Trail map
Detailed content
🎯 Metric with an owner
A number tied to the pain point, today's value recorded before turning it on, a target, a deadline, and one person accountable for it. At the end, the matrix of two questions.
A number you count the same way every time that shows whether the pain point has eased — such as "patients who showed up."
“Messages sent” goes up even if nothing improves. The wrong metric can make a useless agent look successful.
Metric, pain point, effort × outcome, count them the same way.
The metric’s starting point, counted at the source (schedule, spreadsheet, or a week of notes) and written down with a date.
Without the previous number, you can’t prove anything later—not that it improved or that it got worse.
Today's value, source, date; count instead of estimate.
Where you want to get and by when: Renata, from 29 to 34 gifts in 60 days; Marcos, from a morning to an hour in 30 days.
Without a goal and deadline, the agent stays “in testing” forever, and no one decides whether to keep it.
Goal, deadline, midpoint and final check, honest goal.
Who reviews the agent's errors, how often, and in what situation a human takes over the conversation.
An agent can make up a price with complete confidence. If everyone checks, no one does.
Owner, scheduled review, handoff to a person, record.
At the end of the period: did it work? (did the number change) and was it under control? (you can show what the agent did and who approved it).
A result without control is luck; control without a result is a cost. Only together do they justify keeping the agent.
Worked, under control, luck, cost.
A table with one row per week—metric, errors, handoffs, approvals, owner's rating—generated with an AI chat.
Eight rows later, both questions have already been answered, and every error has become a task instead of a surprise.
Weekly dashboard, set day, errors become tasks, "fill in".
📋 The spreadsheet in 4 blocks
Everything you decided in the course on one page, one agent at a time: pain point, control, security, and results. Includes a five-question audit and a review every 30 days.
The four blocks in a diagram and a printable table, with Renata’s example alongside and a blank column for your agent.
Each learning path left you with a decision. Together, on one page, they answer “who’s responsible for this agent?”
Agent worksheet, four sections, print version.
Desired result, what would break at 10×, and today’s proof — a counted number.
An agent without a clear pain point becomes an expensive toy with access to your accounts. Empty block = agent turned off.
Result × request, the 10× question, today’s proof.
Read, change, and send, approval, logs, and a shutoff button — written so a stranger can understand.
“Access to the calendar” is vague. Control also means saying what the agent is not allowed to do.
Five keys, exact location of the record, stranger test.
Yes or no for hidden instructions, secrets, extensions, spending caps, and customer data—with where to find the evidence.
Every “no” is an open door. It becomes a task with a due date before the agent gets more access.
Closed doors, “yes” with proof, task with a date.
Module 5.1 in block 4 and five audit questions: money, customer data, records, who can shut it off, and the error at 3 a.m.
A business partner, a major customer, or an insurer will ask. The completed spreadsheet answers each question in under a minute.
Section 4, audit, a response written before a mistake happens.
A complete prompt that has the chat interview you, fill in the four sections, run the audit, and schedule the next review.
Filling it out alone is a blocker. With the interview, you’ll leave with the spreadsheet ready — without giving the chat any secrets.
Interview, “VERIFY,” review every 30 days.