by armin-ronacher
Coding agents dominate reinforcement learning data, which means the smartest way to build any agent is to lean into the coding agent paradigm — files, code execution, Unix workflows — rather than fight it with custom tools.
by sarah-guo, elad-gil
Your productivity is now measured in tokens commanded per second, not lines typed — Karpathy hasn't written a line of code since December 2024, and the game is now arranging autonomous agent loops that run without you.
by topoteretes
AI agents forget everything between calls — Cognee gives them persistent memory by turning raw data into knowledge graphs and vector stores that survive context windows.
by geoffrey-huntley
Porting codebases between languages is a solved problem if you decompose the work into specs first — reduce code to language-agnostic PRDs, then let an agent loop rebuild it in the target language against a strict compiler.
by gergely-orosz
The abstraction ladder never stops moving — compilers, assembly, even hand-written code are becoming irrelevant. AI-enabled teams of 2-20 people will rival big company output, but the vampiric burnout effect means you might only get 3 productive hours at max speed.
by alexander-opalic
The practice of using AI tools — copilots, agents, and code assistants — to augment software development workflows.
by gergely-orosz
Software engineering is undergoing a printing-press moment: coding as a gatekept craft is dissolving, generalists who ship prototypes and adapt weekly to new model capabilities will define the next era.
by alexander-opalic
How the /batch and /simplify Claude Code skills work — parallel work orchestration and automated code cleanup.
by indydevdan
Pi agent is the open-source, unopinionated counterweight to Claude Code — minimal defaults, full harness control, and extensibility that lets engineers build what Anthropic won't ship.
by zhanna-sharipova
Owning your data isn't enough — local-first must extend to owning your communication channels, and the same infrastructure that protects human social connections will become the foundation for a trustworthy multi-agent future.
by agenticnotetaking
A Claude Code plugin that generates a complete, personalized knowledge management system through conversation — derivation over templating, backed by 249 interconnected research claims.
by matt-carey
MCP's context bloat problem has a clean solution: give the agent two tools—search and execute—running in a V8 sandbox against the full OpenAPI spec, cutting 1.17M tokens to ~1K.
by lenny-rachitsky
Coding is a solved problem — the bottleneck shifts to taste, product sense, and deciding what to build. Boris Cherny makes the case that underfunding teams, betting on general models, and building for the model six months out is how you win.
by alistair-gray
The competitive moat in coding agents isn't the model — it's the orchestration layer. Stripe's Blueprints prove that mixing deterministic workflows with agentic flexibility, backed by 500 MCP tools, scales autonomous coding to 1,300+ PRs per week.
by indydevdan
Skills alone are a dead end — you need four layers (capability, scale, orchestration, reusability) to solve entire classes of problems with agentic browser automation instead of one-off tasks.
by goose-oss
Playwright CLI skills let AI agents generate, run, and debug browser tests through natural language — writing snapshots and recordings to disk instead of filling the LLM context.
by lex-fridman
Fun beats funding — OpenClaw's 180K GitHub stars prove that personality, self-modifying software, and a lobster mascot can outcompete every well-funded AI agent startup that takes itself too seriously.
by anthropic
Anthropic's 30+ page official guide covers the full skill lifecycle—from progressive disclosure architecture (frontmatter → SKILL.md → references) through five reusable workflow patterns, three-level testing, and organizational distribution—positioning skills as the knowledge layer that turns MCP's raw tool access into guided, reliable workflows.
by onur-uzunismail
The most effective coding agent isn't the one with the most features — it's the one that wastes the fewest tokens. Pi's radical minimalism exposes how bloated our baseline tools really are.
by sarah-perez
Spotify's best engineers stopped writing code entirely — their internal 'Honk' system built on Claude Code lets them ship features from Slack on their phones, signaling that AI coding has crossed from experiment to default workflow at scale.
by kieran-klaassen, peter-yang
The simplest form of compound engineering is telling Claude to update CLAUDE.md after every mistake — but the full four-step loop (Plan, Work, Assess, Compound) with specialized sub-agents turns each development cycle into a training run for the next.
by armin-ronacher
The plummeting cost of code production makes new programming languages viable again—and the winning ones will optimize for machine readability over human brevity.
by alistair-gray
Stripe's coding agents produce 1,000+ merged PRs per week by combining isolated devboxes, a customized Goose fork, MCP with 400+ internal tools, and aggressive feedback-left testing — proving that enterprise-scale autonomous coding demands custom infrastructure, not off-the-shelf solutions.
by alexander-opalic
Claude Code's agent teams upgrade the subagent workflow from star topology to mesh — agents can now message each other, coordinate through shared task lists, and collaborate in real time. Three real sessions show the patterns in action.
by playwright
Playwright CLI saves 4x tokens over MCP by writing browser data to files instead of piping it into the LLM context, making it the better choice for coding agents that can read from disk.
by ray-amjad
Claude Code's new agent teams combine a shared task list with an inter-agent messaging system—two simple primitives that unlock parallel collaboration, persistent teammates, and real-time coordination between sub-agents.
by tom-crawshaw
Claude Code's native agent teams replace single-agent sequential work with a lead agent that delegates to parallel teammates, each with their own context window and inter-agent messaging.
by amp
The IDE sidebar is a dead-end interaction model for coding agents—parallel, headless agent swarms that run for 45 minutes without human input replace the one-on-one assistant workflow.
by dev-agrawal
Sync engines collapse distributed state management into straightforward SQL by maintaining local databases that sync bidirectionally, and this architecture extends naturally to AI agents that read and write directly to the database.
by lewis-metcalf
Effective human-agent collaboration requires making problems feedback-loopable: build playgrounds for shared exploration, parameterize experiments for reproducibility, and provide headless CLI tools so agents can iterate in text rather than pixels.
by nicholas-carlini
Sixteen Claude instances working in parallel without human supervision can produce a 100,000-line Rust-based C compiler capable of compiling the Linux kernel—but only when the task verifier and CI pipeline are nearly perfect.
by ryan-lopopolo
When AI agents write all the code, engineering shifts from authoring to harness design—crafting navigable documentation, mechanical enforcement, and feedback loops that keep autonomous agents productive across thousands of pull requests.
by karaposu
Vibe-Driven Development transforms intuitive vibe coding into a structured methodology with four pillars—Radical Transparency, Verbose Communication, Explicit Over Implicit, and Human-in-the-Loop—providing repeatable patterns for AI collaboration.
by sunil-pai
Every AI agent framework should be a sync engine—the pull-based architecture agents use today mirrors front-end evolution, and pushing data into agent context solves the fundamental performance problem of MCP tool fetching.
by boris-cherny
Productivity with Claude Code comes from parallel worktrees, plan-first workflows, self-evolving CLAUDE.md files, and skills—multiply your throughput by running more concurrent sessions.
by armin-ronacher
The best coding agents aren't the ones with the most tools—they're the ones that can extend themselves.
by alex-shershebnev
Building AI coding agents requires only basic tooling—a ReAct loop, tool definitions, and MCP integration—and the developer's role shifts from writing code to managing autonomous virtual developers.
by u-other-tune-947
Custom subagents with explicit model assignments and parallel execution now work in VS Code Insiders—conserving 95% of context window by delegating tasks to specialized agents.
by anthony-fu
A curated skill collection embeds ecosystem expertise directly into AI coding agents—keeping context synchronized with upstream documentation through git submodules.
by microsoft
Teaching AI about your codebase through context engineering makes Copilot suggestions more accurate — this hands-on workshop demonstrates the approach using .NET 10 and Blazor.
by michael-bolin
The agent loop—a simple cycle of LLM calls and tool execution—is the core of every AI agent, but performance requires stateless design, prompt caching, and context compaction to avoid quadratic inference costs.
by aicodeking
Kimi K2.5 challenges proprietary models at a fraction of the cost: trillion-parameter MoE with vision, agent swarm parallelism, and 5th place on AICodeKing's benchmark—beating Claude Sonnet 4.5 and DeepSeek V3.2.
by amp
The assistant era is over—agents now write production code. The next frontier is building 'agent-native codebases' with feedback loops that let agents verify their own work autonomously.
by david-ondrej
Remotion's new agent skill lets Claude Code create professional animations by writing React components instead of using complex UI tools—turning video production into a coding problem AI agents already excel at.
by steve-yegge, gene-kim
IDEs and single-agent tools like Claude Code are transitional—the future belongs to swarms of specialized agents, CNC-style automation, and leaders who code their own solutions.
by kieran-klaassen, trevan
Claude Code's native task system converges with Beads and compound engineering patterns, but the real opportunity lies in contextual intelligence—letting machines query for context rather than forcing artifacts into markdown files.
by simon-willison, wilson-lin
A single engineer armed with 2,000 concurrent AI agents can produce over a million lines of Rust code in a week—FastRender proves linear scaling of coding agents is possible through hierarchical planning and emergent coordination.
by langchain
The /remember command lets agents reflect on conversations and extract reusable knowledge—preferences, workflows, and skills—into persistent storage for future sessions.
by microsoft
VS Code agents handle end-to-end coding tasks across four execution modes: local (interactive), background (CLI), cloud (remote infrastructure), and third-party integrations—with unified session management and seamless handoffs.
by alexander-opalic
Workshop covering the transformation from LLM to Agent, context engineering, AGENTS.md, subagents, and skills in VS Code Copilot.
by ray-amjad
Claude Code's new task system—inspired by Beads—persists tasks to disk with dependency tracking, enabling multi-session orchestration and sub-agent parallelism without context window bloat.
by maggie-appleton
When agents handle code generation at scale, design and architectural planning become the bottleneck—implementation velocity no longer constrains development.
by dominik-kundel, gabriel-chua
Core argument: Agent skills are untestable vibes until you build an eval pipeline — define success metrics, capture traces, write graders, and compare scores over time.
by agentic-lab
Slash commands evolve from saved prompts to full orchestration platforms—most users stop at level one, but level six treats Claude as an architect that spawns specialized subagents with domain skills.
by adam-gospodarczyk
MCP's flaws matter less than the fundamental LLM limitations it exposes: context bloat from tool schemas, degraded instruction following in long conversations, and inference costs that balloon with agentic workflows.
by blader
A meta-skill that extracts reusable knowledge from debugging sessions and saves it as new skills, giving Claude Code persistent memory across sessions.
by aws
Lambda durable functions use checkpoint/replay to enable resilient multi-step workflows that can run for up to a year, automatically recovering from failures without re-executing completed work.
by moss-ebeling
Investing in automated feedback mechanisms—build systems, type checkers, visual renderers, formal verification—gives AI agents the ability to self-correct, freeing engineers from manual review.
by alexander-opalic
Learn how to build a conversational AI that queries your personal knowledge base using Nuxt, Nuxt Content, and the Anthropic SDK.
by geoffrey-huntley
Back pressure—automated feedback on quality and correctness—is the key mechanism that enables AI agents to handle complex, long-horizon tasks.
by mattpocockuk, dex-horthy
Live conversation exploring practical approaches to AI-assisted coding, context engineering, and building reliable agents in complex codebases.
by onmax
Portable AI skills bring Nuxt, Vue, and NuxtHub expertise to coding assistants—skills activate based on file context, making agents domain-aware without manual prompting.
by github
Build AI coding assistants by connecting to Copilot CLI through JSON-RPC, letting you embed GitHub's coding agent into any application.
by amp
Tab completion is obsolete because AI agents now write most code—the human-as-coder premise that autocomplete was built on no longer holds.
by geoffrey-huntley
Coding agents are just 300 lines of code running in a loop—demystifying AI tooling reveals that the model does the heavy lifting, and understanding these primitives transforms you from AI consumer to AI producer.
by william-lyon
Enterprise AI value is shifting from systems of record to decision traces — graph databases that capture not just what happened, but why, creating an institutional memory that makes AI recommendations auditable and precedent-aware.
by shu-ding, andrew-qu
Performance problems compound over time, so optimization work should follow a strict priority order: eliminate waterfalls first, reduce bundle size second, then progressively address server-side, client-side, and re-render issues.
by visual-studio-code
Dev containers provide isolated, reproducible development environments that let you run AI coding agents safely separated from your local machine, with instant portability between local Docker and GitHub Codespaces.
by geoffrey-huntley
Understanding how coding agents work—tool calls, inferencing loops, the basic primitives—is now baseline knowledge for software engineering interviews, not optional curiosity.
by indydevdan
Thread-based engineering provides a mental framework for measuring improvement with AI agents—scale by running more threads, longer threads, thicker threads, and fewer human checkpoints.
by simon-willison
Claude Cowork repackages Claude Code's powerful agentic capabilities for general audiences through accessible design rather than technical innovation—a pragmatic approach to unlock untapped value.
by anthropic
Official documentation on Skills—markdown files that extend Claude Code with specialized knowledge and workflows, triggered automatically based on semantic matching.
by geoffrey-huntley
A hands-on Go workshop that builds a coding agent incrementally through six files, each adding one capability - proving agents need only simple primitives composed in a loop.
The Ralph Wiggum technique: a bash loop that runs AI agents autonomously, resetting context each iteration to stay in the 'smart zone'
by geoffrey-huntley
Software development now costs $10.42/hour when running Ralph loops—the key is deterministically malicking the array, starting with a screwdriver before grabbing the jackhammer.
by covibes
Multi-agent validation prevents self-deception in autonomous coding—the validator didn't write the code, so it can't lie about tests.
by peter-steinberger
A self-hosted Claude-powered assistant that runs across messaging platforms like WhatsApp, Telegram, and Slack—turning natural language into automated workflows for email, calendars, and home automation.
by worldofai
Claude Code 2.1 ships a massive feature set—skill hot-reloading, forked sub-agents, Chrome browser integration, LSP support, and async agents transform the terminal into a multi-threaded agent orchestration platform.
by alexander-opalic
Claude Code 2.1 collapses the distinction between skills, slash commands, and subagents—skills can now fork into isolated contexts and specify their own model, making them the single abstraction for agent orchestration.
by alexander-opalic
Claude Code 2.1 transforms skills from specialized knowledge containers into a unified automation abstraction—hot-reloading, forked contexts, lifecycle hooks, and agent specification make skills the default choice for extending Claude Code.
by visual-studio-code
Agent Skills are portable instruction folders that load on demand, transforming AI agents from general assistants into domain-specific experts through scripts, examples, and specialized workflows.
by drew-breunig
Specification-only libraries work for simple utilities but foundational software still requires actual code, community oversight, and sustained maintenance.
by humanlayer-team
Systematic context management—through frequent intentional compaction and a Research-Plan-Implement workflow—enables productive AI-assisted development in complex production codebases.
by langchain
Context engineering—filling the context window with the right information at each step—determines agent performance more than model choice or complex frameworks.
by jediah-katz
Coding agents perform better when they pull context on demand rather than receiving everything upfront—files serve as a simple, future-proof abstraction for this dynamic retrieval.
by ryan-carson, geoffrey-huntley
Ralph ships code while you sleep—a bash loop that pipes prompts into your AI agent, picks tasks from a JSON backlog, runs tests, commits passing code, and repeats until done.
by ryan-carson
Ralph solves the context window limitation by breaking work into independent iterations—each Amp session gets a fresh context, implements one story, and commits before the next iteration begins.
by repomirrorhq
Running Claude Code in headless loops can port entire codebases overnight—the key is simple prompts, one goal per iteration, and letting agents self-terminate when stuck.
by developers-digest
Claude Code can become a self-correcting system by capturing mistakes in CLAUDE.md—each error becomes a permanent lesson that compounds over time.
by dex-horthy
The Ralph Wiggum Technique evolved from a novel idea to a mainstream development methodology in under a year—early adopters learned that poor specs doom loops and iteration beats over-planning.
by thariq-shihipar
Bash is the most powerful agent tool. The Claude Agent SDK packages Claude Code's battle-tested patterns—tools, file system, skills, sandboxing—for building coding and non-coding agents alike.
by burke-holland
AI coding agents can replace developers for many tasks—Claude Opus 4.5's self-correction capabilities make human code review increasingly optional.
by geoffrey-huntley, dex-horthy
The official Anthropic Ralph plugin differs fundamentally from the original technique: outer orchestrators with full context resets produce deterministic outcomes, while inner-loop plugins with auto-compaction lose critical context.
by alexander-opalic
Ralph is a bash loop that feeds prompts to AI coding agents repeatedly—the key is one goal per context window, deliberate context allocation, and robust feedback loops.
by philipp-schmid
As AI models converge in benchmark performance, the infrastructure managing them—Agent Harnesses—becomes the competitive differentiator for building reliable, multi-day workflows.
by gergely-orosz, chip-huyen
Chip Huyen explains how AI engineering differs from ML engineering, walking through the practical developmental path from prompts to RAG to fine-tuning.
by chip-huyen
A practitioner's guide to building applications on foundation models, covering prompt engineering, RAG, finetuning, agents, and evaluation.
by aicodeking
Vibe Kanban is an open-source tool that lets you orchestrate multiple AI coding agents from a visual Kanban board, treating coding tasks as asynchronous jobs rather than blocking conversations.
by cian-clarke
BMAD's spec-driven methodology beats pure prompting for production-ready AI development because it forces clarity before code and catches requirements gaps before they become technical debt.
by steven-bartlett
Tristan Harris warns that AI companies are racing to build a 'digital god' that could automate all human cognitive labor, with insiders believing this will happen within 2-10 years while publicly downplaying the risks.
by mo-gawdat, steven-bartlett
Mo Gawdat, former Chief Business Officer at Google X, warns that AI represents humanity's greatest existential challenge—bigger than climate change—and outlines his 'three inevitables' for why we're approaching a point of no return.
by dex-horthy
A practical framework for getting AI coding agents to work reliably in brownfield codebases through context engineering, intentional compaction, and the Research-Plan-Implement workflow.
by upro-vi
Community discussion on Ralph Wiggum loop trustworthiness—skeptics question validation limits while practitioners share multi-iteration success stories.
by jesse-vincent
A skills library for Claude Code that provides structured workflows for AI-assisted development, including brainstorming, implementation planning, TDD, and debugging.
by varin-nair
Frontier models cap out at 1-2 million tokens, yet enterprise codebases span several million. Factory's solution: a five-layer context stack that delivers the right information at the right time.
by alexander-opalic
A comprehensive breakdown of Claude Code's extensibility layers—MCP servers, CLAUDE.md files, slash commands, subagents, hooks, and skills—explaining when to use each component.
by antonio-gulli
A practical guide presenting 21 design patterns for building AI agents, covering prompt chaining, tool use, multi-agent collaboration, and self-correction techniques with examples in LangChain, CrewAI, and Google ADK.
by mohammad-ghassemi
Michigan State University lecture covering three agentic design patterns—prompt chaining, routing, and reflection—with practical LangChain and LangGraph implementations.
Patterns and practices for building autonomous AI agents with prompt chaining, routing, and reflection
by lee-robinson
Lee Robinson reflects on how coding agents have surpassed his coding ability and shares four recommendations for software engineers adapting to 2026.
by andrej-karpathy, dwarkesh-patel
Karpathy argues we're building 'ghosts' that imitate internet documents rather than evolved animals, and that practical AI agents will take a decade—not a year—to fully mature.
by steve-yegge
A distributed, git-backed graph issue tracker designed for AI coding agents, replacing markdown task lists with structured, dependency-aware workflows.
by maham-codes
A practical guide to building AI agents using prompt chaining and basic primitives instead of heavy frameworks.
by mark-anthony-cianfrani
A first-principles guide to building AI agents in TypeScript, covering LLM integration, conversation memory, tool calling, and agentic loops.
by erik-schluntz, barry-zhang
Anthropic's guide to building agentic LLM systems, advocating for simple composable patterns over complex frameworks.
by john-lindquist
Claude Code operates as a programmable AI platform with isolation, extensibility, and automation as first-class features—not a traditional CLI assistant.
by worldofai
Step-by-step walkthrough for installing and using the Ralph Loop plugin with Claude Code, enabling autonomous iterative development through a stop-hook mechanism.
by daniel-avila-arias
Hedgineer built a knowledge distribution system using Claude Skills that automatically applies institutional expertise without requiring users to invoke specific commands.
by luke-parker
Replace interactive AI chat with structured execution loops—invest heavily in planning, dump full context each iteration, and let verification backpressure catch errors before they compound.
by yacine-mahdid
Prompt chaining breaks complex LLM tasks into sequential steps, each with a specific job and structured input/output, trading latency for reliability.
by humanlayer-team
A manifesto for building production-grade LLM agents, arguing that effective agents combine mostly deterministic software with strategic LLM decision-making rather than naive 'loop until solved' patterns.
by simon-willison
Simon Willison's comprehensive year-in-review analyzing how reasoning models, autonomous agents, and Chinese AI competition fundamentally reshaped the landscape in 2025.
A curated map connecting key resources for understanding AI agents and context engineering
by nate-b-jones
Traditional front-end engineering (hand-coding pages from Figma designs) is being replaced by composability—designing primitives, schemas, and contracts that let AI, PMs, and designers ship dynamic interfaces without reinventing the wheel.
by geoffrey-huntley
A Bash loop technique ('Ralph') that feeds prompts to AI coding agents repeatedly, enabling autonomous project development with iterative refinement.
by steve-yegge
Managing 20-30 parallel AI coding agents requires factory-scale orchestration—hierarchical roles, persistent state in Git, and structured workflow primitives that survive session crashes.
by andrew-qu
Stripping a text-to-SQL agent down to a single bash tool produced a 3.5x speedup, 100% success rate, and 37% fewer tokens—proving that simpler agent architectures outperform elaborate tooling.
by dennis-stanoev
Understanding AI agents requires building one from scratch: an agent is a wrapper around an LLM that makes decisions and takes actions through a simple tool-use loop.
by alberto-salazar
Subagents offer isolated, autonomous task execution within Copilot Chat—running refactors, tests, and security scans in separate Git worktrees without polluting your main workspace.
by microsoft
AI-assisted TDD in VS Code works best through custom agents that enforce Red-Green-Refactor discipline with explicit handoffs between phases.
by xiwei-xu
Context engineering—not model fine-tuning—should be the central challenge for generative AI systems, solved through a Unix-inspired file system abstraction that treats all context components uniformly.
by anthropic
AI coding assistants boost productivity dramatically but create a supervision paradox: effectively overseeing AI requires the deep technical skills that may erode from overreliance.
by mario-zechner
Minimal coding agents outperform bloated ones because frontier models already understand agentic coding—so the harness should stay out of the way with fewer than 1,000 tokens of system prompt and just four tools.
by ryan-x-charles
Markdown has become a general-purpose programming language—AI agents like Claude Code compile structured specifications into working applications.
by anthropic
Context is a finite resource in LLM agents; treating tokens as precious budget rather than limitless capacity enables reliable long-horizon task completion.
by den-delimarsky
Spec Kit provides a structured four-phase workflow for AI coding agents, replacing vague prompting with specification-driven development.
by geoffrey-huntley
Coding agents require only 300 lines of code in a loop with LLM tokens - understanding these fundamentals transforms you from AI consumer to producer.
by thorsten-ball
Building a functional coding agent requires only an LLM, a loop, and a handful of tool definitions—the complexity lies in refinement, not architecture.