Have you ever closed a session with an LLM convinced that your argument was more solid than before — without a single piece of evidence having changed?
AI-assisted outputs that sound brilliant but, by the end of the day, leave you feeling you have thought less, not more. Documents that are longer, less read, never challenged. Decisions signed off in a hurry — and overturned just as fast in the months that follow. A three-page "AI policy" that nobody remembers reading and that is already obsolete. If this sounds familiar, this book is for you.
Thinking with LLMs, the Right Way is not yet another prompt-engineering manual. It is a manual of critical governance for thinking with LLMs — a system of critical thinking applied to the daily use of generative AI, fuelled by years of classroom practice with teams from every sector — to stop slipping into confirmation mode and start using generative AI as a tool for augmenting thinking, not replacing it.
What you'll find inside:
- The Thinking-With Triangle (Intent / Adversary / Editor): three vertices always active, one rule — when one collapses, you're not using the LLM, the LLM is using you
- The 2×2 matrix clocks/clouds × truth/confirmation: a diagnostic map to work out in 30 seconds whether the session you're about to open is the right case for an LLM
- Four Socratic patterns (Elenchus, Maieutic, Aporia, Dialectic) for investigating a badly posed problem, and Chain-of-Verification with the deductive/inductive/abductive triad for verifying an output before signing it — calibrated to the cost of the error, not the type of task
- Four explicit meta-decisions (What to delegate · How to verify · Who is responsible · Abandonment threshold) for team governance — not a document, an operational pact
- Three drifts to recognise — mirror, inertia, agenticity — with their recovery patterns
Francesco Fullone, author of the the Right Way series (KPI, OKR, Sustainable IT, Business Innovation, Business Design, Theory of Change), founder of Daruma Consulting and a trainer at Bologna Business School, H-Farm, Bi-Rex and Talent Garden, condenses years of practice with LLMs into 8 chapters (plus an introduction) that will take you from the frustration of "I used AI but came out no better at my job" to the posture of the professional who signs every output in their own voice.
This book is for you if:
- You're tired of AI-assisted outputs that don't hold up to a severe re-reading
- You want to understand the difference between thinking-with and delegating-to an LLM
- You have to design your team's AI policy — and don't know where to start beyond generic bans
- You're looking for a method, not yet another tutorial on the prompt of the moment
- You work with an LLM every day and notice that, after six months, you're more productive and less capable
Includes:
- 8 inline teaching figures (the Triangle, the 2×2 matrix, the paradox of the augmented brain, the three drifts, the Socratic-adversarial quadrant, the tilted mirror, the four Socratic patterns, false/real onboarding) with descriptive captions
- Appendix C — Operational roadmap for the first four weeks
- Appendix D — Fillable materials (the Four Explicit Decisions of the role template, a complete Socratic session with an example, an LLM onboarding checklist for the team)
- The open-source adversarial-verify skill for Claude Code (MIT licence): the book's method crystallised into an executable workflow, with an independent release cycle
- A reasoned bibliography with 49 sources indexed for cross-reference — critical philosophy, cognition and decision-making, AI applied to work, adversarial engineering
Corporate workshops — available separately
The book teaches the method; the corporate workshop Thinking with LLMs, the Right Way transfers it into the classroom. Available in three formats — 2h cultural taster, 4h core (recommended), 6h deep — it includes a facilitator guide, a student workbook, six A2/A3 operational canvases, four illustrated table instructions, classroom simulations.
To bring it into your organisation: contact me via DarumaHQ.it
Book contents under CC BY-NC-SA 4.0 licence · adversarial-verify skill under MIT licence.
