A scheduled job that fails sends an alert. A scheduled job that never runs sends nothing.

That second kind is the expensive one. The report is just missing on Monday morning, and someone has to notice.

Tool of the Week: n8n's durable scheduler

The short version: as of n8n 2.45 (released Oct 20), Schedule Trigger nodes store each run in the database before it is due. According to n8n's release notes, a restart no longer drops pending runs. It is on by default, with no setup.

Why it matters for real work. Scheduled reports, file pickups, nightly syncs and recurring checks are the jobs people quietly depend on. If the host restarts at the wrong minute, the old in-memory scheduler had nothing to fire from. A stored run does.

What else changed:

  • In multi-main setups, one main claims each run instead of a single leader firing all of them.

  • A run missed during downtime still fires if it is inside its grace period. After that, the trigger's misfire policy decides. The notes do not describe a full catch-up.

  • "Every N seconds" and "every N minutes" schedules now run N apart from the publish time, not on clock boundaries. Hourly, daily, weekly, monthly and cron schedules behave as before in everyday use.

Check before you upgrade:

  • If you relied on clock-aligned short intervals, N8N_SCHEDULER_TRIGGER_NODE_MODE=legacy keeps the old timing.

  • N8N_SCHEDULER_ENABLED=false returns an instance to the in-memory scheduler.

  • Both variables are deprecated and log a startup warning. A future version will remove them and the in-memory scheduler.

  • Poll triggers and system tasks stay on their current schedulers, so not every automation is covered.

My take: this fixes one failure mode. It does not tell you a job ran. Whatever tool runs the schedule, add one check that confirms yesterday's output exists: a file, a row, a timestamp. The scheduler is the easy part. Knowing it ran is the job.

Source: n8n release notes. This is a read of the notes, not a test report, so confirm behavior on the version you actually run.

Quick Hits

  • Claude Fable 5.2 and Mythos 5.2 (Nov 9). Anthropic says they are the same underlying model. Fable has stronger safeguards for cyber and biology misuse and is available to everyone, with the API model ID claude-fable-5-2. Mythos is restricted to vetted organizations. Pricing is $8 per million input tokens and $40 per million output tokens. Two things worth knowing: Anthropic itself recommends starting with Opus 5.5 for most everyday coding and knowledge work, and queries Fable's classifiers block get answered by an older Opus model instead, so an agent's behavior can change mid-run. Benchmark numbers are Anthropic's own.

  • n8n 2.46 (Oct 27): personal credentials in team projects. You can use your own credentials in a team project without sharing them. Teammates can see the workflow uses your credential but cannot run or publish it with that credential. Execute and Publish are disabled for anyone who cannot use it, and already-published workflows keep running. Useful to know before you hand a workflow to someone else: the credential has an owner, and that owner is a person.

  • Agent Skills moved to the Agentic AI Foundation (Oct 22). Anthropic donated the open Agent Skills standard to the foundation, which sits under the Linux Foundation. It says more than 70 agent platforms have adopted it. The security guidance for writing and installing skills is on the roadmap but not delivered yet, so treat a skill you did not write the way you would treat any code you did not write.

  • Anthropic started its IPO roadshow (Nov 10). It filed its S-1 on Oct 26. Nothing changes in your workflow today. It is a reminder to keep the model name in config and not hard-coded across your automations.

Prompt of the Week

Paste your list of scheduled automations into this. It forces the question that most schedules never answer: how would you know it did not run?

I will paste a list of my scheduled automations. For each one I give: name, schedule, what it produces, where the output goes, and what runs it.

For each automation, answer:
1. If the host is down when it is due, is the run skipped, delayed, or lost?
2. How would anyone find out it did not run?
3. What is the cheapest check that proves it ran (a file, a row, a timestamp)?

Return a table sorted by risk, then list the three checks to build first.
Do not assume how any tool handles missed runs. If you are not sure, mark it UNKNOWN and tell me which setting or doc to check.

Here is the list:
[paste]

One last thing

Writing this is one side of the work. The other is building it: data pipelines, file and reporting workflows, integrations between systems that were never meant to talk, and the automation that takes the manual steps out.

If a process in your week still depends on someone remembering to run it, book a free 15-minute audit and I will find your top 3 time-wasters. If it is not worth automating, I will tell you that.