Skip to content
FlintFlintby Torch
← Level 4Builder · L4

Day 14: is it still running?

Build-along · ~12 min at your own keyboard

Two weeks is long enough for the first silent failure and short enough to fix it. Come back on day 14 and count. How many runs should have happened? How many produced good output? How many alerts fired, and were they right? How many minutes did you spend fixing the output by hand? Then compare the minutes you spent against the minutes the job saved. Decide one of three things. Keep it as it is. Fix one thing. Kill it. Killing a job that does not pay back is a good result. It frees you to pick a better job.

Ladder · Level 4The long version: Automation ↗

Do this

0/7

Checkpoints

  • ▸You should now see counts for expected runs, good runs and alerts, taken from the run history.
  • ▸You should now see minutes of fix-up next to minutes saved.
  • ▸You should now see a decision of keep, fix or kill, with one reason.
The three ways this goes wrong
  • You cannot tell how many runs happened. — Use the run history or your heartbeat alerts. If your tool keeps no history, add a line to a log sheet at the end of every run so the next count is easy.
  • Every run says 'completed' but you rarely used the output. — That job is busy, not working. Count those runs as not good. Name the one cause: bad input, weak prompt, or a job you do not need.
  • The job stopped at day 6 and you did not know. — Your heartbeat is missing or wrong. Fix the heartbeat first, using the last lesson. Then restart the job and check again in 14 days.

Dated facts in this lesson: ChatGPT has Scheduled Tasks (chatgpt.com/schedules). Workspace agents are on Business and Enterprise plans (researched Oct 4, 2026, not yet confirmed by a person: check the source before you rely on it); Claude scheduled tasks live in Cowork, under 'Scheduled' in the sidebar. They run in the cloud with your laptop off, on Pro and above (researched Oct 4, 2026, not yet confirmed by a person: check the source before you rely on it); Gemini Spark (beta) has Schedules: time-based, Gmail monitors and topic monitors. It is on AI Pro and Ultra, for personal accounts only, with up to 50 schedules. An older 'Scheduled actions' feature allows 10 active (researched Oct 4, 2026, not yet confirmed by a person: check the source before you rely on it); Microsoft 365 Copilot automates through Copilot Cowork and Power Automate. A scheduler in consumer Copilot is not confirmed (researched Oct 4, 2026, not yet confirmed by a person: check the source before you rely on it); Grok has Automations (daily, weekly and email triggers) on a paid tier (researched Oct 4, 2026, not yet confirmed by a person: check the source before you rely on it); Perplexity has Scheduled Tasks that run at most once an hour (researched Oct 4, 2026, not yet confirmed by a person: check the source before you rely on it).

Share what happened

Paste your day-14 check in about six lines: expected runs, good runs, alerts fired (and whether each was right), minutes of fix-up versus minutes saved, your decision (keep, fix or kill), and the one reason.

Strip API keys, passwords, and tokens before you paste. Swap them for [KEY].

🔒 Don't paste confidential, personal-health, or payment info.