On Writing

I keep a mental list of rules about writing. Not grammar rules. Not “never split an infinitive” pedantry. Rules about how to make someone actually understand what you mean, and maybe even enjoy the process of understanding it. Rules I’ve collected from writers I admire, from editing my own terrible first drafts, from reading sentences that made me stop and re-read them because they were that good.

This post is that list. Fourteen rules, each one a small essay. Some are about mechanics. Some are about mindset. A few are about the strange relationship between writing and thinking, which turns out to be the most important thing here and the reason writing still matters in a world where you can ask a language model to generate ten paragraphs on any topic in four seconds.

These aren’t universal laws. Orwell’s sixth rule of writing was “Break any of these rules sooner than say anything outright barbarous,” and that’s the right attitude. But these are the rules I come back to when my writing isn’t working, and they fix the problem more often than not.

Rule 1: Writing is thinking. Not a record of thinking. The thinking itself.

Most people treat writing as the last step. You figure out what you want to say, then you write it down. That’s backwards.

The act of writing is where the figuring out happens. You sit down with a vague idea, start putting words on a page, and discover that the idea has holes you didn’t notice when it was just floating around in your head. The holes were always there. Writing made them visible.

This is William Zinsser’s core argument in On Writing Well: “Writing is thinking on paper.” Not recording thoughts. Generating them. The reason your first draft feels like a mess is that it IS a mess, because your thinking was a mess, and now you can see it for the first time. That’s not failure. That’s the process working.

Paul Graham put it differently in his essay Putting Ideas into Words: “A good writer doesn’t just think, and then write down what he thought, as a sort of transcript. A good writer will almost always discover new things in the process of writing.” He’s describing the same phenomenon. The writing isn’t downstream of the thinking. They’re the same activity.

This has a practical consequence that most people miss. If you want to think more clearly about something, write about it. Don’t wait until you’ve “figured it out.” Write your way to figuring it out. Start with what you know, articulate it precisely, notice where your explanation breaks down, and follow the breakdown until you find the gap in your understanding. Then fill the gap. Then keep writing.

This is also why writing still matters in the age of AI. Yes, Claude or GPT can generate ten paragraphs about any topic. But the paragraphs aren’t the point. The thinking that happens during the writing is the point. When you outsource the writing, you outsource the thinking. You end up with a document that sounds polished but that you don’t fully understand, because you never went through the cognitive work of producing it.

Use AI to edit, to challenge your drafts, to catch errors, to suggest restructures. But do the first draft yourself, because the first draft is where the thinking happens.

Rule 2: Cut until it hurts. Then cut more.

Zinsser called clutter “the disease of American writing.” He was right in 1976 and he’s more right now. The internet incentivizes verbosity. SEO rewards word count. AI generates fluent filler at near-zero cost. The result is a world drowning in text that takes 2,000 words to say what could be said in 400.

The fix is simple and painful: cut. Cut adjectives. Cut adverbs. Cut throat-clearing phrases. Cut sentences that restate the previous sentence in slightly different words. Cut paragraphs that don’t advance the argument. Cut sections that exist because you feel like the piece “should” cover that topic, even though you have nothing interesting to say about it.

Stephen King’s version of this: “Kill your darlings, kill your darlings, even when it breaks your egocentric little scribbler’s heart, kill your darlings.” He’s quoting Faulkner (who may have been quoting Quiller-Couch), and the advice is about a specific failure mode: the sentence you love so much that you keep it even though it doesn’t serve the piece. That sentence is the first one to cut.

Here’s a diagnostic test. Take any paragraph you’ve written and try to delete every sentence, one at a time. For each one, ask: does the paragraph lose anything important without this sentence? If the answer is no, the sentence is clutter. You’ll be surprised how often the answer is no.

Cluttered version	Cut version
“Due to the fact that the system experienced a significant number of failures during the testing phase, we made the decision to postpone the deployment.”	“Testing found too many failures, so we postponed deployment.”
“It is important to note that the performance of the algorithm is highly dependent on the quality of the input data.”	“Algorithm performance depends on input data quality.”
“In order to facilitate a better understanding of the architecture, I will provide a comprehensive overview of the major components.”	“Here’s how the architecture works.”

The cluttered versions aren’t wrong. They’re just slow. They waste the reader’s time with empty calories. The cut versions say the same thing in fewer words, which means the reader gets to the point faster, which means they’re more likely to keep reading.

Elmore Leonard compressed this entire idea into his tenth rule of writing: “Leave out the parts readers tend to skip.” If you’re honest about which parts those are, you’ll cut about 30% of everything you write. The remaining 70% will be twice as good.

Rule 3: Use plain words. Not dumb words. Plain words.

There’s a persistent misconception that sophisticated ideas require sophisticated vocabulary. That if you’re writing about complex topics, you need complex language. This is wrong. The best explanation of a complex topic is the one that uses the simplest language that remains precise.

Orwell’s second rule: “Never use a long word where a short one will do.” Not “never use long words.” Use short ones where they will do. If “use” works, don’t write “utilize.” If “start” works, don’t write “commence.” If “buy” works, don’t write “procure.” But if “idempotent” is the right word because no shorter word captures that exact meaning, use “idempotent.”

The test is always: does the fancy word add precision, or does it add prestige? Precision is worth paying for. Prestige is not. Nobody is impressed by your vocabulary. People are impressed by your clarity.

Paul Graham takes this further in Write Like You Talk. His advice: write your draft however you normally would. Then go back through every sentence and ask, “Is this how I’d say it if I were talking to a friend?” If not, rewrite it until it is. He claims that if you do only this one thing, you’ll be ahead of 95% of writers.

That sounds too simple. It’s not. Most bad writing is bad because the writer is performing instead of communicating. They’re trying to sound like a Writer, capital W, instead of trying to be understood. The moment you start performing, you reach for longer words, more complex syntax, more formal phrasing. You start writing sentences that no human would ever say out loud. And those sentences are harder to read, which means fewer people understand them, which means your writing has failed at its one job.

Here’s what plain language looks like in technical writing:

Bad:  "The system leverages a microservices-based architecture
       to facilitate horizontal scalability."

Good: "The system uses microservices so it can scale horizontally."

Bad:  "We conducted an investigation into the root cause of the
       performance degradation and subsequently implemented
       a remediation strategy."

Good: "We found what was causing the slowdown and fixed it."

Same information. Half the words. Twice the clarity.

Rule 4: The first sentence does one job: make them read the second sentence

This is copywriting advice that applies to all writing. The purpose of the first sentence is not to summarize the article, introduce the topic, provide context, or establish your credentials. The purpose of the first sentence is to make the reader want to read the second sentence. That’s it. One job.

The purpose of the second sentence is to make them read the third. And so on. Every sentence earns the next one.

This doesn’t mean every opening needs to be a cliffhanger or a shock. It means the opening needs to create a question in the reader’s mind, one they’ll stick around to get answered. The question can be explicit (“How do you build a 14-agent system that actually works?”) or implicit (a surprising claim, a concrete detail that feels out of place, a confession).

Compare these openings:

Weak opener	Strong opener
“In this blog post, I will discuss my experience building a physics engine and the challenges I encountered.”	“I spent the last few months building a physics-realistic billiards engine. Not a toy demo.”
“Writing is an important skill that everyone should develop.”	“I keep a mental list of rules about writing.”
“This article explores the intersection of AI and software engineering.”	“The agents are not the hard part. The orchestration protocol is everything.”

The weak openers are meta-commentary. They describe what the article will do instead of doing it. The strong openers drop you into the middle of something. They raise questions. They have specificity. The reader thinks “tell me more” instead of “I know where this is going.”

One concrete technique: write your entire first draft, then delete the first paragraph. Your second paragraph is almost always a better opening. The first paragraph is throat-clearing. You needed to write it to get warmed up, but the reader doesn’t need to read it.

Rule 5: One idea per sentence. One point per paragraph.

This is the most mechanical rule on this list and maybe the most useful one. When a sentence is confusing, it’s almost always because it’s trying to do two things at once. When a paragraph is confusing, it’s because it’s making two points instead of one.

A sentence should express one complete idea. If you find yourself writing a sentence with two independent clauses joined by “and” or “but” where each clause could be its own sentence, split them. If you find yourself writing a sentence with three prepositional phrases nested inside each other, restructure.

Overloaded sentence:
"The system processes incoming requests through a series of
middleware functions that validate authentication tokens and
check rate limits and then routes them to the appropriate
handler based on the URL pattern, which is defined in a
configuration file that's loaded at startup."

Split into one-idea-per-sentence:
"Incoming requests pass through middleware that validates
authentication tokens and checks rate limits. The system
then routes each request to a handler based on its URL
pattern. These patterns are defined in a configuration
file loaded at startup."

Three sentences instead of one. Each sentence has one job. The reader processes each idea fully before moving to the next. The total word count is about the same. The comprehension rate is dramatically higher.

The paragraph version of this rule: each paragraph should make one point. The first sentence of the paragraph states the point (or sets it up). The remaining sentences support, elaborate, or provide evidence for that point. When you catch yourself making a new point, start a new paragraph.

Paragraphs are free. Use them generously. A wall of text is intimidating. A series of short paragraphs is inviting.

Rule 6: Read it out loud

This is the advice that every good writer gives and most aspiring writers ignore. Read your draft out loud. Not in your head. Out loud, with your actual voice, making sounds with your mouth that travel through the air.

When you read silently, your brain auto-corrects problems. It fills in missing words, smooths over awkward phrasing, glosses over repetition. Your brain is too good at understanding what you meant to say. But your ear is brutal. When you read out loud, you hear every stumble. You hear where the rhythm breaks. You hear where you run out of breath because the sentence is too long. You hear where the same word appears three times in two sentences. You hear where the prose clunks instead of flows.

Zinsser did this. Graham does this (he reads his essays aloud before publishing). King does this. The technique works because writing is, ultimately, a spoken art that happens to be written down. Good prose has a cadence. It alternates between long sentences and short ones. It varies rhythm to maintain interest. Bad prose is monotonous: every sentence the same length, the same structure, the same beat.

If you don’t want to read out loud (or you’re in a coffee shop and don’t want to be that person), there’s a modern substitute: text-to-speech. Paste your draft into a TTS engine and listen. It’s not as good as reading it yourself, because the act of speaking activates different cognitive processes than listening. But it’s better than nothing.

Rule 7: Strong verbs. Weak adverbs.

This is the single highest-leverage edit you can make to any sentence: replace a weak verb + adverb combination with a strong verb that carries the meaning on its own.

Weak verb + adverb	Strong verb
“ran quickly”	“sprinted”
“said loudly”	“shouted”
“walked slowly”	“shuffled” or “ambled”
“looked at carefully”	“examined” or “scrutinized”
“completely destroyed”	“obliterated”

Adverbs are often a sign that you chose the wrong verb. If you need to modify a verb to make it say what you mean, you probably need a different verb. Stephen King is famously anti-adverb: “The road to hell is paved with adverbs.” That’s overstated, because adverbs have legitimate uses. But as a default instinct, “find a better verb” is almost always the right move.

This applies to technical writing too. “The function efficiently processes the data” becomes “The function processes the data in O(n log n).” The adverb “efficiently” is vague. The complexity class is precise. In technical prose, the strong-verb principle generalizes to: replace vague modifiers with specific measurements.

Vague:   "The API responds quickly."
Precise: "The API responds in under 50ms at p99."

Vague:   "The model performs well on the benchmark."
Precise: "The model scores 92.3 on MMLU, up from 87.1."

Vague:   "We significantly reduced memory usage."
Precise: "Memory usage dropped from 2.4GB to 800MB."

Numbers are the ultimate strong verb. Whenever you’re tempted to write “significantly,” “substantially,” “dramatically,” or “considerably,” ask yourself: can I replace this adverb with a number? If yes, do it. If no, at least acknowledge that you’re being vague and explain why.

Rule 8: Structure is kindness

A well-structured piece of writing is an act of generosity toward the reader. It says: “I’ve organized my thoughts so you don’t have to.” A poorly structured piece says: “Good luck figuring out what I’m talking about.”

The structure of a piece should be visible. Not hidden in some clever rhetorical scheme that English professors find delightful. Visible. Headers, subheaders, paragraph breaks, topic sentences. The reader should be able to skim the piece and get the gist from the structure alone, then read the full text for depth.

This is doubly true for technical writing. A technical document with no headers is a hostage situation. The reader is trapped in a wall of text with no landmarks, no way to skip to the part they need, no way to assess whether the document even covers their question.

Here’s a minimum viable structure for most technical writing:

1. What is this about? (one paragraph, three sentences max)
2. Why does it matter? (context, motivation, the problem)
3. How does it work? (the core explanation, with examples)
4. What are the edge cases? (limitations, gotchas, caveats)
5. What should I do next? (actions, links, references)

This isn’t the only valid structure. But if you have no structure, start with this one. It covers the questions that most readers bring to most technical documents. And it’s infinitely better than the alternative, which is a 3,000-word paragraph that starts at the beginning and ends when the author runs out of things to say.

Tables deserve a special mention here. Tables are the most underused structural tool in writing. Any time you’re comparing two or more things across the same dimensions, a table communicates the comparison faster and more clearly than prose. Don’t write “Option A has low latency but high cost, while Option B has medium latency and low cost, and Option C…” Just make a table:

Option	Latency	Cost	Complexity
A	Low	High	Medium
B	Medium	Low	High
C	High	Low	Low

Three seconds to read. The prose version would take thirty seconds and the reader would lose track of the dimensions halfway through.

Rule 9: Delete your throat-clearing

Throat-clearing is what writers do at the beginning of a piece (or section, or paragraph) to warm up. It’s the written equivalent of “um, so, basically, the thing is…” It serves a purpose for the writer: it gets you into the flow. It serves no purpose for the reader.

Common throat-clearing patterns:

"Before we get into the details, it's important to understand..."
"As we all know..."
"It goes without saying that..." (then why are you saying it?)
"For the purposes of this document..."
"Having established the background, we can now..."
"With that context in mind..."
"As mentioned earlier..."
"To put it simply..."

Every one of these can be deleted. The sentence that follows the throat-clearing is where the actual content starts. Lead with that.

A subtler form of throat-clearing happens at the paragraph level. Many writers start paragraphs with a transition sentence that exists only to connect to the previous paragraph. “Building on this point, we can also consider…” Just start with the point. The reader can handle a transition that happens in whitespace. That’s what paragraph breaks are for.

The test: read the first sentence of each paragraph. If it could be deleted without losing information, delete it. Your writing will feel more direct, more confident, and more respectful of the reader’s time.

Rule 10: Write for the reader who’s about to close the tab

Here’s the uncomfortable truth about all writing on the internet: your reader is one boring paragraph away from leaving. They have thirty other tabs open. Their phone just buzzed. A Slack notification just popped up. They are actively looking for a reason to stop reading.

This changes how you should write. You can’t afford a slow build. You can’t afford a paragraph that’s “necessary context” but not interesting. You can’t afford to repeat yourself, because the reader who’s already thinking about their other tabs will interpret repetition as “I’m not learning anything new” and leave.

Every paragraph needs to earn its keep. Every paragraph should either teach the reader something they didn’t know, make them laugh, surprise them, or create a question they need answered. If a paragraph does none of these things, it needs to go.

This isn’t about dumbing things down or writing clickbait. It’s about respecting the reader’s attention as a scarce resource. The best technical writing is dense with information. Not dense in the sense of impenetrable, dense in the sense of high signal per sentence. There’s always something happening. There’s always a reason to keep reading.

Practical technique: after finishing a draft, go through each paragraph and write a one-sentence summary of what the reader learns from it. If you can’t write that summary, or if the summary is “this paragraph provides context,” the paragraph probably needs to be rewritten or cut.

Rule 11: Steal structures, not sentences

Every experienced writer has a collection of structural templates that they deploy across different pieces. This isn’t plagiarism. It’s craft. You don’t copy someone’s words; you study how they organized their ideas, and you adapt that organization to your own material.

Some structures worth stealing:

The inverted pyramid (journalism): lead with the conclusion, then provide supporting detail in decreasing order of importance. Good for announcements, changelogs, incident reports. The reader who leaves after the first paragraph still got the most important information.

The problem-solution-result (case study): describe the problem in concrete terms, walk through the solution, show the outcome with numbers. Good for project write-ups, “how we built X” posts.

The numbered collection (Elmore Leonard, Umberto Eco): a list of principles, each one a self-contained mini-essay. Good for advice posts, lessons-learned posts. (This post uses this structure.)

The before-and-after (Zinsser): show bad writing, then show the edited version, and explain each change. Good for teaching craft.

The narrative with digressions (Paul Graham): make a central argument, but allow yourself to explore tangents when they’re interesting, then return to the main thread. Good for essays where the thinking process matters as much as the conclusion.

Read writers you admire and pay attention to structure, not just content. When you find a structure that works, add it to your toolkit. The more structures you have available, the more likely you are to find the right one for any given piece.

Rule 12: The three books you should actually read

Most “recommended reading lists” for writers are 30 books long and nobody reads any of them. Here are three. Read these three and you’ll have a better foundation than 95% of people who call themselves writers.

1. On Writing Well by William Zinsser (1976, updated through 2006)

This is the bible of nonfiction writing. Zinsser’s thesis is that clarity is the product of clear thinking and ruthless editing. The book covers principles (simplicity, clutter, style, audience) and then applies them to specific genres: travel writing, science writing, memoir, business writing. Read Part I (Principles) and Part IV (Attitudes) at minimum. The chapters on Clutter and Simplicity should be required reading for anyone who writes professionally.

2. On Writing: A Memoir of the Craft by Stephen King (2000)

Half memoir, half writing instruction. The memoir half is one of the best accounts of a writer’s formation ever written. The instruction half is practical and opinionated. King’s advice on vocabulary (“use the first word that comes to your mind, if it is appropriate and colorful”), on plotting vs. situating, on the importance of reading widely, on revising with the door closed and then with the door open. The formula he gives for revision: “2nd Draft = 1st Draft - 10%.” Simple, memorable, correct.

3. Politics and the English Language by George Orwell (1946)

Not a book. An essay. You can read it in 20 minutes. Orwell’s argument is that bad writing is both a cause and a symptom of bad thinking, and that the deliberate corruption of language is a political act. His six rules for writing have been quoted so often they’ve become cliché, but they’re cliché because they’re right. Rule six (“Break any of these rules sooner than say anything outright barbarous”) is the most important and the most often forgotten.

Honorable mentions: The Elements of Style by Strunk & White (controversial, but the chapter on “Omit Needless Words” is a masterclass), Bird by Bird by Anne Lamott (for when you need permission to write badly on the first draft), and everything Paul Graham has published at paulgraham.com on the subject of writing (start with Write Like You Talk, then read Putting Ideas into Words, then Writing, Briefly).

Rule 13: Develop a practice, not a habit

“Write every day” is the most common writing advice and the least useful. It treats writing like cardio: just do it enough times and you’ll get better. That’s half true. You do need volume. But volume without feedback is just repetitive motion.

A writing practice is different from a writing habit. A habit is: sit down, write for 30 minutes, check the box. A practice is: sit down, write something, then examine what you wrote and identify what’s working and what isn’t. A practice includes reading other writers and paying attention to craft. It includes editing, which is a different skill from writing and arguably a more important one. It includes studying structure. It includes sometimes writing badly on purpose to understand why it’s bad.

Specific things you can do:

Keep a swipe file. When you read a sentence, paragraph, or article that’s unusually good, save it. Don’t just think “that was nice” and move on. Copy it into a document. Periodically review the document and try to articulate what makes each example work. This trains your taste, which is the precondition for improving your craft.

Edit something you wrote six months ago. You will be horrified. That’s good. The horror means your standards have improved. The editing itself will teach you more about writing than writing something new, because you can see the mistakes clearly now that the emotional attachment has faded.

Write the same idea three different ways. Take a paragraph from a draft and rewrite it as: (a) a single sentence, (b) a three-sentence version, and (c) a version for a reader who knows nothing about the topic. Each rewrite forces different decisions and reveals different aspects of the idea.

Read your writing to someone. Not “ask someone to read your writing.” Read it to them, out loud, while they sit there. You will feel every weak moment in your body. You will know exactly where the writing fails because you will feel embarrassed reading it. That embarrassment is the most honest editor you will ever have.

Rule 14: Revise. Then revise again. Then revise.

First drafts exist to be rewritten. This is not a deficiency. This is the process. Every writer whose work you admire, without exception, writes bad first drafts. The difference between them and a mediocre writer is not that their first drafts are better. It’s that they revise more carefully and more ruthlessly.

Here’s a revision process that works:

Pass 1: Structure. Read the entire draft without editing anything. Just read. Then ask: does the order of ideas make sense? Is there a logical progression? Are there sections that should be moved, merged, or deleted? Do the restructuring before you touch any prose. Editing sentences in a section you’re going to delete is wasted work.

Pass 2: Paragraphs. Go through each paragraph. Does it make one point? Is the topic sentence clear? Could any sentences be cut without losing the point? Are there paragraphs that repeat a point made earlier? Merge or cut them.

Pass 3: Sentences. Now go through each sentence. Is it clear? Is it concise? Does it use plain language? Are the verbs strong? Are there unnecessary adverbs or adjectives? Is it the way you’d say it to a friend?

Pass 4: Read aloud. Read the entire piece out loud. Fix everything that sounds wrong. This is where rhythm, flow, and voice get polished.

Pass 5: Cut 10%. Stephen King’s formula. Your draft is done? Now cut 10% of the word count. Not by deleting random sentences. By tightening every paragraph until it’s as lean as it can be. This pass is where good writing becomes great writing.

Five passes sounds like a lot. For a blog post, each pass takes 15-30 minutes. For a short piece, less. The investment is small relative to the improvement. And with practice, the passes get faster because you start internalizing the patterns and catching problems earlier in the process.

---

I want to close with something that might sound contradictory after 14 rules about writing: the rules matter less than you think they do.

Not because they’re wrong. Because knowing the rules and applying the rules are different skills. You can memorize every rule on this list and still write badly, the same way you can memorize every chess opening and still lose. The rules are a starting point. They give you a vocabulary for talking about what’s wrong with a piece of writing and a set of techniques for fixing it. But the thing that actually makes you a better writer is writing a lot, reading a lot, and being willing to look at your own work with clear eyes and say “this isn’t good enough yet.”

The best writing advice I’ve ever encountered is the simplest. It’s not from any of the famous books or essays. It comes from a conversation I had with a friend who writes well. I asked him how he got good. He said: “I wrote a lot of bad stuff, and then I tried to figure out why it was bad, and then I tried to make the next thing less bad.”

That’s the whole game. Write. Examine. Improve. Repeat.

The rules help. The books help. The practice helps. But the core loop is the same thing it’s always been, before AI, before the internet, before typewriters. Put words on a page. Make them better. Do it again tomorrow.