The Monday Morning AI Workflow I Built to Avoid Forgetting Tasks
- Kathleen Spangler
- May 8
- 4 min read
Updated: May 13
The first hour of every Monday used to look the same. Outlook calendar in one tab, Jira in another, and my eyes flickering between them trying to confirm three things at once: which beats had moved over the weekend without anyone telling me, which new beats on the calendar didn't have sprints built yet in Jira, and what tasks each of those new sprints would need.
It was the worst hour of my week, mostly because I was running the whole thing from memory, from the part of my brain that already had six other things on it, and I'd spend the entire hour comparing dates while everything else got staler.
The "Problem"
Open Outlook calendar and look at the next four weeks of beats. Open Jira and pull up each active sprint. For every beat on the calendar, find the matching sprint in Jira and confirm the story due date in the sprint still matched the date on the calendar.
The reason this had to be manual is that marketing calendar updates don't generate Jira notifications. If a release date moved over the weekend because production shifted or partner availability changed, nobody was going to ping me, so I had to find it myself.
For any beat on the calendar that didn't have a sprint yet, I'd build the sprint from scratch. That meant creating each individual task and filling in a description, a story point estimate, a due date, an assignee, and dependency links to other tasks.
The number and shape of tasks varies by release size, but a typical Release Date Announcement sprint runs around 40 tasks across categories like dev builds, screenshots, trailer, copy, press kit, website, social, community, email, and day-of activation. The dependency chain matters because dev build comes first, then screenshots, then press kit assembly, then press release distribution, and if you get one date wrong the whole chain slips.
Building all of that by hand from memory took me three hours per sprint, sometimes longer.
Why I Built It
I needed three specific things on a Monday morning, and only three:
Drift between calendar and Jira dates, so I can fix the dates
New beats without sprints yet, so I can build them
The full task list for any new sprint, so I can spin it up in Jira fast
Velocity, workload, forecasts, and historical analysis are all useful, but they're not what I'm trying to do at 9:00 AM on a Monday when I'm just trying to start the week without surprises.
How I had it Made
The first version of this AI-assisted Monday sweep was overloaded. I asked the AI to cross-reference the calendar against the active sprints, and what came back was technically correct but not useful to me. The output included nine sections (velocity charts, story-point burndown, workload-by-assignee tables, four-sprint trend forecasts, historical drift analysis) and ran over three hundred lines.
I told the AI to cut it down. Three things on a Monday: drift, missing sprints, and the full task list for any sprint I need to build. Strip everything else. The version I got after that is what I use today.
The other thing I had to push back on was assignee inference. When the AI builds a new sprint from scratch it has to guess who should own each task, and early versions either left assignees blank or assigned them at random. The fix was telling it to look at the last three sprints of the same type (Release Date Announcement, Beta Launch, Patch Drop) and infer assignees from the pattern, like copy and PR to the writer, video and screenshots to the video lead, press kit to the PR coordinator, community posts to community, social to social, email to the lifecycle owner, and final review to the producer.
Now when the AI proposes a new sprint, the assignees are already there, and I look it over, fix the one or two it got wrong, and approve.
The Solution
Monday mornings now take about 10 minutes instead of an hour. I paste the calendar into a chat with the AI, ask for the sweep, and what comes back is two short sections plus, if there's a missing sprint, a proposed task list with assignees, due dates, and dependency notes.
If I approve, the AI gives me everything I need to spin up the sprint in Jira in another few minutes. Once it's live I paste the sprint ID back and the AI links the dependency chain in one pass, typically 14 links for a Release Date Announcement sprint, which used to be a 30-minute click-fest.
The bigger thing is what I'm doing with the time I got back. Monday morning is now when I check in with each team member before the day kicks off, instead of when I do data entry.
What's Next
Calendar input is still manual. The AI can't read Outlook directly, so I have to paste a screenshot of the calendar into the chat. I have an IT request in to sync Outlook to Confluence so the script that runs my marketing beats dashboard (the subject of Post 3) can read it directly, and when that lands the manual paste step goes away too.
The assignee inference is good, not perfect. For sprint types we run regularly the pattern holds, but for new sprint types, like when we launch a new product line or take on a beat we haven't done before, the AI guesses and I correct. It's a once-per-novel-sprint cost.
Date math on dependency-heavy chains still surprises me sometimes. The AI is good at the obvious cases like "press kit assembly comes before distribution," but it occasionally proposes a date that doesn't account for a dev-side blocker I haven't told it about, and I catch those in review.
Comments