Every Friday I used to spend 90 minutes compiling a weekly report: checking dashboards, writing summaries, formatting, sending. Now it takes me about 5 minutes to review and approve what the agent has already drafted. Here is the full setup.
The problem
Weekly reports involved:
- Pulling data from three different sources
- Summarizing activity from the past 7 days
- Writing a narrative that made sense of the numbers
- Formatting it consistently
- Sending it to the right people
I hated all of this except the “making sense of numbers” part, which is ironic because that was the part AI turned out to be best at helping with.
The OpenClaw workflow
I set up a cron job that runs every Friday at 8 AM:
- Agent reads from my data sources (CSV exports and a couple of web dashboards)
- Computes week-over-week changes for key metrics
- Identifies anything unusual (spikes, drops, anomalies)
- Drafts a summary in the house style I trained it on
- Drops the draft into a Notion page and pings me
I review it, sometimes tweak a sentence or two, then hit send.
What made it work
The most important part was training the agent on my “house style” by giving it 10 examples of past reports I liked. It picked up the tone, the level of detail, and which metrics I care about without me having to specify everything every time.
What still needs a human
- Interpreting genuinely unexpected results (why did signups drop 30%?)
- Any sensitive context (business decisions, team issues)
- The final “does this sound like me?” sanity check
The agent is great at synthesis and formatting. I am still the one who understands the full context.
Time saved
Roughly 80 minutes per week, or about 65 hours per year. Worth the 3-hour setup.
Anyone else automating recurring reports? What is your stack?
Curated by Selendia AI ⚡
