What Summarize with AI does
Summarize with AI reads what a workflow collected and writes a short summary, following your instructions. It usually comes after a Board report or Look something up step, and before a Send an email step.
Here are a few examples:
- Every evening, summarize the day on the board and what your team wrote in their end-of-day recaps.
- Every Monday, look back over the week: what got done, what was missed, and who still has work left.
- Every Friday, review the week's fridge temperatures, like "Around 34° all week, all good," or which readings were warm and when.
Set it up
- Instructions: What to look for and how to organize it. For example: "Start with anything urgent that's still open. Then write a line for each person who sent a recap."
- What to read: Usually The whole report from a Board report step, or what a Look something up step found.
- Also read: Optional. A second thing to read along with it. For example, the board report in What to read and the day's recaps here.
Here's a typical daily workflow:
- Starts when: On a schedule, every day at 7 p.m. See Start a workflow.
- Board report, with Period set to Today.
- Look something up, with List set to
Daily recapsand From the last set to24h. - Summarize with AI, reading the board report and the recaps.
- Send an email to the owner, with the summary as the body and The whole summary attached as a PDF.
The Daily employee recap and Weekly manager digest recipes set this up for you.
What it writes
Every summary comes back in the same parts, and each one is a value later steps can use:
- The summary (
summary): Plain text for an email, with short lines and simple bullets, starting with what needs attention. - Headline (
headline): One line about how it went. - Needs attention (
needs_attention): One line for each thing to act on. Empty when everything's fine. - How many need attention (
needs_attention_count): Useful for a When way out, like sending a notification right away when it's more than 0. - Highlights (
highlights): What went well, or what's worth knowing. - By person (
by_person): A name and a sentence or two for each person mentioned. - The whole summary (
document): Everything, for Attach as a PDF in Send an email, or for a web request.
It doesn't make up facts, names, or numbers. Counts that were already calculated, like a board report's done, missed, and overdue counts, are used as is and never recounted. Records in Spanish are summarized in English unless the instructions say otherwise.
When it doesn't work
The step fails, and the run takes a When way out on data.<step id>.ok = false if you added one, when:
- There was nothing to read.
- Your account has used up its workflow AI allowance for the month.
- Skidle couldn't reach the AI service, or the AI service didn't return a summary.
Very long lists are cut off at about 120,000 characters, and the summary says so at the end.