The field guide to systems that hold up in real work
Practical notes on automation, integrations, AI agents, and the operational details that decide whether a system works after the demo.
Coverage map
63 articles
Latest field guide
Start with what changed most recently
OpenClaw for business: a practical setup and security guide
A business-first guide to deploying OpenClaw, connecting one useful workflow, limiting access, and deciding what belongs in production.
Read the field guideHermes Agent for business: setup, tools, and production guardrails
OpenClaw vs. Hermes Agent: which fits your business?
A practical deployment checklist for self-hosted AI agents
How to scope AI agent permissions and human approval
Find the answer your system needs
OpenClaw for business: a practical setup and security guide
A business-first guide to deploying OpenClaw, connecting one useful workflow, limiting access, and deciding what belongs in production.
Hermes Agent for business: setup, tools, and production guardrails
How to turn Hermes Agent from a working local chat into a bounded business assistant with controlled tools, channels, memory, and schedules.
OpenClaw vs. Hermes Agent: which fits your business?
A practical comparison of OpenClaw and Hermes Agent across deployment, channels, tools, memory, security, data flow, and operating effort.
A practical deployment checklist for self-hosted AI agents
Self-hosting gives you control over the runtime, but it also makes you responsible for permissions, secrets, recovery, monitoring, and every side effect the agent can trigger.
How to scope AI agent permissions and human approval
Human review only works when permissions are narrow, approval is tied to an exact action, and deterministic controls still enforce the boundary after someone clicks approve.
How to test AI agents before production
A useful agent test plan covers outcomes, tool traces, hostile inputs, side effects, dependency failures, and human review. A handful of polished demos covers none of them.
What building seven MCP servers taught me about useful tool design
Seven public MCP codebases exposed the same engineering lesson: protocol compliance is the beginning. Clear tools, narrow permissions, honest errors, and operable deployments determine whether a server is useful.
How to secure an MCP server for production
A production MCP server needs more than valid protocol messages. This guide separates MCP capabilities from the authentication, authorization, isolation, validation, and operational controls your implementation must enforce.
Local vs. remote MCP servers: how to choose
Local and remote MCP servers use the same protocol but create different data paths, security boundaries, deployment work, and collaboration options. Choose based on who needs access and where the protected system lives.
Browse every automation article
A complete HTML index for readers and search engines. The filters above are still the fastest way to narrow the list.