Pick up here. Everything below is established. Do not re-derive. Do not run ahead.
Book title: Mind the Gap
Subtitle (working): What AI sounds like. What AI actually is. And why the difference matters.
Author: Paul Roebuck
Status: Project established. Proposal not yet written. No chapters drafted. Starting point confirmed.
Humans are wired to interpret fluent language as human. AI now speaks fluently enough to trigger every social and psychological instinct we have — while operating on entirely different mechanics underneath.
That gap — between what AI sounds like and what it actually is — is the most important thing to understand about the world we now live in.
The author is a psychotherapist who has spent thirty years listening to exactly that kind of gap in humans. Same instrument. New room.
Read the full Layers map (uploaded separately as Paul_Roebuck_The_Map_of_the_Layers_v4.docx). Summary below.
The bio block (locked compression):
For thirty years, personally and professionally in boardrooms and more lately as a psychotherapist, I've been asking one question: why do people do what they do — and what stops them doing it differently. I listen to what people say, and hear what they don't. I pay attention to behaviour for a living.
Key facts:
The Layers map visual asset: exists as a full infographic (uploaded). Seven floors, one architecture.
Twelve chapters. One per slide of the visual deck (uploaded as images — full deck visible).
| Chapter | Slide Title | Core Idea |
| 1 | Human Recognition | Claude behaves in ways the brain instantly recognises. Layered context, not human mind. |
| 2 | The Human Interpretation Trap | We project human meaning onto system behaviour. Fluency triggers trust. Trust creates the trap. |
| 3 | The Hidden Architecture | Every response built from stacked layers of active context. Claude re-processes, not remembers. |
| 4 | Drift | Conversations migrate. The danger is not drift — the danger is that the answer still looks correct. |
| 5 | Compression: The Silent Transformation | As context fills, older content compresses. Not forgetfulness. The context window. |
| 5.5 | The Context Window Has a Limit | The model does not forget like a human. It runs out of room. |
| 6 | Fluency Creates False Confidence | Fluency is not understanding. Coherence is not correctness. Always apply healthy verification. |
| 7 | Context Window, Models and Privacy by User | Different access. Different limits. Different responsibilities. (Slide 8 in deck — user tiers) |
| 8 | Context Window Visualised | More window = more to work with. But every window has a limit. |
| 9 | Sub-Personalities for AI Fluency | AI fluency is not one personality. It's a set of operating modes. |
| 10 | Same Language or Different Game? | You have a choice. Keep talking like a human. Or speak the system's language. |
| 11 | The Future Choice: AI or Human? | Businesses will offer a choice. Most customers will choose AI for most things. |
| 12 | Audio, Wearables and the Next Interface | The conversation is leaving the screen. Everything this book has taught gets harder to notice. |
Note on Chapter 12: Slide 12 is the weakest in the current deck — it lists rather than lands. The chapter argument is: when the screen disappears, so does your ability to check. Same system. Different vulnerability. Slide needs redeveloping before chapter is written. Paul confirmed this — revisit later.
WaaS image: SaaS vs WaaS (Worker-as-a-Service). "You buy software. People do the work." vs "You buy outcomes. AI does the work." Different model. Different world. Belongs in the commercial consequence section — later chapters. When AI does the work at scale, misreading the system isn't personal anymore. It's organisational.
Fuse Energy / Plugs Facebook posts (April–May 2026): Paul posted in a consumer energy Facebook group, teaching ordinary customers how to talk to Fuse Energy's AI agent (called Plugs) in its own "dialect." This is the book's thesis in live action — a working-class customer using his own AI to talk to a company's AI. Real proof of concept. NGE-FOF shape-sorter image was part of this post. Are you the wrong shape for the world you're in? — this is a chapter frame.
Seven posts. Arc title: The Interpretation Gap.
| Post | Slide | Hook | Length | Breadcrumb Code |
| 1 | Human Recognition | You already interpret AI as if it's human. You can't help it. | Short (3–5 sentences) | BREADCRUMB-IR-01 |
| 2 | The Interpretation Trap | "It remembered me." No, it didn't. | Long (20+ sentences) | BREADCRUMB-IT-02 |
| 3 | Hidden Architecture | Every response built from layers you've never seen. | Long | BREADCRUMB-HA-03 |
| 4 | Drift | The most dangerous moment is when the answer still sounds right. | Long | BREADCRUMB-DR-04 |
| 5 | Compression | Your AI isn't forgetting. It's running out of room. | Short | BREADCRUMB-CO-05 |
| 6 | Fluency and False Confidence | The most coherent-sounding response may be the least accurate. | Long | BREADCRUMB-FC-06 |
| 7 | Context Window Has a Limit | Bridging post — pairs with Post 5. | Short | BREADCRUMB-CW-07 |
🟡 Amber: Post 6 is the strongest card. Consider whether it anchors the series end or leads a second wave.
Format intelligence (researched): Document posts (carousels) generate 39% more reach than average. Either very short (1–5 sentences) or long (20+) — moderate length underperforms. Golden hour = first 60 minutes after posting. Tue/Wed/Thu optimal.
Paul to set up a Claude Project. Upload all assets into it:
Then: One-page book proposal. Paul says go — Claude drafts.
Ideas and direction: Paul Roebuck / Words: Claude AI (Anthropic) — claude-sonnet-4-6 | May 2026