Hooks & Openings
The first line decides whether the rest gets read
Universal tool
The opening is the whole game. If the first line doesn't earn the second, nothing downstream matters — not the proof, not the offer, not the writing. This page is how to open. The anatomy of a hook, the mechanism that makes it pull, the nine types you can reach for, and a system to make them by the dozen.
Hook anatomy, taxonomy and the 70-20-10 split adapted from Alex Hormozi's hooks playbook; the source's hustle-culture tone is deliberately dropped. The three magnetizers (job 02) and the mystery open (job 05) are adapted from Cialdini's research on attention — findings cited, no prose reproduced. Every example is written in your voice on your topics.
01
The anatomy — call-out + value
Every hook does two jobs at once. It calls out the right reader so they think "that's me," and it promises that reading on will pay off. Miss the call-out and the right person scrolls past; miss the value and they don't see the point. You need both.
Call-out — who it's for
Building agents that keep forgetting things?
Value condition — what they get
Here's the memory ladder that fixes it.
Assembled hook
Building agents that keep forgetting things? Here's the memory ladder that fixes it.
Hook anatomy
Two parts, always
A call-out that says "this is for you" + a condition that promises value.
The reader engages only when felt cost < expected value. The hook tips that math.
Yours: the call-out is the reader's problem in the reader's words — the same vocabulary move as the whole system.
02
The magnetizers — why the anatomy works
Job 01 says what a hook is made of. This says why those parts pull. Three things magnetize attention reliably, and two of them are already doing the work in the anatomy above — knowing which is which is what lets you fix a hook that isn't landing instead of rewriting it at random.
Magnetizer 01
The self-relevant
Information about me gets processed ahead of everything else, even when I'm not trying. It's why your own name cuts through a noisy room.
PowersThe call-out. And it's why the call-out has to be their problem in their words — a demographic label isn't self-relevant, a frustration is.
Magnetizer 02
The unfinished
An incomplete thing holds attention that a completed one releases. The mind keeps a tab open until the loop closes.
PowersThe value condition, and every bullet you'll ever write. A promise of payoff is an open loop with a benefit attached.
Magnetizer 03
The mysterious
A posed puzzle does both at once — it's about something the reader cares about, and it's unresolved. The strongest of the three.
PowersIts own hook type. Job 05 below.
The magnetizers
Self-relevant, unfinished, mysterious
The mechanism under the anatomy. Diagnose with these before rewriting.
Hook not landing? Ask which magnetizer is missing. Usually it's self-relevance — the line is about the topic instead of about them.
Every loop you open must close in the piece. An unresolved one works once and costs you the next hook.
The line: directing attention works partly below awareness. That's fine when you're pointing at something true — true framing survives being noticed. Concealed intent doesn't. See gate 1 on
The Honest Line.
03
Make the reader feel important, not you
The oldest lesson in hooks: an opening that's about the writer falls flat, and one that makes the reader feel seen pulls them in. "Here's how I did it" centers you. "Here's what this changes for you" centers them. This is your reader-first brand voice, showing up as a hook rule.
✓ About the reader
You're one prompt away from fixing this.
The thing your agent keeps getting wrong.
✕ About the writer
How I built a 100-agent system.
My framework for agent memory.
Reader-first
Their win, not yours
The hook makes the reader feel important. Your credentials come later, if at all.
"Here's what changes for you" beats "here's what I did" every time.
Same idea as magnetizer 01, stated as a rule: a hook about you cannot be self-relevant to them.
04
The hook types — a toolkit to reach for
When the blank line stares back, reach into the toolkit. Nine verbal hook types, each a different way in. These pair with the three LinkedIn modes — the mode is the post's shape, the hook is how it opens. Same post, many possible doors.
LabelPeople building with agents, this one's for you.
QuestionWhy does your agent keep confidently making things up?yes/no or open — pulls the reader to answer in their head
ConditionalIf your agent forgets everything between sessions, the problem isn't the model.
CommandStop giving your agent more context. Give it better structure.
StatementMost agent failures aren't model failures. They're prompt failures.
ListFive root causes sit under all 22 agent failure modes.
NarrativeA family reunion taught me how far ahead my AI knowledge actually was.
ExclamationThe term "AI slop" is almost always aimed at the wrong culprit.
MysteryTwo teams shipped the same agent architecture. One is still running it. The other tore it out in a fortnight.pose the puzzle, then earn the answer — see job 05
Hook taxonomy
Nine doors in
Label, question, conditional, command, statement, list, narrative, exclamation, mystery.
Not a blank page — pick a type and fill it with your one thing.
Write the same hook three ways using three different types before picking. The type you reach for first is a habit, not a decision.
Yours: pairs with the three modes. Mode = the post's shape; hook = how it opens.
05
The mystery open — pose, hold, resolve
The strongest of the nine, and the one that fits this brand best. It came from studying which popular-science writing actually held readers: the sections that worked opened by describing something genuinely puzzling and inviting the reader in to sort it out. That's your genre — technical ideas explained to people who don't already have them, in public, every day.
PoseTwo teams shipped the same agent architecture. One is still running it eighteen months later. The other tore it out in a fortnight.
HoldSame model, same framework, same size of team. The difference wasn't engineering.
ResolveOne of them wrote down what the agent was allowed to forget. That's the whole difference, and it's the thing nobody scopes.
Mystery open
Three moves, one rule
Pose a real puzzle. Hold the tension. Resolve it with the thing you came to teach.
The resolution has to be the lesson. If the payoff is separable from the puzzle, you wrote a clickbait line and a post, not a mystery.
It has to be genuinely puzzling to them, not to you. Test: could a reader guess the answer from the opening? Then there's no loop.
Yours: a mystery makes the reader reason rather than receive. It's the teach-first stance at the scale of one sentence — and it slots straight into the six-beat, where the turn is the resolution.
06
The 70-20-10 system — make them at volume
You don't gamble on one hook; you produce a batch and let the winners compound. Most of the effort goes to what already works, some to adapting formats that worked elsewhere, a little to wild swings. Winners graduate up; losers get logged so you don't repeat them. It's your reuse philosophy, aimed at openings.
70% Reuse hooks already proven — yours or others', reworked into your voice.
20% Model a format that won in another niche; adapt it to agents/AI.
10% Wild swings. Winners graduate to the 70%; losers get logged, not repeated.
70-20-10
Bet mostly on winners
Most effort on what works, some on adjacent, a little on wild.
Winners graduate up. Losers documented so they're not repeated.
Yours: this is the content engine's reuse rule — cut, don't re-invent — applied to hooks.
07
The swipe file — let the data teach you
The 70-20-10 system needs a memory, and that memory is a running file of your own best hooks with how each one performed. Before writing a new opener, you review the file. "What works" becomes data you can see, not a feeling you half-remember.
| Hook | Ran on | Signal |
|---|
| Not 22 problems. Just five. | Card + LinkedIn | reshared |
| Complexity is earned, not adopted. | Memory Ladder card | saved |
| The prompt is usually to blame, not the model. | LinkedIn | high |
| Your agent isn't broken. Its memory is. | Signal Thinking | test |
Review the file before writing anything new.
The swipe file
Your winners, tracked
A running list of your best hooks and how each performed.
"What works" is data you review, not memory you trust.
Yours: the same early-signal tracking the content engine already runs — which cards got reshared, by whom.