Anthropic
Claude
I lean on it for long documents and code that has to hold together.
Where it has an edge
- Long briefs, specs and reports
- Code that touches several files
- Writing that has to sound like a person
- Reading a screenshot or a PDF and telling me what matters
How I prompt it
I give it the job, who will read the result, and the one thing that must not happen. Then I paste a real example of the output I want. Claude follows structure well, so a short outline beats a long list of rules. When it gets something wrong, I point at the exact sentence instead of rewriting the whole prompt.
Images
Claude reads images but does not make them. I use it to look at a screenshot or a layout and tell me what is broken, not to produce artwork.
Where it slips
It can hedge, and it will write three paragraphs where one would do. Both are fixable with a tighter prompt, but you have to notice.