2,930 saved links, 2015–2026 (peak: 2023). Your second-biggest tag and the vault’s catch-all — anything about writing software that didn’t have a more specific home landed here for a decade. Treat this page as a router: the language and tooling clusters live on their own pages, and what remains is the story the recent saves actually tell — a near-total tilt toward AI-assisted coding. The fact that simonwillison.net sits among your top sources for a tag this old is the tell: “programming” quietly became “agentic engineering” somewhere around 2025.
Related: Python · Go · JavaScript & Web · Git · Linux & CLI · AI Agents
The tipping point — what happened to the job
The 2026 saves are one long argument about what a software engineer even is now.
- Software engineering at the tipping point (video) — the note says: “if you have 40 minutes, watch this. Don’t speed it up.”
- The AI-Native Developer (ACM Queue) — which parts of development are still worth doing yourself; the academic version of the question.
- Addy Osmani on the “new SDLC” — the phases, redrawn for the autocomplete-to-autonomous-agents progression; his comprehension debt piece is the necessary counterweight.
- Anthropic’s research on how agentic coding amplifies some skills and substitutes for others — actual data instead of vibes.
- Armin Ronacher: what I still don’t let agents ship — code needing long-term maintenance stays human; the most credible line-drawing you’ve saved.
- Vibe coder vs. software engineer and Willison: vibe coding and agentic engineering are converging — the definitional skirmish.
- Rachel Thomas on the psychology of vibe coding — why it trips people up even while making them faster.
- Pragmatic Engineer research: how GitHub fumbled its agentic-coding lead · Bloomberg on AI coding as an existential threat to the Indian tech sector — the note admits: “I’ve been waiting for articles that point this out.”
- Building in the age of collaborative coding — how mature teams work with agents; the note’s plea: “More please!”
Agentic engineering — the practice
The craft beneath the hype: harnesses, feedback loops, and supervision. This is layers-first thinking applied to coding — exactly your angle.
- The better question: “what feedback can the agent use without asking me?” — types, tests, lint, traces; the whole discipline in one reframe. Same author on the evolution from prompts to harnesses and supervised loops.
- Harness engineering, the resource — “a capable model is not enough”; and the distinction that stuck: “one gives you a conversation, the other gives you a harness.”
- The harness design philosophies of Claude Code and Codex, compared · VS Code’s behind-the-scenes look at its own harness optimization.
- Birgitta Böckeler: sensors in an agent harness — static analysis as the agent’s nervous system; part of martinfowler.com’s steady AI-coding series (the running fragments).
- 10 lessons for agentic coding · kdy1: running 15+ agent sessions without cognitive overload · Steinberger’s refactor loop — “refactor, live-test, autoreview, commit, repeat.”
- Philipp Schmid: you wouldn’t ship code without tests, so why ship skills without evals? and Pinterest’s testing process for agent skills — quality control for the new artifacts.
- The Context Development Lifecycle (CDLC) — the DevOps parallels, drawn by someone who lived DevOps; and tracking agent lineage in git — “a commit doesn’t carry enough info to piece everything together.”
- Worktrees and parallel branches became agent infrastructure — why git worktrees matter to modern developers; the deeper git material lives on Git.
The Claude Code shelf
Your densest 2026 cluster by far — courses, setups, and the skills ecosystem. The broader agent-building world is on AI Agents; this is the daily-driver material.
- Courses: Frontend Masters × Anthropic — the free Claude Code course (Lydia Hallie on the agentic loop, CLAUDE.md, skills, hooks, subagents) · Claude Code 101 and Claude Code in Action · the Coursera version.
- Setups worth stealing: “I spent 6 months tuning Claude Code” — the exact setup · Marco Lancini’s setup — guardrails, context/plan/code workflow, StarCraft-themed statusline · Beyond the prompt: stop prompting, start operating.
- Reference repos: claude-code-guide — the high-level manual · awesome-claude-md — 104 CLAUDE.md examples from real projects · the 10-repos-instead-of-tutorials list.
- Skills, the new package format: Anthropic’s official skill-creator skill · superpowers — the agentic skills framework that hit ~100k stars · Addy Osmani’s production-grade engineering skills · 180+ Claude Skills for software engineering · rules distilled from classic programming books, for agents — the canon, compiled into CLAUDE.md form.
- Claude Code at scale — lessons from multi-million-line monorepos and decades-old legacy systems; directly relevant to your enterprise-architecture trajectory.
Parallel agents & orchestration
The frontier of the tag: one engineer, many agents. The serious frameworks live on AI Agents — these are the coding-specific takes.
- Adrian Cockcroft: AI swarms doing “days of work in 15 minutes” — spend less time coding, more time orchestrating.
- Running Claude Code agents in parallel — the practical how-to.
- Superset — 10+ parallel coding agents on your machine · an IM-style workspace where Claude Code, Codex, and OpenCode collaborate · a Telegram bot driving a five-role Claude Code dev team from planning to PR.
- LinkedIn engineering: managed Claude vs. Agent SDK — the note relishes that it’s “by the creator of Kafka, not LinkedIn slop.”
The craft that doesn’t change
Agents or not, the fundamentals still decide whether the system survives.
- How to write effective software design docs — lessons from years of design docs at big shops; arguably more important now that specs drive agents.
- Laws of Software Engineering — the eponymous laws, collected; briefly #2 on Hacker News.
- How branches influence the performance of your code — “the writeup so good”; mechanical sympathy survives the agent era.
- Debugging event-driven systems: the 5 problems teams create for themselves — dead-letter queues and eventual consistency, the unglamorous truth.
- Design Patterns That Deliver — patterns rebuilt as a written course with production examples.
- LikeC4 — live architecture diagrams from code — diagrams that can’t drift.
- Sierra’s AI-native engineering interview process — how hiring changes when coding agents are the default.
CS foundations & learning resources
The evergreen shelf — see Learning Resources for the full library.
- Knuth used Claude to solve an open problem in graph theory — the father of algorithm analysis, pair-programming with an LLM. If he’s in, the debate’s over.
- Visual DSA: VisuAlgo · Georgia Tech’s CS visualization tool — every data structure animated · Algorithm Visualizer — live visualization from your own code.
- By doing: Coding Challenges — 80+ real-world build-it-yourself projects (including “build your own AI agent”) · learn-by-playing games for Git, Linux, Python, and — your home turf — Kubernetes.
- project-based tutorials, curated — the classic “build your own X” list.
- The Valley of Code — with the author’s own advice that learning to code now means building alongside an agent and having it explain as it works.
- Justin Math’s free textbook — from “the most advanced high school math/CS sequence in the USA”; University of Helsinki’s Python MOOC is on Python where it belongs.
Voices worth following
The bylines that keep recurring across 2,930 links — your de facto editorial board.
- Simon Willison — the tag’s dominant voice. His Agentic Engineering Patterns guide is the spine: chapters on using git with coding agents, having agents build interactive explanations to fight cognitive debt, and “hoard things you know how to do” — career advice disguised as agent advice. Plus his Pragmatic Summit fireside on agentic engineering.
- Drew Breunig — the system-prompt archaeologist: what the Claude Code source leak revealed about context assembly and system prompts across 6 coding agents, compared.
- Peter Steinberger — “I ship code I don’t read” — the note: “probably one of the best interviews I’ve ever seen. Top levels of mastery here.”
- Margaret-Anne Storey — the cognitive debt essay: “the humans involved may have simply lost the plot.”
- Mikayla at Zed: “your name is on the code — make sure you can stand behind it” · Thorsten Ball’s Register Spill · Isaac Flath: “AI wrote every line, but I directed everything”.
- Mario Zechner’s DIY coding-agent reading list — if you ever build your own harness, start here.
- Tim Dettmers on being productive and impactful with limited resources — the note calls it a “wonderful website/journal.”