Four writers, coded.

The curriculum move.

• 22,000+ subscribers
• ~4 posts a month
• ~1,400 median words
• Ex-Whoop, Core Product and AI

She does not publish an insight and stop. She publishes a sequence with a completion state. In June 2025 a post called making AI a habit shipped a Google Sheet titled “30 Days of GPT”, day one through day thirty, ten minutes each. Ten months later that became couchto5k.ai, and the guest post she wrote about it exists mainly to say this:

That last clause is the whole model. The post is not the product. The post is the trailhead. A curriculum answers “what do I do tomorrow, and the day after”; an essay answers “what should I think.” Only one of those has a return visit built into the format, and a paid course to graduate people into.

The sentence-level mechanics

• Lowercase titles, always. She switched some time in late 2025: The Self-Promotion Horror Show in January 2025 became monkey grapes by May 2026. Lowercase reads as a text message, not as thought leadership.
• The title and subtitle do two different jobs. Title sea legs, subtitle your dreams have a cold start problem. Curiosity, then promise. Eight of her titles are odd-phrase constructions that mean nothing until you have read the piece.
• Long, short, shorter. A 30 to 45 word opening sentence, then a fragment. “Business slop!” “Bleak!” “Sorry Meg.” The fragment lands in the first four sentences and it is where the voice lives.
• “You” owns the problem, “I” owns the failure. She almost never writes “you should.” She writes “you can”, and bare imperatives: “Start with one problem.”
• Fixed sign-off: xoxo, hils.

She opens as the person who was wrong

• “Every time management system I’ve ever abandoned failed for the same reason: it was designed for the person I wanted to be, not the person I actually am.”how I run my life in claude code
• “I remember the first time someone recommended I use an AI copywriter. It was 2021, so pre-Chat GPT.”skills are cheap now. what’s your superpower?
• “There is a YouTube video I think about almost every day. Two Capuchin monkeys complete tasks in their cages.”monkey grapes

Always specific, always slightly embarrassing, never a humblebrag. Two sentences of that buys permission to be prescriptive for the next 1,500 words. It is the cheapest credibility mechanism in the entire dataset.

The credibility machine.

• 1,000,000+ subscribers
• 2 to 4 posts a month
• 10 to 30 hours a post
• Ex-Airbnb PM lead

The received wisdom is that every Lenny claim arrives with a benchmark attached. The counting says that is half right, and the correct half is more useful. Fifteen of his 25 posts carry ten or more distinct figures, the highest rate here. But the proof attaches at the section level, not the sentence level, and about a quarter of his output carries almost no external evidence at all. One 2026 post is a hand-drawn cartoon with zero numbers and zero citations.

What actually holds across all 25 is narrower and far more portable:

• “This data is courtesy of Pave, which gets its data by integrating with human resources information system software and equity management systems at over 7,500 top companies. The data below is based on actual comp from over 23,000 product managers globally.”
• “We took it upon ourselves to find out what’s actually happening on the ground by running one of the largest independent, in-depth surveys on how AI is affecting productivity for tech workers (1,750 respondents).”
• “Palle spent over 100 hours digging through all available public data.”

He writes the methodology sentence before the finding. That is a formatting decision, not a data-access decision, and it costs nothing.

Four proof reservoirs, three of them free

1. Licensed data partners. Pave, TrueUp. Bought with audience size. Not available to you.
2. His own survey panel. Repeated annually so that edition two can sell the delta: “A year ago, we ran our first large-scale survey.” Available at smaller n.
3. Crowdsourced reader submissions. His two highest-engagement self-authored posts of 2025, on vibe coding and on Claude Code, are literally “I asked on X, LinkedIn, and my subscriber Slack” plus curation. Each entry is a name, a one-line quote and a live link. This works identically at 500 followers and at 500,000.
4. Guest writers. Eleven of the 25 posts I coded are guest-written. He gives distribution plus ten hours of editing across five or six iterations; the guest gives proprietary evidence he could never generate himself.

When he cannot source something, he says so

The AI glossary is “drawn from my own understanding, a bunch of research, and feedback from my most AI-pilled friends.” The 500,000 milestone post says “which I’m told makes it among the top five largest.” And when there is genuinely no data, he changes format entirely rather than faking rigour.

The short-form discipline.

• 89,000+ subscribers
• 200,000+ on LinkedIn
• 1 post a week
• 3 to 5 hours a post

She works both forms, which is why she is worth studying. Going from 1,800 words to forty, the first thing she cuts is evidence. What survives is the verdict and the enemy.

Her churn benchmarks post is 1,894 words. The short version is: “None of the NRR and monthly churn rate high level bullsh1t.” That is the whole argument compressed into a refusal. The second thing to go is objection-handling. Long form gives her room for a subheading literally titled “Wait, but Elena, YOU have a personal brand!?” Short form has no room to argue with itself, so she asserts and lets the comments do the rebuttal.

Position in line one, always

• “B2b referrals don’t work… or do they!?”Seven words containing a consensus, a reversal and a promise
• “Many B2B businesses are swimming in self-serve users, but when it comes to generating pipeline from them for Sales? Crickets.”
• “IC work. So hot right now.”Opening line of a 5,000-word report
• “Who just hit 70K subs!? ME. / That is all. / Carry on.”

She is a compulsive coiner of portable objects

Her own explanation of why:

The Four W’s of Product-Led Sales. The Product-Market Fit Treadmill. Mom-and-Pop SaaS. The Minimum Lovable Product Era. Five Laws of Growth. The essay builds the object; the short post ships it. A noun someone can repeat in a meeting without citing you is worth more than a good paragraph.

Credentials go inside a confession, never in a bio

• “I’ve been in growth for over a decade, long enough to collect some scars.”
• “A year ago, I would’ve laughed at the idea of running growth at a $130M ARR company without touching activation.”
• “Let’s get one thing straight: I’m not a writer.”

And on why she owns the list rather than renting the audience:

The data storytelling manual.

• 4 to 5 charts a week
• 25 minutes a post
• ~220 words a caption
• 50,000+ companies in the dataset

He has documented his own method in more detail than the other three combined, so this section is mostly his words.

1. The single highest-leverage rule

The chart title states the conclusion, never the axes. Not “Time to first search by cohort” but “New recruiters are hitting their first search eleven days faster than last year.” It is a free edit and it is the difference between a chart that sits there and a chart that travels.

2. Start from a question, not from the data

And his definition of the word everyone overuses: “An insight is the sort of aggregated result of those numbers that tells you to take an action or not take an action.” If the chart does not change what someone does on Monday, it is trivia.

3. One chart, one point

Twelve of the thirteen posts I transcribed carry exactly one graphic. Three points is three weeks of content, not one crowded post. And for anyone starting: “Start with one chart that looks good and put it out on social. Stop trying to build a big report up front.”

4. Argue against your own number

This is the move that builds trust, and it is counterintuitive. He publishes a “things to remember” block that undercuts his own benchmarks: “Medians are not some special number. They simply reflect the middle of a wide range of rounds. Guideline only, each deal is different.”

He also publishes findings against Carta’s commercial interest and does not soften them, writing that venture pricing is “unfair and not truly data-driven” and that the twenty percent dilution norm “has very little to do with the capital needs of the business.” And he writes the counter-read himself:

Every chart post gets a “what this doesn’t tell you” block. It converts skeptics into commenters, and comments are where the next 200 ideas come from.

5. Sample size and exclusions, in the second sentence

“The data, from across more than 45,000 US startups that use Carta, is startling.” “Software companies only included, all rounds raised in the last 6 months, no bridges, extensions, or weird stuff.” Exclusions stated as plainly as inclusions. When the data is thin: “Be open and honest about how much data you’re talking about. Use phrases like ‘our data suggests’ to soften it a little bit.”

6. The caption structure

85 to 340 words, median around 220, fact and opinion never data alone. The finding in one line, then scope and method, then a bolded subhead block of terse bullets, then his opinion explicitly marked (“IMO”, “My bet is”, “Personally I’d like to see”), then a punchline or an open question. He separates fact from opinion typographically, so a reader can reject one without rejecting the other.

7. And if you have no proprietary data at all

He proved it himself. His post on S&P 500 concentration is entirely public market data and is one of his cleanest. His four routes: audit your own organisation for a latent dataset, because “very often, any company of a significant size will have a data asset that’s worth exploring”; curate public reports and add your own read; deliberately pick a thin-data niche, because less data means less competition; or survey your own audience, where 150 responses is defensible provided you disclose the n.