What the training covers in detail

Every line is something you do yourself at the keyboard, on your own material. What we pick follows the work in front of you, rather than the order of this list.

Part A: What Claude can do

The basics, on your own computer

1. Chat versus Claude Code. A chat window explains how to tidy a folder; Claude Code opens the folder and tidies it.
2. Speaking instead of typing. One spoken sentence in the morning is enough, and the result is ready when your working day starts.
3. The interface. The question it asks before moving anything, the display that shows how full its memory is, and the mode where it plans first and acts after.
4. Files and folders. Mixed downloads become named folders, and a stack of PDF reports becomes one table of key figures and dates.
5. Office documents. Notes become slides for the meeting, and a folder full of receipts becomes a table with totals, the formulas included.
6. Web research. Competitors’ price lists land in one table, and every number carries the link it came from.
7. Mixed formats through one entrance. PDFs, presentations, spreadsheets and mail text are made readable first, because a converter settles a question of format better than a language model does.

Making it yours

1. Connectors. A date from a mail lands in your calendar and a call summary in your notes, while both apps stay closed.
2. A memory for the project folder. Units, date format, language, tone and the rule that every number carries its date and source are set once, and from then on they apply to every analysis in that folder.
3. Skills. A workflow that went well once becomes a named tool, and from then on one sentence runs every step.
4. Tools from outside. What others have built is set up in a minute, together with the question of how you can tell whether you still understand it.

Runs on its own and stays in place

1. Several helpers at once. Three checks run side by side and hand back one comparison table.
2. Scheduled runs. The overview is ready early on Monday, the list of open items on Friday, and your part is approving rather than assembling.
3. Chains with checkpoints. A workflow runs as separate steps rather than one big pass, a person approves in between, and a log records where every result came from.
4. Linked knowledge. Contacts, conversations and topics connect, and a question runs across all your notes rather than through a single folder.
5. Ready for handover. A workflow is written down so that a colleague can take it over on Monday.

Part B: Understanding and judgment

How it works, and where it turns unreliable

1. What generative AI is. A spam filter, a chatbot and an image generator are three different things, and only one of them is meant here.
2. How text comes about. You start a sentence and the model finishes it; technically, that is all that happens.
3. Same question, different answer. The same request sent twice gives two different texts, and that difference is part of how it works.
4. Capabilities and limits. A forty-page contract gets summarised well, while today’s exchange rate is out of its reach.
5. Three kinds of error. Invented, misread and left out each call for a different remedy, and the omission is the riskiest because it stays silent.
6. Numbers from text and tables. A bar chart is the least reliable source in any report.
7. Context and memory. After a long stretch of work, Claude contradicts what was settled at the start, and the reason takes two minutes to understand.
8. Probable versus fixed. Sums, conversions and filing belong to fixed rules; the AI takes on the fuzzy work: reading, sorting, wording, assessing.
9. The realistic bar. A workflow has to be measurably better than the handwork it replaces, and that bar is reachable, where perfection would keep every task manual.

Getting what you want

1. Clear instructions. “Make it shorter” leads somewhere different from “cut to 200 words, keep the three key figures, drop the introduction”.
2. Reference material. Put your last offer next to the request, and the new one sounds like your company.
3. Output format. As a table, four columns, one row per supplier, saved as a file in a named folder.
4. Roles. The same question answered from the point of view of your tax advisor changes what comes back and what gets flagged as risky.
5. Examples. One offer you are happy with is enough, and the next twenty follow its shape.
6. Breaking down big tasks. First list every requirement, then stop, then answer the first five.
7. Steering instead of starting over. In round three it is “shorter, less marketing tone, keep the numbers”.
8. AI as a sparring partner. Ask it to find out everything it needs first, and it comes back with five questions worth answering before anything gets written.

Judgment, and staying in charge

1. When it gets stuck. One sentence almost always helps: “Explain step by step what just happened, and change nothing until I say so.”
2. What stays handwork. The rejection with a personal touch, decisions about people and the two-line mail stay with you.
3. Checking and owning it. A quoted number gets checked against its source in twenty seconds, and the result goes out with a name on it.
4. Confidentiality. What stays out, what goes in instead, and where your inputs actually end up.
5. Where permissions live. Your operating system, file storage, mailbox and calendar decide who sees what, and Claude works within exactly those permissions.
6. Bias. The same assessment comes out differently when a name changes, and that matters wherever people are involved.
7. Understanding before adopting. An installed tool works until the day it stops, and on that day someone should be able to say what it actually does.
8. Staying capable by hand. You redo one analysis by hand, to see whether the tool you built still calculates correctly.
9. Keeping up. When a new model is announced, a ten-minute test shows whether anything changes for your own work.

Everything you adopt is something you can explain.

Book your free first session