Your single biggest AI topic: 1,891 saved links, almost all from 2023 onward, with a sharp acceleration through 2025–2026. The corpus tracks the field’s evolution from “chatbot with tools” to production agentic systems — and it leans heavily toward the practitioner side: harness engineering, Microsoft’s agent stack, MCP, and running agents in production. That maps directly to your DevOps → AI/ML transition: the corpus says your angle on agents is operating them, not training them.
Related: LLMs · RAG · Prompt Engineering · MLOps
Start here
The five links that give the best current mental model of the space:
- Learn Harness Engineering — “The better the harness, the better the agent.” The free site you saved twice; the practitioner’s frame for everything below.
- Simon Willison: Agentic Engineering Patterns — evolving book of patterns, updated weekly since Feb 2026.
- 10 Lessons for Agentic Coding — dense, experience-based rules of thumb.
- Connecting LLMs to the Real World: Tool Use, Function Calling, and MCP — the conceptual ladder from function calling to full agent connectivity.
- Anthropic’s guide to building productive agents — the vendor guide worth reading end-to-end.
Harness engineering
The theme your 2026 saves circle obsessively: the model matters less than the scaffolding around it — feedback loops, sandboxes, permissions, context plumbing.
- Agent Harness Engineering Explained Local vs Frontier LLMs — video overview of harness vs. model quality.
- Practical guide to AI agent harness engineering and the awesome-harness-engineering list.
- Building an agent harness — a from-scratch walkthrough.
- VS Code: behind the scenes of harness optimization — how a major product team tunes theirs.
- Kong: lessons from 34,000 tests — don’t over-engineer your harness.
- The agent harness distributed-feedback problem — why multi-program systems are harder.
- Birgitta Böckeler: sensors in an agent harness — static analysis as agent feedback.
- “What feedback can the agent use without asking me?” — the right question to ask; types, tests, lints as the answer.
- Guarding against disruptive actions in looped agents — safety rails for autonomous runs.
MCP (Model Context Protocol)
The connectivity standard that dominates your 2026 saves.
- Pamela Fox: MCP workshop slides + exercises and her PyCon “Build your first MCP server in Python” tutorial.
- Phil Schmid: using MCP servers with agents — practical advice amid the hype.
- Agentic Workflows and MCP — how the pieces compose.
- Microsoft’s MCP servers you saved: Azure Resource Manager MCP, Data Factory MCP, Fabric Ontology MCP.
- GitHub cut agent token usage 62% — partly by pruning unused MCP tools; the ops view of MCP sprawl.
Agent Skills
The 2026 pattern for packaging reusable agent expertise.
- Addy Osmani: Agent Skills — specs, tests, reviews as enforceable workflows.
- Anthropic’s free Agent Skills course — skills vs. tools vs. MCP, from zero.
- Google’s official Agent Skills repository and the 13 launch skills (cross-agent standard: Claude Code, Gemini CLI, Copilot, Cursor).
- Collection of verified Agent Skills for Claude Code.
- waza: Go CLI for agent skill evals — A/B baselines, pairwise judging.
- Structuring Agents, Skills, and MCPs: Best Practices from Anthropic.
Frameworks & platforms
Your corpus covers every major stack, with a strong Microsoft center of gravity.
Microsoft (your deepest coverage):
- Microsoft Agent Framework documentation and Foundry Agent Service + Agent Framework overview.
- Foundry adds runtime, tooling, governance for production agents and CI/CD for AI agents in Foundry.
- AgentLoop middleware —
.ralph(),.with_predicate(),.with_judge(). - Building agentic systems with Agent Framework + Foundry IQ and on-prem agents with Azure Local + Foundry Local.
- Azure Functions serverless agents runtime · Logic Apps sandboxed code interpreters · Azure Skills: 25 skills for coding agents in Azure.
Google: Agent Development Kit codelab · How Gemini Managed Agents work · Agent Executor open-sourced · multi-tenant agentic reference architecture.
Others: Vercel’s eve — “Next.js for agents” (filesystem conventions over config) · Pydantic production-grade agents · Cloudflare Agents SDK · CrewAI tutorial.
Build one from scratch (the best way to actually understand them): teaching-grade agent in TypeScript · multi-agent systems from scratch · why building from scratch teaches you the stack around the model.
Multi-agent systems & orchestration
- Multi-Agent Interaction Patterns (Microsoft repo).
- Grab case study: multi-agent engineering support at scale — the real-world benchmark you saved twice.
- Shopify: from chatbot to specialized agent swarms.
- Git worktrees and why your agent should use them — parallel agent isolation.
- Orchestration platforms to watch: Alook (agents as a managed team) · IM-style multi-agent workspace · real-time node graph for Claude Code orchestration.
Production: evals, observability, security
The DevOps-shaped section — where your existing expertise transfers directly.
- Why agents fail: Why AI Agents Fail in Production · Hidden Technical Debt in Agentic Systems · Salesforce’s lessons from enterprise agent deployments.
- Evals: AI Agent Evaluation · three working eval demos · measuring agent hallucination.
- Observability: AI Agent Observability · Tracing Agent Sessions with OpenTelemetry & Aspire · structured logging of agent decisions.
- Security & guardrails: OWASP Top 10 for Agents · AI security skills for agent-assisted testing · why guardrails aren’t enough · human-in-the-loop approval for custom agents.
- Memory: AI Agent Memory · visual survey of agent memory · where to store agent memory: files, blocks, skills.
- AI for SRE (your two worlds meeting): Google: deploying agentic AI in SRE · building an AI SRE agent · the AI SRE agent revolution.
Agentic coding
- Karpathy’s “Idea File” — hand it to your agent, it builds the tool.
- Using agents is literally engineering management — the mental model shift.
- The AI Coding Agent Manifesto (Wix) — beyond vibe coding for production.
- Spec-driven development with coding agents (free course).
- “How I Use AI to Code” — shaping the harness and feedback loops is the work that pays off.
- Armin Ronacher: what I don’t trust agents to ship — the sober counterpoint.
- The 100x engineer: one senior + their agents.
Learning paths
- Hugging Face AI Agents course (free) and 10 free tools to learn AI agents.
- 22-chapter course: production-grade agent systems — designed to be studied with an AI assistant.
- Microsoft Agent Academy + hackathon.
- GitHub Agentic AI Developer certification (GH-600) — a credential that fits your profile.
- Roadmap to Become an Agentic AI Engineer in 2026.
- Claude Cookbook — 81 practical guides across agents, tools, RAG, evals.
All links on this topic
Requires the Dataview plugin: