Communication

The reason writing feels hard is that it is not transcription — it is the act of finding out what you think. An idea that feels complete in your head is usually three-quarters fog, and the page is where the fog becomes undeniable. This is why outsourcing first drafts wholesale to AI is so seductive and so costly: the machine removes the struggle, and the struggle was the thinking. Use AI to challenge, compress, and polish — after you have fought the idea onto the page yourself.

The professionals who seem unusually clear in meetings are almost always people who wrote first, somewhere, even three sentences in a notebook.

Practice. Take a decision or opinion you currently hold "clearly" in your head. Write one page explaining it to a skeptic. Mark every place you got stuck — each one is a spot where the thought wasn't finished. Finish it.

Every effective communication has exactly one point — the sentence you would keep if you could keep only one. Most failed memos, talks, and pitches fail here: they have four points, which the audience compresses into zero. The discipline is brutal triage. Decide the single thing your audience must believe or do, state it early in plain words, and demote everything else to support. If you cannot say your point in one sentence, you do not have a point yet; you have a topic.

The test: an hour after your communication, what does the audience repeat to someone else? That paraphrase — not your document — is what you actually shipped.

Practice. Take your last three significant messages (memo, deck, long email). For each, write the one sentence it existed to convey. If you can't, rewrite one of them around a single point and send the new version.

School taught you to build to a conclusion; work punishes you for it. Busy people read the first three lines and decide whether the rest exists. So invert: answer first, then reasons, then evidence — the pyramid that consulting firms have drilled for fifty years because it matches how decision-makers actually read. The same structure works out loud: "I recommend X. Three reasons." Nobody has ever complained about being told the point too early.

Structure is also a kindness to disagreement: when the conclusion is upfront, people can engage with your reasoning instead of hunting for your position.

Practice. Rewrite one real document conclusion-first: the answer in line one, three supporting reasons as headers, evidence beneath each. Send both the old and new versions to a colleague and ask which they'd rather receive.

A good memo does what a meeting pretends to: it forces complete thought, gives every reader the same information, survives as a record, and respects everyone's calendar. The form is stable — context in two sentences, the question, your recommendation, the reasoning, what you need and by when. One to two pages. The organizations that run on memos (and they include the most effective companies in the world) are not writing-cultures by accident; writing is how they think together.

In an AI-saturated office the memo gains power: it is the artifact you can hand to both humans and machines, and the discipline it enforces is exactly the discipline prompting rewards.

Practice. Replace one recurring meeting this week with a memo in the form above plus fifteen minutes of written comments. Compare the decision quality and the time spent, honestly.

First drafts are written for the writer; editing converts them into something written for the reader. The moves are mechanical and merciless: cut the throat- clearing opening, replace abstractions with specifics, break any sentence doing two jobs, delete every word the sentence survives without, and read it aloud — the ear catches what the eye forgives. Half the length is usually hiding inside the draft, and the half that remains is the half that was working.

Edit for the skimmer too, because everyone is one: front-load sentences, meaningful headers, bold sparingly. A document has to work at three speeds — glance, skim, and study.

Practice. Take 600 words you've written and cut to 300 without losing a single necessary idea. Then have AI attempt the same cut, compare its choices to yours, and note what it couldn't know to keep.

Everyone's AI-assisted writing is converging on the same smooth, confident, forgettable register — and readers' detectors are improving weekly. Voice is now a signal of authenticity, which makes it an asset worth protecting. The working arrangement: you supply the point, the structure, the specifics, and the sentences that sound like you; AI supplies critique ("what's the weakest argument here?"), compression, and the tenth-draft polish you'd never have patience for. Author and editor — never the reverse.

The tell of reversed roles is content that could have been written about any company, by anyone, with nothing at stake. If a reader can't find you in it, neither can they find a reason to trust it.

Practice. Write a short piece your way, then ask AI for its three strongest criticisms — not a rewrite. Address them in your own words. Compare the result to a fully AI-rewritten version and note what each kept and lost.

A presentation is a spoken argument with visual support — not a document performed aloud. Build it in that order: the one point, the three beats that establish it, and only then slides, each carrying a single idea your voice elaborates. Reading your slides to a room transmits one message: this person did not prepare to talk to us. AI generates handsome decks in minutes now, which makes the merely-handsome deck worthless and the genuinely-argued talk conspicuous.

Rehearse out loud, at least twice, standing up. The difference between the first and third spoken run-through is the difference the audience experiences.

Practice. Take your next presentation and write the talk first: one point, three beats, in spoken language. Then build the fewest slides that support it. Rehearse aloud twice and deliver. Notice how questions change.

Humans do not remember bullet points; they remember what happened to someone. Story is the compression format the brain ships with: a person, a problem, a struggle, a change. The skill transfers everywhere — the customer anecdote that makes the strategy real, the incident narrative that makes the lesson stick, the founding story that makes the pitch human. The data earns belief; the story earns attention and memory. You need both, in that order of construction and the reverse order of delivery.

Keep an inventory. The working communicator has five true stories, polished by retelling, deployable on demand. Collect yours deliberately — the moment something instructive happens, write it down before it smooths itself into forgettable.

Practice. Write down three true stories from your own work — person, problem, struggle, change, each under ninety seconds aloud. Tell one this week where you would otherwise have presented a list. Watch the room.

The ambush question — in the meeting, after the talk, from the investor — is where credibility is actually priced. The skill is not having instant answers; it is structure under pressure. Buy three seconds ("good question — let me think about that properly"), answer the question that was asked, give your point first and one reason, and stop talking. Rambling is the tell of panic. And the most credibility-building sentence available remains "I don't know — I'll find out by Thursday," delivered without flinching.

Preparation makes spontaneity: before any high-stakes room, write the five hardest questions you could be asked and say your answers aloud once. You will be asked two of them.

Practice. Before your next important meeting, write the five hardest questions and rehearse answers aloud — point first, one reason, stop. Use an AI session as a hostile questioner for a live drill first.

Before any consequential communication, answer four questions about the receiving mind: What do they already know? What do they actually care about — not what should they care about? What do they fear? And what do I need from them? Most failed communication fails before the first word, in the unexamined assumption that the audience is you with less information. They are not you. The CFO hears "exciting technology" as "unbudgeted risk"; the engineer hears "simple change" as an insult.

This modeling is also why communication can't be fully delegated to AI: the machine knows the general audience, but you know that this particular VP was burned by exactly this kind of project two years ago.

Practice. For your next significant message, write the four answers first — knows, cares, fears, needed. Then write the message. File both; after the response arrives, score your model against reality.

Every idea exists at multiple altitudes: the one-sentence version for the hallway, the one-paragraph version for the exec, the one-page version for the decision, and the deep version for the people doing the work. Professionals can fly all four and — harder — notice which one the moment requires. The classic failure is altitude mismatch: drowning a decision-maker in implementation detail, or hand-waving at an expert who needs the mechanism. Both read as either disrespect or incompetence, and both are fixable in advance.

Build all four versions for anything that matters. AI is genuinely good at altitude conversion once you've written the deep version — it can compress honestly, but it cannot know what was important enough to survive the compression. Review what it dropped.

Practice. Take your current main project and write it at all four altitudes: one sentence, one paragraph, one page, full depth. Test the sentence on someone senior and the page on someone technical; revise where their eyes glazed.

Persuasion has had a stable recipe for 2,300 years: ethos — why you can be believed; logos — why the argument holds; pathos — why anyone should care. Modern professionals over-invest in logos and wonder why airtight arguments lose. The fix is sequencing, not manipulation: establish stake and credibility, make the case plainly, and connect it to what the audience already cares about. The line between persuasion and manipulation is also old and also simple: would you be comfortable if the audience could see your full reasoning? If yes, proceed.

In an era of synthetic persuasion at scale, the honest persuader gains a strange advantage: people are learning to discount polish and search for stake. Show yours — what you'd lose by being wrong is the most persuasive disclosure there is.

Practice. Take a real case you need to make. Write it three times: pure logic, pure credibility-and-stake, pure why-it-matters. Then braid the strongest of each into one version. Deliver it, and note which strand the audience responded to.

A growing share of what you write will be read first — sometimes only — by AI: the agent executing your spec, the assistant summarizing your memo for an exec, the search system deciding whether your documentation answers a question. Writing for machines is writing for the most literal reader you will ever have: state assumptions explicitly, define terms on first use, keep one instruction per sentence, and never rely on tone to carry meaning. Conveniently, this discipline improves the text for humans too.

The summary test is the new proofread: run your document through an AI summarizer and check whether your one point survives. If the machine mangles your meaning, some human skimmer already did.

Practice. Take an important document and ask AI to summarize it in three sentences. If the summary misses or distorts your point, revise the document — not the summary — until a fresh pass gets it right.

Distributed teams, compressed calendars, and AI teammates all point the same direction: the default medium of serious work is now the written, asynchronous artifact. The async-first discipline: decisions get a document, documents get deadlines for comment, comments get responses, and the outcome gets recorded where the next person — or the next agent — will look. Meetings still exist, but they are promoted to what they're good at: conflict, nuance, and trust. Status died with a document; let it rest.

The compounding benefit is institutional memory. A team that writes its decisions can onboard a human in days and brief an AI system in minutes. A team that decides verbally re-litigates everything forever.

Practice. Pick one decision currently bouncing through meetings and chat. Write the one-page decision doc, set a 48-hour comment window, decide, and record it. Track how often the topic resurfaces compared to the verbal baseline.

Attention is allocated by reputation. People decide to read your message because your last five were worth it — that is the entire algorithm. Earning the asset is slow and mechanical: send less, make every send dense with signal, never bury the ask, and never waste a reader's click on something a subject line could have carried. Spending it is fast: one rambling all-hands email, one cried-wolf "urgent," and your next ten messages are skimmed by reflex.

In the flood, this reputation becomes career infrastructure. The person whose messages are always worth opening has a communication channel money can't buy — and AI, which makes sending easy, makes the restraint that protects it rarer and more valuable.

Practice. Audit your last twenty sent messages as their recipient: which were worth opening? Set yourself two rules that would have filtered the rest (e.g., "no FYI without a because," "ask in the first line") and hold them for a month.

Teaching is the final exam of understanding — and a daily job requirement wearing other names: onboarding, documentation, the architecture walkthrough, the board explanation. The craft: start from what the learner already knows and build a bridge, one new idea at a time; use an analogy to create the shape and then immediately show where the analogy breaks; and check understanding by asking them to use the idea, not to nod at it. Curse-of-knowledge is the enemy — the expert's inability to remember what it was like not to know.

The people who explain well become disproportionately influential, because every person they teach becomes an advocate for the idea — and for them.

Practice. Take the most complex thing you understand at work and explain it to someone two domains away, in ten minutes, with one analogy. Have them explain it back. Every distortion in the echo is a flaw in the bridge — rebuild and re-test.

Most feedback fails in one of two ways: so cushioned the message never arrives, or so blunt the listener spends the conversation defending instead of hearing. The working formula is specific, behavioral, and prompt: what happened (observable, not interpreted), the impact it had, and what different would look like — delivered close to the event, in private, with the assumption that the person wants to be good at their job. "You're careless" is an attack; "the last two releases shipped without the checklist, and both had rollbacks" is information.

Receiving is the same skill reversed, and it is the rarer one: ask for the example, resist the rebuttal for one full day, and thank the giver even when it stings — because the alternative is a future where no one tells you anything.

Practice. Deliver one piece of real feedback this week in the what-impact-different form. And solicit one: ask a colleague "what's one thing I do that makes your work harder?" — then just say thank you and sit with it for a day.

A meeting is a communication system, and most are designed by accident. Running one well is a facilitation craft: a stated purpose ("we leave with a decision on X"), the right people and no others, pre-reading sent early enough to be read, drawing out the quiet expert before the loud generalist fills the air, naming disagreement instead of letting it fester politely, and ending with who-does-what- by-when said aloud. Thirty minutes run this way outperforms ninety run by inertia.

With AI handling notes, summaries, and action-item tracking, the human job in the room concentrates on what machines can't do: reading the silences, balancing the voices, and making the call when discussion has done its work.

Practice. Take one meeting you own and redesign it: written purpose, trimmed list, pre-read, a named decision moment, spoken action items. Let AI take the notes. Ask attendees afterward whether the difference was noticeable.