Codex rewrites three files you never mentioned and reports "Done." The problem is almost never the model — it is the assignment. Seven observable checkpoints for writing clearer task briefs, spotting drift in real time, and verifying output before you merge anything.
Every AI agent for business assumes you are holding the queue. Sixteen roles, more than thirty agent windows, and I talk to one of them. The dispatcher pattern that stops you being the bottleneck — the four things it must ask permission for, and the one thing it must refuse to do itself.
Skills, Subagents, and Hooks answer three different questions and get confused for one another constantly. Reusable procedure, parallel execution, or an enforced check? Three questions asked in order route any task correctly, with beginner pitfalls named.
Keeping AI systems running in production: deployment, permissions, MCP wiring
A system prompt cut from 5,000 tokens to almost nothing, and what that reveals about where the weight actually lives. Read it for the architecture, not for the recipe.
An architecture study, not a recommendation: wrapping the Claude Code CLI as a local OpenAI-compatible endpoint, five layers deep, with the account and Terms of Service risk stated before the design, and what a paid seat does not let ai automation tools reuse.
The bad failures are the quiet ones. The agent did not crash, it just stopped, and nobody on the team noticed until Monday. Opening the log is the wrong first move: the agent already wrote down what it was doing, in plain language, in its own channel. Two small scripts put that in front of you.
Delegated work without a receipt is not delegated, it is just moved somewhere you keep checking. Three ways to close the loop: Hooks for a quick script, Channels for two-way chat, and a Hermes MCP bridge for cross-device push, with config code for each.
Permission prompts are not friction - they are what makes file-editing safe for real projects. Six modes compared, allowlist and denylist rules, the five-layer config priority chain, the OS sandbox as a second defence, and the one setting business process automation should never use.
One clever message turns a permission nobody remembers granting into an incident. Three stacked layers keep an ai agent for business safe even after the model has been talked around, and the day to install them is the day you open the first channel.
Picking wrong here is not a technical mistake, it is a switching cost you pay for months. Three axes turned out to matter, and only one of the six held up on all three.
Twelve MCP servers tested over three months: five earned a permanent spot, seven got removed. The short list for non-programmers, with real usage frequency, an honest reason for each cut, and a copy-paste .mcp.json.
Syncthing, a Git repo with cron, or iCloud Drive plus one symlink - I ran all three across six months. Only one avoided becoming a part-time sysadmin job. Honest comparison, exact commands for the winner, a Gist backup plan, and when heavier workflow automation software is right.
When I rebuild one with AI agents, you get the write-up — including the parts that didn't work. No weekly roundup, no "5 tools you need."