A Complete Example
A Complete Example: From Manual to System
Let us walk through a real example end to end, so you can see the full process.
The scenario: Weekly client status report
Sarah is an account manager at a professional services firm. Every Friday, she prepares a status report for her top five clients. Here is her current manual process:
- Check email for each client's latest communications.
- Check the project management tool for task status and deadlines.
- Check Slack for any team discussions about each client.
- Review last week's report to see what was flagged.
- For each client, write a 3–5 sentence summary of what happened this week.
- Flag any risks, delays, or decisions needed.
- Format everything into a standard template.
- Send to her director for review before the Monday leadership meeting.
This takes Sarah 2–3 hours every Friday.
Step 1: Describe the workflow
Sarah tells the agent:
"Every Friday I prepare a client status report for my five key clients. I check email, our project management tool, and Slack for updates on each client. I compare to last week's report to spot changes. For each client I write a short summary of what happened, flag risks or delays, and note any decisions that need to be made. I format it into a standard template and send it to my director."
Step 2: Identify constants and variables
| Constants | Variables |
|---|---|
| The five clients she tracks | This week's emails, messages, task updates |
| The template format | What happened for each client this week |
| Quality criteria: what counts as a risk | New risks and decisions |
| The output destination (director) | Changes from last week |
Step 3: Build one step first
Sarah starts with the hardest, most time-consuming step: gathering and summarizing updates across sources.
She asks the agent:
"Here are this week's email threads for Client A [pasted]. Here are the project management updates [pasted]. Summarize what happened this week in 3–5 sentences. Flag any risks, delays, or decisions needed. Compare to last week's summary [pasted] and note what changed."
Step 4: Test with real data
The agent produces a summary. Sarah checks:
- Did it catch the delayed deliverable mentioned in the Tuesday email? ✓
- Did it flag the budget concern from the PM tool? ✓
- Did it miss the client's tone shift in Thursday's email? ✗ — It summarized the content but missed the client's frustration.
Step 5: Refine
Sarah adds to her instructions:
"When reviewing client emails, pay attention to tone and sentiment, not just content. If a client sounds frustrated, concerned, or unusually formal, flag this as a relationship risk even if the content seems routine."
She re-runs the step. Now it catches the tone shift.
Step 6: Compose
Sarah chains the steps:
- Gather: Pull this week's emails, PM updates, and Slack messages for each client.
- Compare: Check against last week's report for changes and trends.
- Summarize: Write a 3–5 sentence summary per client with risks flagged.
- Format: Place summaries into the standard template.
Step 7: Name and reuse
She calls it "Friday Client Brief."
Every Friday, she provides the agent with the week's raw inputs. The agent produces a draft report. Sarah reviews it in 20 minutes — checking for accuracy, adding context only she knows (like a private conversation with a client), and adjusting tone.
Result: 2–3 hours compressed to 20–30 minutes. And the quality is more consistent because the agent never forgets to check all sources or compare to last week.
The compounding effect
After running this for a month, Sarah notices patterns the agent surfaces: Client B has had a risk flagged three weeks in a row. She would not have noticed the trend in her manual process because she started fresh each week. The system creates institutional memory she did not have before.
This is knowledge compounding — the system gets more valuable the longer you use it.