You're Not Automating. You're Just Doing It Fast.
You learned to prompt. You learned to build a skill. Now you can tell Claude "do X" and it nails it. So why does it still feel like work? Because you're still the one who decides when. You see the email land. You remember it's Monday. You open the terminal and run the thing. You're the trigger, the scheduler, AND the supervisor—doing by hand the one part a computer is actually best at: showing up on cue. That's not automation. That's a really fast manual process. ——— ⚓ There are only two triggers to learn To get yourself out of the loop, you only need two things: - Cron → fire a job on a schedule (every 15 min, every weekday at 6am). For "do this every X." - Hooks → fire when something happens (a session ends, a file changes, a tool runs). For "do this when X." Master those two, plus chaining a few skills together, and Claude Code becomes the orchestrator. Not you. ——— ⚓ A real example A solopreneur kept hand-fixing malformed credential numbers people typed wrong at signup. Annoying, recurring, not life-or-death—the perfect first thing to automate. We built a small agent (not a website rewrite) that runs on its own: spot the bad record → check it against the official registry → fix it if there's a confident match → email a summary. He went from janitor of his own database to getting a quiet email that says handled. That time went straight back into his actual business. The kicker: once you build one identify → verify → remedy → notify chain, you own the method, not just the fix. The next chore drops into the same shape. ——— ☠️ Do it safely - Pick ONE narrow, repetitive, low-stakes task first - Only let it act automatically on a confident match—otherwise report - Make failures LOUD (alert on errors, not just successes) - Test against a fake record before it touches live data ——— 🗝️ The takeaway If you're deciding when your tasks run, you haven't automated yet—you've just gotten fast. The next step is handing the when to cron and hooks.