Practice 02: Right-Size the Task
Where most AI initiatives actually die
Practice two is where this course has watched the most well-intentioned AI initiatives stall, and it has nothing to do with technology. (That's this course's own observed pattern, not a cited industry statistic; nobody has an agreed number for it.) A good first task to delegate has three properties: the inputs are already in hand, the desired output is known in advance, and the result can be verified within minutes.
The mission vs. the bounded job
"Build us an AI system to manage all our customer data."
Vague, unverifiable, touches every system in the company. There's no defined output and no way to check progress. It stalls in week two, and the conclusion drawn is usually "AI doesn't work here," when the actual problem was scope, not capability.
This is delegation 101, not an AI-specific insight: a new hire doesn't get handed the company strategy on day one. They get handed a task they can finish, and that can be checked.
Split it: never simplify it
When a job really is too big, there's a rule worth adopting directly: split it, never simplify it. It's the same discipline covered later in this course's look at how the most advanced teams industrialize these five practices at scale (Lesson 10); here, it applies just as well to one person's desk.
The trap is quietly shipping a watered-down "v1 for now" of the whole vague thing: something that gestures at everything and finishes nothing, still unverifiable, still stalled. That's true whether the one doing the shrinking is an agent or a person.
The discipline is to cut the oversized job into two bounded jobs instead, and finish each one properly. Each piece keeps all three properties: inputs, defined output, quick to verify. Each one ships and gets checked, and each one earns the trust that buys the next, bigger task. A shipped small thing beats a stalled big thing every time.
The anatomy of a brief that works
Five ingredients turn a wish into a brief an agent can actually execute, and that can actually be checked. This is the template behind every worked demo in this course.
Which files, where. "The two exports in the data folder," not "our sales data" in the abstract.
The exact artifact wanted back. Not "analyze this," but "one Excel file, three named tabs: Matched, Exceptions, Summary."
The constraints. "Amounts in EUR. Don't touch the source files."
How the agent proves it worked: control totals, a citation per row. This is practice 03, arriving early. It gets its own lesson next.
Where the agent must pause. "Show me the plan first" (practice 01) or "leave it drafted, don't send" (practice 05).
Notice: not one word of these five is technical. It's a work order, the same one written for a capable new hire on their first real task.
Checking the discipline
Practice 03, with two full worked demos: a spreadsheet reconciliation and a contract folder becoming a renewals tracker.
Reply here and it goes straight to Rod. Same as replying to one of his emails.
