Why a few people get far more from AI
Andrej Karpathy names three causes for the gap between casual and heavy AI use: the size of the problem, the kind of work, and access. Act on each, starting with one problem.
11/11/26 planningskillscost controlThe whole catalogue
Andrej Karpathy names three causes for the gap between casual and heavy AI use: the size of the problem, the kind of work, and access. Act on each, starting with one problem.
11/11/26 planningskillscost controlAnswer five questions before asking Claude to build: the result you want as a number, the smallest first version, what to leave out, which parts must work, and what slows you down.
11/10/26 planningverificationautomationTurn a frustrating task into a small tool: filter the data, check model judgments, and keep the files needed to rebuild it.
11/05/26 Claude Codedelegationmodel choiceverificationChoose where Claude should work, what it should make, which instructions to save, and when to use a goal or scheduled task.
11/05/26 Claude Codemodel choiceskillsmemoryautomationagent safetyGive an agent one job, limit its tools, and check its work before granting more independence. Keep a human owner responsible for each automation.
10/30/26 delegationagent safetyverificationautomationUse a visual brief to compare options, settle missing decisions, and check what Claude needs before it starts a larger build.
10/21/26 planningClaude Codewriting promptsverificationDefine what the app should do, agree on the names used in the business, and check the agent's work with sample data.
10/16/26 planningwriting promptsverificationteamsmodel choiceChoose one recurring task, give an existing agent your files and procedures, and check the result. Use Harness Firmware to keep the verified setup with your project.
10/09/26 Claude Codeskillsknowledge basesinstruction filesverificationCompare a proposed AI workflow change with your current process, including setup, correction, maintenance, and the time you spend checking results.
10/03/26 automationcost controlplanningverificationChoose a task, give AI a specific tutoring job, and check what you can do without help. Use recurring mistakes to plan practice and spaced review.
10/03/26 careerknowledge basesverificationTurn a repeated correction into a focused Claude Code skill. Choose when it runs, test its output, and remove instructions that add no value.
09/29/26 skillsClaude CodeverificationFind why an AI task needs repeated corrections, change one cause, and compare the results before keeping the new workflow.
09/25/26 Claude Codeverificationcost controlinstruction filesCorrect a running task, check a remote handoff, and ask for evidence of what could break before accepting the result.
09/24/26 Claude CodedelegationplanningBuild a Claude setup in an order where each step builds on the one before: a working folder, shared pages, connected accounts, skills, scheduled runs, and a check before building new tools.
09/22/26 Claude CodeskillsautomationSet a coding agent a result, a check it can run, and a clear stopping point so a long task ends with evidence you can review.
09/22/26 delegationverificationClaude CodeDefine the choices, scores, and yes-or-no questions Jev can answer, then use the results to route work. Compare examples from four videos, recognize tasks that need deeper context, and test accuracy, cost, and speed before automating.
09/20/26 automationmodel choicecost controlverificationTurn a result you already like into a reusable Codex skill. Define its task, decide where judgment belongs, and test it on fresh input.
09/19/26 skillsCodexverificationAudit Claude Code instructions, preserve project requirements, and test a smaller setup against the work you already trust.
09/15/26 instruction filesClaude CodeverificationUse failed runs to repair skill descriptions, scripts, and checks. Test each change before keeping it, with Harness Firmware as a working example.
09/13/26 skillsverificationinstruction filesCompare agents on one repeatable task, check the tools each can use, and keep your skills usable when you switch.
09/12/26 model choiceskillsverificationinstruction filesAn AI operating system is a folder of files your agent reads before every message. The four parts it needs, in build order, and the habits that fill it from your head.
09/07/26 knowledge basesinstruction filesCodexTurn a tested task into a reusable skill, choose where it runs, and catch failed work before it wastes more time.
09/07/26 automationskillsverificationClaude Design asks the questions a designer would, keeps your colors and fonts in a design system your whole team reuses, and fixes whatever a teammate circles in a comment. When it beats a plain chat, and the path from brand system to finished asset.
08/28/26 designteamsA language model reaches for the most familiar answer, and the better it sounds the less you check. Four staged prompts force the opposite: break the problem into parts, sort facts from assumptions, recombine the checked parts, and design cheap tests. You keep the judgment.
08/27/26 writing promptsplanningTheo let Claude Code save memories on its own, then audited them: 45 files, written three times more often than read, most never opened again. How to run the same audit on your machine, the three ways saved memories rot, the ladder that finds a better home for each lesson, and what a harness built to pass the audit looks like.
08/25/26 memoryinstruction filesClaude CodeAgent workflows run one step after another by default, which is the slowest shape. Ask at each step whether it needs the result of the step before it, then re-shape the work: fan out independent steps, route requests down branches, loop until a check passes, and cap what each shape can waste.
08/20/26 automationplanningcost controlSkills built on one person's machine stay there, and the copies you send around go stale the moment anyone edits one. Put the team's skills in one GitHub repository, install it as a plugin, and every edit reaches every teammate's agent.
08/19/26 skillsteamsBoris Cherny, who created Claude Code, maps four steps between chatting with an AI and a system that starts work you never assigned. Find your step from the bottleneck you feel, and make the one change that moves you up.
08/17/26 delegationautomationAllie K. Miller runs 34 AI agents, and her day is reviewing what they propose, deciding, and handling escalations. What an agent needs before it can start work you never assigned (written goals, fresh context, tools, triggers), the four-stage ramp from one agent to a workforce, and the watchdog jobs worth assigning first.
08/12/26 delegationteamsautomationGrok Bot, the new beta app from SpaceXAI and Cursor, gives you a team of AI agents that keep working with your laptop shut, each on its own cloud computer. How work gets routed between them, how recorded skills and triggered routines run themselves, and which work should stay at your desk.
08/11/26 automationdelegationteamsCody Schneider builds agents that find leads and post content while nobody watches. The method: turn the human process into scripts on a schedule, and call the model only where a judgment is needed. Both of his builds, tool by tool.
08/05/26 automationcost controlClaude CodeIn ChatGPT, any chat can search every other chat, send work into them, and collect the results. How to find a lost conversation by asking, delegate work between chats, publish a result to a shareable page, and reach it all from your phone.
08/05/26 delegationcost controlCodexNearly everyone gets anxious before a job interview, and practicing out loud is what lowers it. Give an agent the real documents, a skeptical interviewer to play, and your voice, then have it break character and coach you.
08/04/26 careerBuzz is an open-source team chat from Jack Dorsey's Block where AI agents are members of the channel. What that buys you (swap the model, keep the context), what it can already do, and whether your team should try it yet.
08/04/26 teamsmodel choiceAn agent sent a discount code to 150,000 people because a to-do item looked like an instruction. A rule in the prompt is a suggestion; a rule built into the tools is a restriction. Scope every key to the least the task needs.
08/04/26 agent safetydelegationOne successful run proves an agent worked once. Collect real examples with answers a person already judged good, grade each run with a script or a judging model, and rerun the set after every change so you know whether it helped.
08/04/26 verificationOne big question in one chat makes the model decide, research, write, and grade itself in a single pass. Map the work instead: parallel research jobs, a checker that did not write the answer, a merge, and your approval before anything expensive.
08/03/26 planningdelegationverificationWhen a model can finish the task in minutes, what people will pay you for changes: choosing the task well, doing things others do not know AI can do, feeding your results back into your tools, and proving you can deliver.
08/03/26 careerAnything your agent reads (a web page, a resume, a calendar invite) can carry hidden instructions it will follow. Why the attack has no clean fix, and the five defense layers that shrink what a successful one can do.
08/03/26 agent safetyModels got good where the work grades itself, and they aim at the most likely answer, which in design and writing is the average one. The fix: break your quality standard into checkable rules and keep human judgment for taste.
07/31/26 model behaviorverificationA rerun loop makes the agent redo its own work until a check passes. Script the repetitive half of a skill, end it with a pass or fail verdict, wrap both in a loop, and keep a human at the checkpoints where errors are expensive.
07/23/26 automationverificationWrong agent answers usually trace to the files the agent reads. The four ways a growing knowledge base drifts from reality, and the habits (a router file, read-only audits, scheduled pulls, backtracking) that catch the drift.
07/23/26 knowledge basesverificationAn enhance-prompt skill turns a rough request into the full prompt the model will actually run, and shows it to you first. Reading that rewrite catches misunderstandings before any work starts.
07/22/26 writing promptsplanningWhen an agent hits the same problem twice, encode the fix as a lint rule, CI check, or steering-file instruction that every future session inherits.
07/21/26 instruction filesautomationClaude CodeA routine is a saved instruction Claude runs on its own, on a schedule or an incoming signal. What it replaces, and why an unattended run has to be written tighter than one you watch.
07/19/26 automationTwo ways to start an AI task: deepen a rough prompt when you know the goal, or brainstorm first when the goal or the tradeoffs are still open. A quick test for which one you are in.
07/19/26 writing promptsplanningRetrieval augmented generation in plain terms: put the relevant facts in front of the question. How to feed the chat, docs, and code your team already uses into a knowledge base anyone can ask, and why weighting beats volume.
07/19/26 knowledge basesteamsWhen a new model family ships, treat your routing as expired. Re-measure which size and effort give the most score per dollar, spot the size you can skip, and drop the speed modes that do not pay.
07/19/26 model choicecost controlCodexWhether AI can have original ideas is a debate that keeps moving the goalposts. A way to judge it honestly: read the trend line, and see why learning to predict text can build real understanding.
07/19/26 model behaviorYour limit is a compute budget: tokens consumed times the model that reads them. Audit where a session spends, trim the context every chat preloads, and send grunt work to cheaper models so the budget lasts the day.
07/18/26 cost controlClaude CodeThe prompt and the reusable skill file behind the about page on this site, with the reason each part earns its place.
07/18/26 skillsAutomating the bottleneck means finding the constraint that sets your pace and pointing Claude at it: four working rules distilled from Anthropic's own engineers.
07/18/26 automationverificationClaude CodeOpenAI and Anthropic prompting guides are one playbook when read side by side: lean out the prompt. Where they converge, where they diverge, which habits to drop.
07/17/26 writing promptsA memory store is a workspace-scoped directory your agent reads and writes with normal file tools, carrying preferences and prior mistakes across sessions.
07/17/26 memoryGuardrails are the support structure that keeps an hours-long agent run honest: an audit instruction, a memory that survives resets, and fresh-eyed verifiers.
07/17/26 verificationmemoryagent safetyThe newest Claude models are highly responsive to system prompts, so shouting backfires: soften trigger language, migrate deprecated levers, keep the model in bounds.
07/17/26 writing promptsClaude is a brilliant new hire with no context on your job. Brief it like one: explicit instructions with motivation, examples, XML tags, roles, and context ordering.
07/17/26 writing promptsClaude's July 2026 lineup is four models: Fable 5, Glasswing-gated Mythos 5, Sonnet 5 at intro pricing, and Opus 4.8, plus the three migration changes that bite.
07/17/26 model choicemodel behaviorAI Firmware is a boot layer of kernel rules, installable skills, and committed memory that makes every new agent repo start senior instead of relearning old lessons.
07/17/26 instruction filesskillsClaude CodeCodexReasoning tokens are billed at the output rate and re-read every turn after. How to compress what a model says without compressing what it knows.
07/17/26 cost controlA skill file is career-long judgment packaged as instructions your agent can run. How to adopt someone's method, and why you scan it like untrusted input first.
07/17/26 skillsagent safetyModels get repriced, capped, and retired; their way of working can be kept. Extract the top model's discipline into skill files, routing tables, and planner-and-worker setups any model can run.
07/17/26 skillsmodel choice