Scheduled tasks that earn their keep.

Scheduled tasks are one of LADLE's most-underused features. Once you have three or four running, you notice the compound value — a small daily rhythm of prompts the assistant answers without you having to remember to ask.

The eight that keep earning

**1. Daily standup.** Weekdays, 08:00. Prompt: "It's the start of a workday. What are the three things I said I'd do today (based on the last chat in the 'Weekly Planning' project)? Give me one sentence of context on each." Runs into a dedicated chat titled "Daily Standup." Skim it over coffee.

**2. Weekly plan.** Weekly, Monday 07:30. Prompt: "It's a new week. Based on the last three weekly-plan chats, what's the one project that most needs attention? Give me a two-day mini-plan for it." Best in a dedicated chat so the model can see its own prior recommendations.

**3. End-of-week debrief.** Weekly, Friday 16:30. Prompt: "It's end-of-week. Draft three questions I should ask myself about how the week went. Keep them concrete." Dedicated chat.

**4. Weekly writing prompt.** Weekly, Sunday 20:00. Prompt: "Give me one non-obvious writing prompt on the intersection of [your beat]. Two sentences max." New chat per run so each prompt is standalone.

**5. Daily learning nibble.** Weekdays, 12:30. Prompt: "Teach me one non-obvious thing about [your current learning topic]. Under 200 words. Explain why it's non-obvious." New chat per run.

**6. Weekly reading tail.** Weekly, Sunday 09:00. Prompt: "What have I mentioned wanting to read in the last week's chats that I haven't actually read yet? Just the list." Best in a dedicated chat.

**7. Monthly retrospective.** Weekly, first Sunday of the month, 10:00. Prompt: "Based on my last month of chats, what shifted? What stayed the same? What am I probably about to be wrong about?" Dedicated chat.

**8. Daily focus reset.** Weekdays, 14:00. Prompt: "It's mid-afternoon. What's the single most valuable thing I could do in the next 60 minutes given [your context]? Answer in three sentences." New chat per run.

What makes a task worth keeping

Three tests, in order:

**One:** does the reply actually make you think? If you glance at it and immediately close the tab, the prompt is too generic. Rewrite it more specifically.

**Two:** does the cadence match the natural rhythm of the thing? A monthly retro on a Sunday works because Sundays already carry that reflective quality. A daily standup at 3pm doesn't work — the day's already shaped by then.

**Three:** does the task compound? A scheduled task in a dedicated chat gets to see its own prior replies. That's the magic — the model can reference "you said last week that X" without you having to remember. Prompts that don't take advantage of this should probably be new-chat-per-run.

When a task has gone stale

Two signs:

- You're deleting the run without reading it. Prompt is generic; kill or rewrite. - The reply hasn't changed meaningfully in three weeks. The model has hit a local optimum. Add a specific ingredient to the prompt (a project name, a recent decision, a constraint) that forces it to compute something new.

The compound value

The reason to run several scheduled tasks — not just one — is that they collectively give you a background rhythm of prompted reflection. It's not that any single task is life-changing; it's that the aggregate creates a habit of stopping, reading a short piece of writing about your own work, and letting it prompt one small decision.

That habit is undervalued. Scheduled tasks are the cheapest way to buy it.

The failure mode

Running too many. If you have 30 scheduled tasks firing across the week, you'll stop reading them. Six is a good target. When you add a seventh, kill one that's earned its retirement.

What LADLE doesn't do (yet)

Scheduled tasks don't have access to tools right now — no web search, no code execution, no external calendar reads. The task is a one-shot Anthropic call with the standard system prompt. If you need "check my calendar and tell me what to prioritize," that's on the exploring roadmap; for now, do it manually.

Scheduled tasks also don't retry aggressively. First failure → back off one hour → retry. Second failure → the task marks errored and stops running until you re-enable it. This is by design (a broken task shouldn't spam retries), but it means you should occasionally glance at /scheduled to see if any errored quietly.

Setup takes five minutes

Go to /scheduled → New task → paste one of the prompts above → pick a cadence → pick a hour → pick your timezone → save. The task fires at the next matching moment. If you set daily 8am and it's already 10am, the first run is tomorrow morning.