What this rung really is
Level 5 is where the agent stops being something you start and becomes something that's already running. Up to now you opened it, gave it a job, and watched. Here you set a rule that auto-saves its work, and you put it somewhere always on — so it can run on a schedule, while you sleep, and leave you a finished briefing instead of a job to do.
Two pieces make that safe. An auto-save rule means every unattended run leaves a trail you can review or roll back, even one that ran at 3am. An always-on home means the work happens before morning, not when you finally open your laptop. Clear this and your day starts with the work already done — not waiting to be started.
One more thing makes always-on trustworthy: it has to tell you when it breaks. An unattended agent that silently loses its connection at 2am isn't always-on — it's just off, quietly. So the last piece is a watch that pings you the moment something drops, and enough capacity headroom that a long run doesn't die halfway from a hidden limit.
Set it running — then walk away
The move at this rung is always-on: wire an auto-save rule so nothing is ever lost, then give the agent an always-on home so it runs on a schedule — overnight, while you sleep — and leaves you a finished briefing.
This rung sits on everything you wired in rungs 1–4. You add always-on hosting, auto-save & a watch on top — you don't start over.
Do this rung, for real
Everything you need is here — no tabs to chase. First, the jargon this rung throws at you, in plain words. Then the steps, with the exact things to paste or say.
Every morning you used to spend the first hour pulling rankings, GSC movement, and competitor changes for your top five clients into a status note. At L5 you wire that as a scheduled run on an always-on machine. It fires at 5am, pulls the data, drafts the note, and saves itself as it goes. You wake up, open one page, and the briefing is already there — caught up before your coffee. That's L5: the work moved from your morning to the machine's night.
An auto-save rule (a hook)
A rule that fires automatically at set moments — for example, saving the agent's work every time it finishes a step — without anyone there to ask for it.
Why it matters here — Once the agent runs while you're asleep, “save the work” can't depend on you. An auto-save rule means every unattended run leaves a trail you can review or roll back — the safety net that makes walking away reasonable.
An always-on home (hosting)
Running the agent on a machine that's always on instead of only your laptop, so it can do its job on a clock whether your computer is open or not.
Why it matters here — A morning briefing only works if something produced it before morning. An always-on home is the difference between “a thing I run” and “a thing that runs.”
A watch (monitoring)
A small always-running check that pings you the instant an unattended agent loses a connection or a scheduled run fails — so you find out at 2am from a message, not at noon from a missing report.
Why it matters here — An always-on system you can't tell is broken is worse than no system, because you trust it. The watch is what makes “I let it run overnight” safe: silence stops meaning success and starts meaning you get told.
Capacity headroom (a gateway)
Running across enough model capacity that a long overnight job doesn't hit a hidden limit and stop halfway. The clean way to do it is the budgeted API-key path, with sensible caps and fallbacks.
Why it matters here — The most common way an unattended run dies is quietly running out of quota at 3am. One caution: pooling personal subscription logins behind a proxy is against Anthropic's terms — so use the budgeted API-key path, not a shared-login workaround.
Wire the auto-save rule
Set the agent to save a snapshot automatically every time it finishes a step. Now an unattended run can't end with lost work — there's a labelled restore point at every stage, even one that ran at 3am. You can ask the agent to set this up for you.
“Set up a rule that commits the work automatically every time you finish a step, so an overnight run always leaves a trail I can roll back.”
Give it an always-on home
Move the run off your laptop onto a machine that stays on — a small hosted server. It doesn't need to be big; it needs to be always reachable, so the job can fire whether or not your computer is open.
Put it on a schedule
Tell it what to do and when — “run this every morning at 5” — as a single unattended command, no conversation. Pick one job you already do by hand each morning and let the machine produce it overnight. The briefing is waiting when you wake up.
“Every morning at 5, pull rankings and GSC movement for my top five clients, draft the status note, and save it.”
Put a watch on it
Tell the agent to ping you the moment a connection drops or a run fails, and to save a snapshot it can resume from if the machine restarts. Now “it runs while you sleep” includes “and it wakes you only if something actually broke.” A silent connection drop at 2am is the difference between always-on and just-off — the watch is what closes that gap.
“If any connection drops or a scheduled run fails, message me right away and save a restore point — otherwise let it run.”
How you know it's working
Before L5: the agent only works when you do. You open it, kick off the job, and wait at the keyboard — and the morning report only exists if you spent the morning making it.
- An unattended run produced something real on its own — a briefing, a monitored check, a scheduled job — with no one at the keyboard.
- Every run auto-saves as it goes, so even a 3am job leaves a trail you can review or roll back.
- The agent lives somewhere always on, reachable at a stable address, not trapped on your laptop.
- Your day starts with the work already done, not a task list waiting to be started.
- When a connection drops or a scheduled run fails, you get a message — you never discover it hours later.
- A crash mid-run doesn't cost the work — the agent resumes from its last saved state.
Make it stick. Take the one thing you do first every morning — the rankings pull, the inbox triage, the status note — and move it to the night. Run it unattended once; when the briefing is waiting the next morning, it's a habit, not an experiment.
Our skills for this rung
Linked items are founding-circle skills — clone the repo and run ./install.sh from the skills folder. Unlinked items are practices you build by doing.
An always-on artifact (morning briefing / monitoring hook / hosted agent). Show the live URL.
Clear this and you've genuinely cleared the rung — not read about it. Keep the proof; it's how you place yourself on the ladder.