Claude AI at Work
by Isaac Otu
Chapter 1
Most people who use Claude treat it like a search engine that talks back. They open a chat window, type a question, get an answer, and close the tab. That is a real use of the tool, and it is worth maybe a tenth of what the tool can actually do for a working professional.
I run a small research and training organisation in the corrosion engineering space, called the Centre for Corrosion Studies. I am not a software developer. I have a materials engineering background and a day job that has nothing to do with writing code. Over the past several months, working almost entirely through Claude, I have built a research and content pipeline that drafts technical papers and training decks to a fixed house style. I have built a scientific computing tool that fits real corrosion data to published models and cites its sources. I have a fleet of specialist assistants that each know their own job and hand work off between each other, publishing to a small collection of live websites. None of that came from learning to program. It came from learning to work with the tool properly, which turns out to be a different skill from knowing how to ask it a good question.
This book is about that second skill. Not prompting tricks, and not a tour of features. The actual difference between someone who gets a paragraph out of Claude and someone who gets a working system.
If your only experience of Claude is a single chat window with no memory of yesterday's conversation, you are using roughly the same slice of it that a first-time visitor uses. That slice is genuinely useful. It also throws away almost everything that makes the tool worth building a habit around.
A single chat forgets your standing rules the moment you close it. It has no access to your actual files unless you paste them in by hand, and pasting does not scale past a page or two. It cannot check its own work against a real file on disk, run a script, or verify a citation against a live source. It cannot come back tomorrow and pick up exactly where you left off. It has no idea there was a yesterday.
None of that is a flaw in the tool. It is a description of what a bare chat window is for: a quick question, answered once, forgotten immediately. The tool is capable of far more than that, but only if you set it up to use the rest of what it has.
The gap between casual use and real use is not a matter of typing better prompts. It is a matter of giving the tool three things a bare chat window does not have: memory, tools, and standing rules.
Memory means the model can read your actual files, your actual project history, and your actual house style, rather than working from whatever you happen to retype each time. In my own work, this is the difference between asking for "a training slide about corrosion" and having the model already know that C2S never uses em dashes. It already knows the citation numbering convention, and it already knows the structure of the four existing training decks a new one should match. I do not repeat that context every time. It is simply there, because the system is set up so that it is there.
Tools mean the model can act on the world instead of only describing it. It can read a file, search a folder, run a calculation, fetch a real webpage, and check its own output against something concrete rather than against its own confidence. A model with no tools can only tell you what it thinks is true. A model with tools can go and check.
Standing rules mean instructions that persist across every conversation without you retyping them. A rule like "never invent a citation" or "never use an em dash" or "flag anything you could not verify" only earns its keep if it applies automatically, every time, rather than living in your memory as something you have to remember to say.