Reframe: From Chatbot to Coworker

Good morning, and welcome to the difference between a chatbot and a coworker. That difference is what this course is really about.

For an entire career, delegation meant one thing: handing work to people. And people are scarce, expensive, busy, and inconveniently asleep at night. So everyone accumulates a pile of work that isn't quite worth delegating to a person: the two-hour reconciliation, the contract folder nobody's gotten to, the report that stays "good enough" because improving it would cost someone a full day. That pile (not big strategic decisions, but this backlog of small things that were never worth a person's time) is where a week quietly disappears.

The promise of this course

The skill on trial in this course is not prompting. Prompting is what you do with a chatbot: you ask, it answers, and you do the actual work yourself. What this course teaches is delegating: handing a bounded task to something that can read your files, do the work, and show you what it did.

Who's teaching it

Rod Rivera has spent fifteen years shipping machine learning inside large companies (Alibaba Cloud, Huawei, Samsung) and now teaches AI product engineering at Nebius Academy and as Professor of the Practice at ITAM. The gap he lives in professionally is the same one this course closes: between what AI looks like in a demo and what it takes to make it work on an ordinary Tuesday inside a real company.

The finding that reframes everything

Before any practice gets taught, one research finding is worth sitting with. Anthropic's internal research (a June 2026 study titled "Agentic coding and persistent returns to expertise," examining roughly 400,000 real Claude Code sessions) found that what predicts success with an AI agent is not coding skill. It is domain expertise. People who deeply understand the work they're delegating succeed at roughly double the rate of novices: experts close to 33% of sessions, novices closer to 15%. When the agent drifts off course, an expert notices and steers; a novice doesn't know enough yet to catch it.

The same study found something sharper still: users make about 70% of the planning decisions on a task, but only 20% of the execution decisions. Read that again: the human decides what, the machine decides how. Read as a job description, that split looks closer to a manager's than a programmer's. The study measured the decision share, not job titles, so that framing is this course's own reading of the number. Anyone who already manages a team, a budget, or a project already has the core skill this course is about. It just hasn't been pointed at software before.

Anthropic's own usage data backs this up from a different angle (a separate internal breakdown of tool usage, not the study above): only about 9% of Claude Code usage is software development. Roughly a third is ordinary business-process work (reports, reconciliations, checklists) and another 16% is content and drafting. A tool built for programmers is used mostly by everyone else, and that split is a second, independent signal that this isn't a coding class in disguise.

None of this is happening at the margins. Bitkom, the German digital industry association, reported 41% of German companies actively using AI in 2026, up from 17% just two years earlier, in its own annual industry survey. The question in most organizations has already moved past "should we try this?" to "who here actually knows how to brief it well?"

Your real limit was never your ideas

Before looking at the tool itself, it's worth being honest about what has always capped output. It wasn't a shortage of ideas.

Here's what changed. The thing on the other end of a delegation now comes in unlimited copies, costs cents per task, runs overnight, and never gets bored of a repetitive job. Put those properties next to that backlog and the logic flips completely: the work that was "not worth a person's time" becomes the first thing worth handing off.

Which moves the bottleneck itself. The scarce resource is no longer someone to do the work. It's someone who can brief it well and check it honestly. That skill already has a name: management.

What briefing it actually looks like, once

Here's one concrete pass at what "brief it well and check it honestly" actually looks like, before this course spends five lessons making it a habit: say the pile includes a monthly expense reconciliation, matching a card-provider export against the accounting system and flagging what doesn't line up. Briefed loosely ("clean up this month's expenses"), an agent might quietly drop a row it can't confidently match instead of flagging it, and hand back a tidy file that looks finished but silently buried a real discrepancy. Briefed with the check built in ("match every row, and list anything you can't confidently match instead of dropping it") the agent instead comes back with, say, an illustrative €340 of transactions it couldn't reconcile with confidence, named individually with the reason each one didn't match. That number isn't a measured result from a real company; it's a stand-in for the shape of what a fresh session will actually hand back: not "done," but a short, named list of exactly what it couldn't resolve on its own. Reading that list is the entire verification step, not re-doing the reconciliation, just checking the handful of rows the agent already flagged as uncertain. That's the discipline this course spends the next ten lessons making concrete, one practice at a time.

A chatbot answers. An agent works.

A chatbot is a one-question, one-answer loop: you ask, it answers, you copy the answer out and do the actual work yourself. It never touches your files.

An agent is different in kind, not just degree. Given a task, it reads the actual files in a folder you choose, makes a plan, works through the steps (for minutes or hours), checks its own output, and hands over finished work plus the evidence that it checked itself.

That reframe is the foundation for everything that follows in this course. Every practice ahead is really a management practice, applied to a new kind of hire: fast, tireless, and slightly overconfident.