From Monday's One-Line Request to Friday's Handover

This course follows one sample project from start to finish.

Monday: an operations director sends you one line: "Should we bring a second supplier into the northern region before the contract renews?"

Friday: your recommendation reaches her. The test of the week is simple. She acts on it without calling you to ask what it means.

She has three realistic moves:

• Sign: renew with the current (incumbent) supplier alone.
• Wait: hold off on adding anyone until more is known.
• Split the volume: bring in a second supplier alongside the incumbent.

Your job is the page that lets her choose with confidence. You are not making the decision for her.

You'll learn to use Manus, an AI agent, on work like this. An AI agent is software that plans and carries out multi-step work on its own computer, rather than answering one question at a time. Manus can produce a polished first draft in minutes, so the draft is no longer what you deliver. What you deliver is the reason the director should believe it.

The stated request is the words someone sent you. The unstated assignment is everything those words leave out, and you have to work it out before anything gets produced.

The supplier question leaves out three things:

1. Who acts, and what they must decide. The director chooses between sign, wait and split.
2. Assumptions nobody has written down. For example, that a second supplier could meet the quality standard, that northern volumes are big enough to split, or that the renewal date can't move. Each one may be true. None has been checked.
3. What the page needs so she can act without calling you: the options, the evidence with sources, what is missing or disputed, and the specific decision you're asking her for.

If you skip this, you'll produce a polished answer before you know what the requester actually needs. Try writing it down for the sample project:

STATED REQUEST: [paste the line exactly as sent]
WHO ACTS: [name/role]  DECISION: [the choices they face]
UNWRITTEN ASSUMPTIONS: [list each; mark checked / unchecked]
THE PAGE MUST SHOW: [options, evidence, gaps, the decision asked for]

A request that sits still is easy to answer. A real week keeps moving, and this one moves on every working day:

• Tuesday: the incumbent revises its lead times, and the new figures clash with the quality team's data.
• Wednesday: the director states a preference that cuts across the agreed scope.
• Thursday: the qualification test for the second supplier slips, and an action from yesterday's call has no owner.
• Friday: the recommendation still has to stand up.

Every professional gets some version of this, whether it comes from a client, finance or a sponsor. The course comes back to these events. You'll see how to bring conflicting evidence to the surface instead of quietly blending it in. You'll keep a preference from silently widening the scope. And you'll mark an action as owner "TBD" instead of guessing a name and presenting the guess as agreed.

The deliverable changes during the week, but the way you handle it stays the same.

The whole course rests on one split of the work:

• You own the brief. A brief is a written statement of who the work is for, what decision it supports, the deliverables, what's excluded, and the acceptance criteria (the tests the result must pass). You settle exactly what is being asked before anything is produced.
• Manus works the middle. That means research, sorting sources, first-pass analysis and drafting, all done in phases you can review.
• You own the handover. The handover is the moment you release the work to someone who will rely on it. You decide exactly what is ready before anything leaves.

Reid Hoffman, co-founder of LinkedIn and a partner at Greylock, describes a similar split in his investing. He finds AI very useful for due diligence and competitive analysis, but he doesn't hand it the decision about whether to invest. In the sample project, Manus can gather and compare evidence on suppliers. The recommendation and the release stay with you, and the decision stays with the director.

There are two easy ways to fail when AI enters this kind of work.

Sitting it out. You keep doing everything by hand and decline to learn the tools. The work changes anyway, and someone else redesigns your job, usually without your judgment in it.

Forwarding the draft. You let the agent produce a fluent first pass and send it on as if it were finished. It reads well, so it goes out. Then a colleague, or the director, finds the wrong lead-time figure, the guessed owner or the missing source, and they end up doing your cleanup.

The path between them is a habit good professionals already have: own the brief and own the handover, and let the agent work the middle. That way you can take on bigger work than you could alone, and you stay the one who vouches for it.

Cassie Kozyrkov, a decision-intelligence adviser and formerly Google's chief decision scientist, warns that being able to do a task doesn't mean being able to do the job. She calls confusing the two "a category error". A Manus report with neat headings and a sources section has completed a task. The job is done only when every file is present and correct, every claim holds up, and the director can use it.

Here's a real case of the gap. In 2025, Australia's employment department replaced a review it had commissioned from a large professional-services firm, after the firm confirmed that some footnotes and references were wrong. Generative AI had been used, a corrected report was published, and officials told a Senate hearing that part of the fee was refunded. The course treats this as a checking failure, not a fraud story. The rule it draws is older than software: nothing leaves under your name that you haven't traced to its source.

A note on voices: none of the people quoted in this course is talking about Manus or endorsing it.

You can follow the sample supplier project throughout. Or you can swap in a comparable assignment from your own role, such as a vendor decision, a service change or a client recommendation, as long as two conditions hold.

1. You have permission to use it with an AI tool. Client confidentiality, NDAs and some clients' consent rules apply. Keep client-confidential records and employee information out of Manus unless your organization has approved it for that use.
2. Scope, commitments and review responsibility stay clear. You know what's in and out of scope, nothing the agent produces becomes a commitment without approval, and you know who reviews the result before it's used.

If either condition is uncertain, use the sample. You'll learn the same method without the risk. Next lesson covers what Manus actually is today, so you can judge whether it fits.