Skip to content
Period 4 / 12

Why Claude Code, Not the Chat App

Anthropic sells the same underlying agent through two different doors, and which one gets used matters for a business, not just an individual. Picture a colleague who wants to try automating a task against a folder of real customer contracts: which door they open first decides whether that folder ever should have left their desktop. This lesson is a short detour before the five practices, because the choice of tool decides how much of what follows you can actually rely on.

The same engine, two doors

Same engine, two doors

Claude Cowork is the simplified app: point-and-click, a pleasant first taste for a personal project. It gives up three things a business needs, though. Less control over standing rules and workflows, since every job stays a one-off chat rather than something reusable. A thinner audit trail. And, as of mid-2026, no EU data-residency option, so inference runs through US infrastructure. This specific claim traces to the instructor's own speaker note from a July 2026 source deck, flagged there as accurate "as of July 7" and changing weekly, so treat it as a snapshot worth re-checking against current Anthropic documentation before relying on it, not as a settled, permanent fact.

Why Claude Code wins for a business

Claude Code is the professional tool, and it's what this course is built on, for three reasons. It carries the full toolkit (Plan Mode, a persistent instructions file, reusable skills), which turns out to be the exact mechanism behind the five practices ahead. It makes a workflow repeatable: in Claude Code a workflow becomes a file that can be versioned, shared, and re-run; in an app it stays a chat that happened once and is gone. And, as of mid-2026, it has compliance headroom the app doesn't: Claude Code can run against EU cloud deployments, such as AWS Frankfurt, when the data demands it. The same snapshot-not-permanent caveat above applies here too: check current deployment options before committing a workflow to this claim.

'Code' describes the tool, not a requirement

"Code" in the name describes what the tool can do, not what anyone needs to know to use it. It can be driven in plain English or German, and there is no programming involved in any lesson in this course.

The honest verdict

The honest verdict

Cowork is a good door for learning on personal, non-sensitive data. It is not, as of mid-2026, where customer or company records belong. Claude Code is the door this course walks through because the demos ahead use real business documents.

Quick check — A colleague wants to try automating a task against a folder of real customer contracts. Which door fits, as of mid-2026?

What this looks like in your function

The worked examples in this course happen to be finance-flavored (a spreadsheet reconciliation, a contract folder, a monthly report) because they're vivid and easy to verify. The underlying pattern is the same everywhere: files in, checked work out.

Finance & Controlling: reconciling exports, variance commentary, month-end checklists.

Legal & Procurement: clause extraction, renewals and risk trackers, obligation summaries. (This is exactly the shape of the contract-folder demo later in this course.)

One deliberate omission from that list: CV screening. The moment an agent's output touches a hiring decision, it sits in high-risk territory under the EU AI Act. That's worth knowing before reaching for this pattern there, and it's covered again in the compliance lesson later in this course.

The demo is a template, not a limit

Every one of those functions runs on the same five practices this course teaches. Reading the demos ahead, it's worth mentally swapping in whatever folder sits on your own desk. The demo is a template, not a limit.

Continue to Lesson 04

The five practices themselves, laid out as one map before each gets its own lesson and worked demo.

Have a question about this lesson?

Reply here and it goes straight to Rod. Same as replying to one of his emails.