seen enough to know what was actually worth specializing in. You are not behind. You were running the right experiment all along.

I would add MCP’s and Agents to this list as first class items enterprise’s and its users are very excited about

@sukh_saroy This is so Claude written. At least take some time to fix the AI tell tales from your post

A new study just blew up the entire "vibe coding" movement. Researchers from UC San Diego and Cornell tracked 112 experienced software developers using AI agents in their actual jobs. The finding is the opposite of every viral demo on your timeline. Professional developers don't vibe code. They control. Here's what they actually found. The researchers ran two studies. 13 developers were observed live as they coded with agents in real production work. 99 more answered a deep qualitative survey. Every participant had at least 3 years of professional experience. Some had 25. The viral pitch of agentic coding goes like this. Hand the agent a vague prompt. Don't read the diff. Forget the code even exists. Trust the vibes. Andrej Karpathy coined the term. Tens of thousands of developers on X claim to run "dozens of agents at once" building entire production systems hands-off. The data says almost nobody serious actually works that way. Here is what experienced developers do instead. → They plan before they prompt. They write out the architecture, the constraints, and the edge cases first, then hand the agent a tightly scoped task. → They review every diff. Not because they're paranoid. Because they've seen what happens when you don't. → They constrain the agent's blast radius. Small, well-defined tasks only. The moment a problem touches multiple systems or has unclear requirements, they take over. → They treat the agent like a fast junior dev that needs supervision, not a senior engineer that can be trusted alone. The researchers also found something darker buried in the data. A separate randomized trial they cite showed that experienced open source maintainers were 19% slower when allowed to use AI. A different agentic system deployed in a real issue tracker had only 8% of its invocations result in a merged pull request. 92% failure rate in production. 19% productivity drop for senior devs. The viral demos lied to you. The paper's biggest insight is in one sentence: experienced developers feel positive about AI agents only when they remain in control. The moment they let go, quality collapses, and they know it. This matches what every serious shop has quietly figured out. The developers shipping the most with AI right now aren't the ones vibing. They're the ones with the strictest review processes, the tightest task scoping, and the clearest mental model of what the agent can and cannot do. Vibe coding makes for great Twitter videos. It does not make great software. The next time someone tells you they let Claude build their entire SaaS in a weekend, ask them how much of that code they've actually read. The honest answer separates real engineers from the demo crowd.

@ShahRathin has compiled the GOLD standard, a blueprint, a guidebook, a playbook for startups to come in future. I wish I had this when I was doing startups. Kudos

Wow. @lennysan has done an amazing job of listing the biggest takeaways.

@OfficialLoganK @GoogleAIStudio RIP your DMs

We are hiring a bunch of Members of the Technical Staff for @GoogleAIStudio who can blend PM, design, eng, and more If this is you, pls DM me, we will move fast for the best people.

I need someone to build me an extension or plugin or something that can take a post on X and generate an md file or pdf or something that I can then share with Claude. This is ridiculous

@felixrieseberg @bcherny The problem of posting only on X is that my Claude web or code cannot check this. I have to jump through the hoops to get this to them. Do you also post on Anthropic blog? @felixrieseberg

Hi! I'm here with *another launch*, it just happens to be extremely niche, nerdy, and probably only for a handful of people. In the desktop app, Claude Cowork and Code now have a little Bluetooth API for makers & developers, allowing you to build hardware devices that interact with Claude. I, for instance, built a little desk pet that alerts me whenever Claude is waiting for permission.

Context degrades, summaries drift, and the AI quietly forgets things you never asked it to forget. I thought I understood the problem well. Then I read this piece by Chrys Bader and realized I was only seeing half of it. He frames memory as an unsolved spectrum between raw (lossless but inert) and derived (compact but drifting like a photocopy of a photocopy). Every memory system is choosing a position on that spectrum, and neither extreme works. Best thing I have read on this topic.

I am an early adopter and loving the convenience this is bringing to my work

My three AI's don't talk to each other. Grok lives in my Tesla. Perfect voice, native to the car, always listening. I ask it about traffic, markets, news. It is fast and sharp. But it knows nothing about me. Copilot lives in my work. Documents, email, meetings, code. It understands my files and my calendar. But it has no idea what I am building on my own time or what I care about outside of work. Claude lives in my personal projects. It knows my family. It knows the trademark I filed this week. It knows the book I am writing, the stories I have been collecting for months, and the repos where all of it lives. When my wife made an observation from the driver's seat today, the AI connected it to a theme we had been developing for weeks. Three extraordinary tools. Three completely isolated worlds. This afternoon, I spent 45 minutes in the passenger seat of my Tesla, talking to my phone instead of talking to my car, because the AI that knows my work lives on a mobile app and the one built into the dashboard does not know me. The AI that could have connected it to my work calendar was somewhere in the cloud. None of them could see what the others were doing. We have been so focused on making AI more capable that we forgot to make it more continuous. The next breakthrough is not a smarter model. It is an AI that remembers you, no matter where you talk to it. @elonmusk @DarioAmodei

Le mec qui a créé Claude Code (@bcherny) vient de montrer comment son équipe dresse l’IA. Un fichier. CLAUDE.md. Tu le poses à la racine de ton projet. Dedans : les erreurs passées, les conventions, les règles. Claude le lit à chaque session. Résultat : l’agent s’améliore sans que tu retouches une ligne de code. Chaque bug corrigé devient une règle permanente. Boris Cherny utilise ça tous les jours chez Anthropic. Je vous mets son template ici. Prêt à copier/coller et à adapter à votre guise : ### 1. Plan Mode Default - Enter plan mode for ANY non-trivial task (3+ steps or architectural decisions) - If something goes sideways, STOP and re-plan immediately — don't keep pushing - Use plan mode for verification steps, not just building - Write detailed specs upfront to reduce ambiguity ### 2. Subagent Strategy - Use subagents liberally to keep main context window clean - Offload research, exploration, and parallel analysis to subagents - For complex problems, throw more compute at it via subagents - One task per subagent for focused execution ### 3. Self-Improvement Loop - After ANY correction from the user: update `tasks/lessons. md` with the pattern - Write rules for yourself that prevent the same mistake - Ruthlessly iterate on these lessons until mistake rate drops - Review lessons at session start for relevant project ### 4. Verification Before Done - Never mark a task complete without proving it works - Diff behavior between main and your changes when relevant - Ask yourself: "Would a staff engineer approve this?" - Run tests, check logs, demonstrate correctness ### 5. Demand Elegance (Balanced) - For non-trivial changes: pause and ask "is there a more elegant way?" - If a fix feels hacky: "Knowing everything I know now, implement the elegant solution" - Skip this for simple, obvious fixes — don't over-engineer - Challenge your own work before presenting it ### 6. Autonomous Bug Fixing - When given a bug report: just fix it. Don't ask for hand-holding - Point at logs, errors, failing tests — then resolve them - Zero context switching required from the user - Go fix failing CI tests without being told how ## Task Management 1. **Plan First**: Write plan to `tasks/todo.md` with checkable items 2. **Verify Plan**: Check in before starting implementation 3. **Track Progress**: Mark items complete as you go 4. **Explain Changes**: High-level summary at each step 5. **Document Results**: Add review section to `tasks/todo. md` 6. **Capture Lessons**: Update `tasks/lessons. md` after corrections ## Core Principles - **Simplicity First**: Make every change as simple as possible. Impact minimal code. - **No Laziness**: Find root causes. No temporary fixes. Senior developer standards.

This is one of the cleanest setups I have seen

The most unreasonable demand in all of technology: know what you want before you begin. After months of building with AI daily, I've found the bottleneck isn't the model. It's the human's ability to discover what they mean. That gap between instinct and intent scales from a text box to global governance. Wrote about where this leads. …ilsinghal-ai-trust-commons.medium.com/discovering-in… #AI #AIGovernance