Working With AI Agents

Stop chatting with AI. Start delegating to it. AI assistants have quietly become agents. They read your files, search the web, run calculations, work inside your other applications, and keep going at a task while you do something else. This book is about using that well: not in theory, but on your own Tuesday, with the eleven emails, the unread contract, and the spreadsheet nobody has cleaned up. Built around Claude, current as of the 2026 models Every example uses Claude by Anthropic, including the current model generation and what each one is actually good for. You will learn when the cheap model is the right answer, when it is not, and what everything costs. What you will be able to do - Write requests that produce usable results the first time, using a six-part pattern - Handle email, research, documents, spreadsheets, learning, and life admin with prompts you can copy - Set up projects, files, and memory so you stop explaining your situation every time - Connect your calendar, files, and tools safely, and understand what MCP is - Let work run without you, on a schedule, with sensible limits - Build five complete projects: a weekly routine, a research brief you can trust, an inbox that handles itself, your own agent in code, and a multi-agent workflow - Verify results properly, protect your data, and recognise prompt injection before it costs you For everyone, with a technical door at the back Parts one to three need no technical knowledge at all. Two chapters in part four show how to build your own agent with a few dozen lines of Python or TypeScript, and the book tells you clearly where you can stop. Practical, not visionary No predictions about the future of work. Instead: twenty ready-made prompts, twelve worked examples, three checklists, a model and price reference, and the failure modes that look like successes. Scroll up and get your Tuesday back.