Building production AI agents in 2026: what MCP, agent SDKs, and Search are changing
AI agents are moving from demos to production workflows, and MCP plus newer SDK features are making the connector layer and runtime rules more important.
Category
AI Agents & LLM Apps
27 matching blog articles. Articles on agent workflows, model integration, prompting, and operational AI app design.
Category wiki
AI Agents & LLM Apps is the broad subject lane for this part of the archive.
Definition
Practical notes on building agentic systems and LLM-powered applications.
What belongs here
Articles land in AI Agents & LLM Apps when the main subject is agent workflows, model usage, prompt systems, and LLM application patterns built for production..
How to read it
Treat this category as the broad lane first, then use tags to narrow that subject down to the concrete technologies, platforms, or patterns used inside it.
Common tags in this category
Human-in-the-loop systems are architecture, not a checkbox
Human-in-the-loop AI works only when the review point is designed into the workflow state, risk model, audit trail, and recovery path.
Hermes vs Workflow Tools: When to Use Cron, Skills, and MCP
Hermes is not a replacement for deterministic workflow tools, but it is a strong layer for flexible tasks that need judgment, tools, memory, and scheduled execution.
How Hermes Turns a Terminal Agent into a Cross-Platform Automation Layer
Hermes becomes more useful when it is treated as an automation layer that lives across chat platforms, scheduled jobs, and remote machines.
Hermes Agent and the Case for Self-Improving Personal AI Agents
Hermes Agent stands out because it treats memory, skills, and session recall as core infrastructure for personal AI agents, not optional extras.
Building pi in a World of Slop — Mario Zechner
A useful talk from Mario Zechner about building Pi with a stronger product philosophy in a market full of shallow AI tooling and repetitive agent hype.
The "PI" (coding) agent is so much more than just another amazing coding agent!
A useful video overview of the Pi coding agent and why its extensible, terminal-first approach stands apart from more closed coding-agent tools.
How to Evaluate Pi's Model Catalog Before Picking a Coding Agent Setup
Pi's model catalog is useful because it makes provider choice, context limits, and price tradeoffs visible before developers commit to one coding-agent workflow.
How Pi Packages Turn a Terminal Agent into a Personal Toolchain
The Pi package ecosystem matters because it turns a minimal terminal coding harness into something much closer to a personal agent toolchain.
What pi.dev Is and Why It Matters for Terminal Coding Workflows
Pi positions itself as a minimal terminal coding harness, and that focus matters for developers who want agent tooling to fit their workflow instead of replacing it.
Richard Sutton — 'LLMs are a dead end' (a different, useful view)
Why Richard Sutton — a father of reinforcement learning — calls current LLMs a 'dead end', and what that means for researchers and practitioners.
How to Plan an OpenClaw Agent Workflow With Channels, Memory, and Guardrails
OpenClaw works best when channels, memory, and guardrails are planned as part of the workflow, not added after the first prototype.
When to Use LangGraph, LangChain, and LangSmith in One AI Stack
LangChain, LangGraph, and LangSmith solve different problems, and the stack is clearer when each layer has a specific job.
Redis in 2026: Cache, Sessions, Realtime, and AI Memory
Redis is still the fastest way to add caching and state, but its 2026 value now extends into vectors, AI memory, and streaming.
How Ubuntu 26.04 Is Fixing AI Development
Ubuntu 26.04 LTS improves the security, container, and retrieval layers that AI teams keep fighting during development and deployment.
What Cloudflare Shared Dictionaries Mean for Faster Agentic Pages
Shared Dictionaries are a compression idea that can make repeated content and agent-heavy pages cheaper to deliver.
Why LangGraph Is the Right Layer for Stateful AI Workflows
LangGraph is built for long-running, stateful agent workflows where durable execution, human review, and controlled resumption matter.
How to Go From Prototype to Production With LangChain
A practical production checklist for LangChain apps that need tracing, evaluation, integration boundaries, and a realistic path to deployment.
LangChain vs LangGraph vs LangSmith: Which Layer Do You Need?
LangChain, LangGraph, and LangSmith solve different problems in the same ecosystem: application building, orchestration, and observability.
What Is LangChain and When Should You Use It?
LangChain is a fast way to build custom LLM applications and agents, especially when you want a practical starting point instead of a blank orchestration layer.
How to Choose OpenClaw Channels, Models, and Trust Boundaries
How to decide which OpenClaw channel, model, and trust boundary setup fits personal assistants, team assistants, and more sensitive workflows.
OpenClaw Security Checklist for Personal and Team Use
A practical security checklist for OpenClaw deployments, including allowlists, sandboxing, reverse proxies, secrets, and trust boundaries.
How to Set Up OpenClaw Safely on a VPS or Local Machine
A practical guide to installing OpenClaw, running onboarding, and choosing a deployment model that matches your privacy and availability needs.
OpenClaw vs n8n and Zapier: When to Use an Agent Instead of a Workflow Tool
OpenClaw and tools like n8n or Zapier solve related but different problems: OpenClaw is agentic and chat-first, while workflow tools are deterministic and trigger-driven.
How OpenClaw Works: Channels, Gateway, Memory, and Skills
OpenClaw combines a gateway, persistent memory, chat channels, and skills so the assistant can accept requests and act on them over time.
What Is OpenClaw and Why It Matters for AI Workflows
OpenClaw is a chat-first, open-source AI agent platform that runs on your machine and can execute real tasks instead of only generating text.
Apache Swap Thrashing: Why Over-Tuned Servers Crash
Learn why aggressive Apache limits can push a server into swap thrashing and how that leads to slowdowns, instability, and crashes.