The AI Practitioner's Playbook: From first prompt to results you can rely on

The AI Practitioner's Playbook Anyone can get an impressive-looking answer out of an AI assistant within ten minutes. Almost nobody learns the part underneath. This is the training manual for that part: a complete, sceptical, practical course in working with ChatGPT, Claude, Claude Cowork and Claude Code, built for professionals whose output has to survive review. No prior experience is assumed. You can start in twenty minutes and build the skill over ninety days. Most people plateau at the surface. They get a plausible paragraph, paste it somewhere, and never learn the things that actually matter: knowing when the answer is wrong, phrasing a request so it is right the first time, knowing which of four products to reach for, knowing what must never be pasted into a chat box, and knowing when to stop trusting the machine. The difference is not technical background. It is a handful of mental models and one repeatable loop, and this book is both. Brief it: say what you want precisely enough that the result can be judged. Run it: pick the surface that fits, whether a chat message, an agent with file access, or a chain of tools. Check it: verify the artefact, not the narrative, because that is what separates professional use from hoping. Improve it: feed what you learned back into the brief. Every chapter serves that loop. INSIDE - How a language model actually works, and the seven ways it fails, with the symptoms and the fix for each- A complete prompting method: six components, nine techniques, and a diagnostic loop for repairing bad output- Product parts for ChatGPT, Claude, Claude Cowork and Claude Code, with particular emphasis on the two agentic tools: agent briefing, permission models, project memory, verification gates- Verification tiered by consequence, prompt-injection defence, cost control, and a governance chapter built around the EU AI Act- A worked end-to-end case study, competence checks for every part, 30 reusable prompt templates and a 90-day learning plan It is written for professionals and knowledge workers whose work carries consequences: consultants, analysts, finance and tax teams, lawyers, project leads, and anyone whose output has to hold up under scrutiny. It assumes no prior experience whatsoever, and it does not assume you enjoy the technology, only that you need it to be reliable. One thing sets it apart. Every substantive claim about how these tools behave is labelled by evidence type, whether vendor documentation, research-supported, practitioner heuristic or organisation-specific, and sourced in the bibliography. Every product part carries a verification date, so you can see how current a claim is rather than guess. A book that asks its readers to verify should hold itself to the same standard.