4,112 saved links, 2017–2026 (peak: 2025). This is your biggest tag by a wide margin — the catch-all that fires whenever a save is about AI but not about any one thing. At this size it isn’t a topic, it’s a firehose, and this page doesn’t pretend otherwise: treat it as the hub that routes to the specialized pages, plus the handful of themes that genuinely live between them — how engineers work now, harness engineering, enterprise architecture, and the strategy takes that shape your Enterprise AI Architect pitch.
Related: LLMs · AI Agents · RAG · Generative AI · Deep Learning · Prompt Engineering · AI Research · NLP · Machine Learning · MLOps
Where things actually live
Most of what lands under this tag has a better home. Agent frameworks, memory, MCP, and multi-agent orchestration → AI Agents. Model releases, context windows, and inference mechanics → LLMs. Retrieval and context engineering → RAG. Image/video/creative tooling → Generative AI. Papers and benchmarks → AI Research. Prompting technique → Prompt Engineering. Theory and courses → Deep Learning and Machine Learning. Getting models served and monitored → MLOps. What follows is the residue those pages don’t cover — and it turns out the residue is where you spend most of your time.
The current moment — how engineers actually work now
The 2026 saves converge on one shift: agents went from autocomplete to colleagues, and the interesting writing is about what that does to you.
- How I use LLMs as a staff engineer in 2026 — the workflow snapshot everyone else is reacting to.
- “How I Use AI to Code” — the payoff is shaping the harness and feedback loops, not reviewing diffs one by one.
- a raw list of how one person actually uses AI, hours a day — rare honesty about the how, not the hype.
- why AI hasn’t replaced software engineers — the decide-execute-deliver sandwich: humans stay at both ends.
- The AI-Native Developer (ACM Queue) — which parts of the job are still worth doing yourself.
- The cost side: Addy Osmani on “cognitive surrender” · the grief when AI writes most of the code · AI-assisted engineers are burning out — is this fine? · how to not feel constantly behind.
- “Reality has a surprising amount of detail” — a 2017 essay resurfacing as advice to AI labs; still lands.
Harness engineering — 2026’s word
The vault watched a term get coined in real time. The model is table stakes; the scaffolding around it is the product.
- Tejas Kumar: harnesses from first principles — why, what, and how to build one; start here.
- awesome-harness-engineering — tools, patterns, evals, memory, permissions, observability — the field’s index.
- a practical guide to AI agent harness engineering and Outcome School’s harness engineering explainer.
- the harness compatibility matrix — “so that I would not go insane.”
- don’t over-engineer your harness — lessons from 34,000 tests — the counterweight.
- Deep Agents — a batteries-included harness if you want to build your own Claude Code, and the primitive behind Claude Code, Codex, and Gemini.
- The investor angle: why the harness is the moat for model providers · GitHub publishing Copilot’s benchmark numbers against rival harnesses.
- the evolution: tests → prompts → agents → harnesses → supervised loops and Fowler’s fragments on AI coding and harness engineering.
Skills — the new packaging unit
Expertise is shipping as markdown folders now. Half your recent repo saves are skills for something.
- Google’s official Agent Skills repository — when both major labs converge on the same primitive, pay attention.
- Addy Osmani’s production-grade engineering skills — trending for a reason — and Ryan Singer’s shaping skills from a “my favorite skills” thread worth mining.
- awesome Claude Skills — the curated list.
- Pinterest: a testing process to optimize skill performance in any repo — skills as engineering artifacts, with evals.
- a skill-quality validator and the hybrid pattern: routing hints in AGENTS.md, heavy logic in skills.
- security skills for agent-assisted testing — appsec, cloud, containers, threat modeling.
Enterprise AI architecture — your lane
The cluster that maps directly onto where you’re headed: reference architectures, gateways, context layers, and the patterns underneath them.
- Google Cloud’s multi-tenant agentic AI reference architecture — centralized platform, autonomous business units; the layers-first shape you think in.
- an enterprise AI gateway with API Management + AI Foundry — the gateway pattern, concretely.
- what an enterprise context layer actually is and agent memory: the missing layer in enterprise AI systems — two takes on the layer everyone’s discovering they need.
- the AI agents stack, 2026 edition — default tools per layer and where the lock-in risk lives.
- the Architect’s V Impact Canvas — designing for deterministic systems coexisting with non-deterministic AI.
- local-first inference escalation with confidence gates — a genuinely clever cost architecture.
- Anthropic’s “Building Effective AI Agents” architecture patterns PDF and The Hitchhiker’s Guide to Agentic AI — the two comprehensive references.
- Codebase-level patterns: reducing friction in AI-assisted development · designing codebases that are safer for AI · architecting for agentic AI development on AWS.
- The skeptic’s corner: multi-agent AI is the new microservices — not everything should be an agent, and security before agents.
AI meets the day job — SRE, ops, delivery
Your DevOps depth is the differentiator here, and the corpus knows it — this is also the research shelf for your agentic incident-management project.
- AI in SRE: where Google deploys agentic AI in operations and a candid talk with Google’s VP of SRE on how AI changes SRE — and how it doesn’t.
- OpenSRE — an open framework for AI SRE agents, plus training and eval environments; connects the 40+ tools you already run.
- “the AI SRE agent revolution: 2026 as the year of autonomous incident resolution” and building an AI SRE agent to analyze production — the practitioner’s version.
- why AI agents fail in production — read before believing the previous two.
- Observability: Datadog’s Lapdog — trace what your agent is actually doing · GenAI observability with OpenTelemetry · AI agent observability, explained.
- Delivery: what an AI-native CI/CD experience looks like · tracking agent lineage and state in repos — a git commit doesn’t carry enough info anymore.
- “Software 2.0” for DevOps — planning loops, verification-first workflows, agents coding while you sleep.
Strategy, economics & the org question
The takes you save for leadership conversations — what agents do to companies, careers, and the industry’s economics.
- Chesky on redesigning Airbnb for the AI era — your note calls it hall-of-fame material, and Sidu Ponnappa on running an AI-native company.
- AI-Native Leaders: the organizational playbook and DORA on measuring the ROI of AI in software development — J-curves, not miracles.
- Hassabis: agents are a “practice run” for AGI and what 81,000 people’s economic hopes and worries about AI look like.
- The sober counterweights: half of gen-AI projects overrun and get abandoned — same as big IT ever was · Russinovich and Hanselman on AI hollowing out the junior pipeline · the era of free AI is ending.
- Business models: AI labor is not SaaS · Cognizant’s AI generating $200M in new pipeline · labs buying dead startups’ Slack threads as RL-gym feedstock — the strangest save in the tag.
- The India thread: AI coding as an existential threat to the Indian tech sector vs. India’s low-resource blueprint for sovereign AI — both sides of a question close to home.
- Thoughtworks Tech Radar: 81 of 118 entries are AI — the moment, quantified.
Learning the craft
The full curriculum lives in Learning Resources; these are the AI-engineering-specific standouts.
- the free, open-source AI Engineer book and the “Agentic AI Engineer in 2026” roadmap — the two maps of the territory.
- the JAX scaling book — how LLMs interact with hardware at scale; the production-engineer’s view.
- Hamel’s 11 links for learning AI evals and his case that evals are the hottest new skill — the skill your saves keep circling.
- an AI primer: 11 topics every software engineer should know — context, embeddings, tokens, evals, agent loops.
- AI Hero’s engineering course with an evals framework · the open-sourced AI Engineer workshop · GitHub’s Agentic AI Developer certification (GH-600).
- And the meta-thread: building Karpathy’s “LLM Wiki” — AI-maintained knowledge bases and an example LLM wiki built in Obsidian — you saved the blueprint for the very page you’re reading.