The Operator Journal
How we actually run a business on AI — receptionists, agents, intake, cold email. No theory, no hype. The systems we use on ourselves first, with the receipts and the failures.
Run Claude Code Headless: -p Mode, CI Pipelines, and Scripts That Don't Hang
Headless mode is how you put an agent inside a shell script or a CI pipeline. It is also where the sharp edges live, because everything the interactive UI does for you — permission prompts, visible progress, a human noticing the agent is stuck — is gone.
Claude Code Slash Commands: Custom Commands, $ARGUMENTS, and Where They Went
A markdown file becomes a /command you can run with arguments — the cheapest leverage in Claude Code. But commands are a moving target since the skills merge: locations, frontmatter, and collision rules all changed. Current state, verified against the docs.
Vercel Deployed, but Your Domain Still Serves the Old Version
The deployment exists, the dashboard says Production, promote says 'already current' — and the apex domain is serving a build from an hour ago. We have logged this repeatedly in production. One command fixes it; the verification routine is what stops it from lying to you again.
Playwright and Your Real Chrome Profile: Why It Stopped Working, and What Works Now
The old trick — debug port on your everyday Chrome, connect over CDP, automate with your real logins — was deliberately killed for security. If your script cannot attach or launches signed out of everything, this is why. Here are the patterns we actually run in production.
npm README Not Updating? The Registry Serves What You Published, Not What You Pushed
You fixed the README, pushed, GitHub shows the new version — and npmjs.com still serves the old one, indefinitely. Nothing is broken and nothing is coming: the registry does not read your repo. The mechanism, the auth trap, and the receipt that proves docs actually shipped.
Claude Code Hooks: Why Yours Isn't Firing, and How to Write One That Does
A hook that does not fire produces no error, no log line, no red text — it just quietly doesn't guard. Every silent hook we have hit in production comes down to one of seven checks. Here they are, fastest-catch first.
Claude Code Subagents: What They're For, and When They Make Things Worse
Used right, subagents let one session read a fifty-file codebase without drowning its own context. Used wrong, they are a way to pay tokens for a worse answer. The difference comes down to one question.
How to Make Claude Code Remember Things Between Sessions
Tomorrow's session starts from zero and cheerfully re-suggests the thing you rejected yesterday. Claude Code has four real persistence mechanisms; the setups that feel like the agent knows the project use all four on purpose.
How to Write a Claude Code Skill (SKILL.md), From Zero
A skill loads on demand instead of paying context rent in every session — that is the whole point. Here is the anatomy, a real example, and the three mistakes that make a well-written skill never get used.
Why Not Just Use ChatGPT? An Honest Answer From People Selling the Alternative
The objection is correct: ChatGPT will write you any playbook we sell, in an afternoon, for free. What it writes is version one. Here is what version one did to us, three times, with dates.
The 366-Clip, $0 Mistake: Verify the Demand Before You Batch
Eleven days of confident production against an enrollment record nobody re-checked on the platform itself. What it cost, where the same shape showed up twice more in the same month, and the twenty-minute gate we now run before any batch.
Is an AI Operator Course Worth $97? Run These Five Checks First
Every check in this guide comes from a failure we caught in our own paid catalog — stub PDFs, a countdown that sat false for three weeks, files four months staler than their sources. Run them on any course. Including ours.
Why AI Automations Break in Production: Six Failure Modes From Our Own Log
Ours broke too. We wrote every incident down — 83 entries and counting — and the same six shapes keep coming back. None of them look like an error. All of them look like success.
Why Your AI Agents Act on Stale Facts (and the One-File Fix)
A fact that lives in two files is already two facts. The agent executing the stale copy doesn't get suspicious — it gets confident. Four failures from our own ledger, and the one-file pattern that ended the class.
You Bought the Prompt Pack. Nothing Changed. Here's Why.
A prompt is one-fifth of an automation — the cheapest fifth to acquire. The missing four (trigger, inputs, destination, review) are why nothing changed, and none of them come in a pack.
What to Delegate to AI Agents — and What to Keep
Most people sort the work by how much they hate it — which hands an agent the highest-stakes tasks first. The two questions that actually sort it, the verb gate that holds the line, and what the week becomes.
How to Build an AI Stack: The Five Layers That Matter
Most stacks are assembled by accretion — a subscription here, a workflow tool there — until you own three tools in one layer and nothing in the two that actually fail. The decision problem, layer by layer.
How to See Which MCP Servers Are Actually Running
A server can be configured and not connected, connected in one session and absent from the next, and running as a process long after every client that asked for it has quit. Three checks, in order.
Claude Code MCP Server Not Loading: The Diagnostic Order
Most advice for this is a list of things to try in no particular order, which is how people end up reinstalling Node at 1am. Each step's answer determines whether the next one can even be true.
Claude Code Permission Rules That Look Right and Do Nothing
A permission rule can be valid, accepted at startup, and never consulted. Nothing distinguishes a rule that is protecting you from a rule that is decoration. Here are the shapes that fail.
Orphaned Claude Code Processes: Kill Only the Dead Ones
The usual advice works the way a grenade works. It also kills the session you are in, its subagents mid-write, and processes that merely mention a .claude path.
How to Run Multiple Coding Agents on One Machine (Without Collisions)
One agent is a tool; four on one laptop is an operations problem. The four collisions you will hit — shared checkouts, unowned ports, leaked processes, stale state — and the procedure that prevents each.
Your Coding Agent Says It's Done. It Isn't.
Asking the agent to double-check is the same model grading its own homework. The fix sits outside the model and touches the outside world before the turn is allowed to end.
Why Your Context Window Is Half Full Before You Type
Sessions compacting sooner, odd tool choices, a model that got worse after you set it up properly. Often the cause is ~95K tokens of preamble you never chose. Here is the audit.
What Does That Repo Tell Your Agent to Do?
You review the code you clone. Almost nobody reviews the part of it that talks to the agent — including the nested instruction files that only apply deep in the tree.
How to Write a CLAUDE.md File That Actually Gets Followed
Most CLAUDE.md files are written once, bloat to 400 lines, and quietly stop working. The load order, the four scopes, and the categorical mistake underneath it: CLAUDE.md is context, not enforcement.
How to Make AI Writing Sound Human (Without a Banned-Words List)
The tells, why forbidden-word lists fail, and the actual fix: writing that contains information only you could have supplied. Includes the two prompts we run on our own copy.
Claude Code for Non-Developers: Run Your Business With It (No Code)
It is marketed at engineers, but it is the most capable operations assistant $20-a-month buys. The non-developer setup, the real jobs, and the safety rules — from a company that runs on it daily.
ChatGPT vs Claude for Small Business: An Operator's Honest Comparison
Not a benchmark afternoon — a comparison from daily operational use. Where ChatGPT's breadth wins, where Claude's stamina wins, and why the documentation you feed either matters more than the logo.
How Much Does AI Automation Cost for a Small Business? (Real 2026 Numbers)
The number-by-number answer the agency pages won't give you: $30–$150/month for a working stack, two evenings of documentation, and a spending order that makes most five-figure quotes unnecessary.
How to Set Up an AI Receptionist for a Small Business (Step by Step)
The full build sequence, in the order that avoids the hang-up-inducing mistakes: script first, conditional forwarding, calendar constraints, handoff rules, and the ten-call gauntlet before launch.
Claude Prompts for Cold Email: 7 Complete Prompts You Can Copy
Copy-paste prompts with the constraints built in: banned phrases, hard word caps, and a no-invented-facts rule. The difference between output you send and output you rewrite.
AI Automation for Solo Founders: What to Automate First (and What Not To)
Written from inside a company that actually runs this way. The one rule (repetition, not judgment), the 30-day sequence, and the failure we shipped so you don't have to.
The Best MCP Servers for Claude: The 7 We Actually Run
Not a 50-item link dump: the working stack — files, browser, GitHub, search, memory, Slack, Notion — plus the three setup mistakes that eat afternoons and the security paragraph most lists skip.
How to Plan a Year of Social Media Posts for a Service Business
A year of posts is not 365 ideas — it's structure: inventory once, four formats, twelve themes, one batching prompt, and a 20-minute weekly loop. Sized for a busy owner, not a marketing team.
AI Prompts That Actually Run a Business (With Two You Can Copy)
Chat prompts get you essays. Operating prompts get you work. The anatomy that separates them, two complete prompts to copy today, and the four reasons yours keep failing.
Stop Prompting, Start Writing SOPs: Reliable AI for Business Operators
Brilliant Tuesday, useless Thursday — that's not the model, that's the spec. How to convert recurring tasks into procedures, with a worked before/after example.
Do AI Receptionists Actually Work for Small Business? An Honest Breakdown
Not a sales pitch — a field report from the phone line we actually run. Where AI receptionists earn their keep, where they blow up, and the one setup detail that decides which one you get.
How One Person Runs a Whole Company on AI Agents
One person, a fleet of agents, and a company that ships every day. The honest version — what actually works, what broke, and the docs-first habit that does 90% of the job.
The Cheapest Way to Automate Customer Intake in 2026
From the $0 version to the full 24/7 build. What we run, what it costs, and the five-minute rule that matters more than any tool you buy.
AI Cold Email That Doesn't Land in Spam
AI writes the email in seconds. Then it lands in spam. Here's the gap between 'generated' and 'delivered' — and how to close it before you send a single message.
Affiliates
Write about this stuff? Earn 30% on every sale.
If you run a newsletter, a channel, or a site covering AI and automation, our playbooks are built to be recommended. Self-serve signup, 30% flat commission, links you can use the same day.
Get the Operator Field Kit — free
Six production prompts, the five-step operator setup, and nine rules from our own failure log.
- 6 complete prompts — printed in full, not previews
- The five-step setup, each step with a pass/fail test
- 9 rules from the failure log that produced them
The kit, then the occasional operator note. One click unsubscribes and we never sell the address.