Essay · November 15, 2026 · 4,200 words · about 19 minutes

Post It When It's Done

The best time to post is a small real effect and a big oversell

This is a test of something I believed going in: that "the best time to post" is a running lie. I went looking for where the advice comes from, what the platforms and the researchers say, and what my own numbers show, and I say how sure I can be of each piece. By the end I take the word "lie" back. I am not a data scientist or a marketer. Vendor recommendations carry their page dates and will change, which is part of the point.

Words worth knowing: 11 plain-English explanations

The words this essay leans on, each explained twice: first in plain English, then in one line for the expert. In the essay, a word with a dotted underline opens the same explanation.

Impression
One time a post was shown on someone's screen.
For the expert a platform-reported display count; it does not mean the post was read, and definitions differ by platform.
Comes up in: How big is it next to the work
Engagement
People doing something with a post: reacting, commenting, sharing.
For the expert a sum of interaction counts, sometimes divided by impressions or followers to give a rate; definitions vary, and not every vendor page gives one.
Comes up in: Where the best time comes from
Feed ranking
How a platform decides which posts you see first.
For the expert a scoring model that orders candidate posts by predicted probability of actions such as clicks, comments or dwell, sometimes with post age as one input.
Comes up in: What the platforms say
Recency
How new a post is.
For the expert post age used as a feature or filter in candidate selection or ranking.
Comes up in: What the platforms say
Boosted post
A post its author paid the platform to show to more people.
For the expert paid distribution of an existing post; its impressions mix paid and unpaid delivery.
Comes up in: The strongest case against me
Confounding
When two things look linked because a third thing drives both.
For the expert a variable associated with both treatment and outcome, here who posts and what they post at a given hour, which biases an observational estimate of the hour's effect.
Comes up in: What these studies cannot show
Effect size
How big a difference is, as opposed to whether there is one.
For the expert the magnitude of an estimated effect, ideally with an interval, as distinct from its statistical significance.
Comes up in: What these studies cannot show
Regression to the mean
An unusually good or bad result is usually followed by a more ordinary one.
For the expert extreme observations of a noisy measure tend to be nearer the average when measured again, so the top slot in one sample will usually look less special in the next.
Comes up in: What these studies cannot show
Cumulative advantage
Getting attention early makes it easier to get more.
For the expert a rich-get-richer process in which early success raises the rate of later success, independent of quality.
Comes up in: The strongest case against me
Half-life
The time it takes for something to fall to half.
For the expert here, the time for the pull of a new item's novelty to halve.
Comes up in: The strongest case against me
Simultaneity
Two things happening at the same time.
For the expert in special relativity, a relation defined only relative to a frame of reference.
Comes up in: Time might not even be real

I work alone and I ship a lot. When a thing is done I post it on LinkedIn, and that might be 8 in the morning or 4 in the morning, because that is when it got done. I do not hide that I work late.

Plenty of guides to social media say this is wrong. There is a best time, they say, and 4 a.m. is usually not it. My view going in was simple: the optimal time to post is a running lie, do things when you want, and the people who want to see them will. I also said time might not even be real, which I will get to.

A belief I like is exactly the kind I should check. So this essay tries to break mine.

Where the best time comes from

The best-time pages I found come mostly from companies that sell tools for scheduling posts. They analyze large sets of posts, in some cases posts their own users sent through the tool, and publish the hours that did best, in editions titled by year. Here is what the current pages say for LinkedIn, as read on November 15, 2026.

Who (page date)Best time to post on LinkedInBased on, as stated
Sprout Social (March 2026)Tuesday, 11 a.m. to 5 p.m.nearly 2 billion on 307,000 profiles, six networks together
Buffer (September 2026)3 to 8 p.m. on weekdays, top slot Wednesday 4 p.m.; mornings lower; Monday and Tuesday lowestover 4.8 million posts sent through Buffer
Hootsuite (November 2025)Tuesday and Wednesday, 8 to 9 a.m. in the page's key takeaways and FAQ; Tuesday and Wednesday, 4 to 6 a.m. in the body of the same pageover 1 million posts, all networks together
Later (May 2026)Tuesday to Thursday, 8 to 10 a.m. and 5 to 6 p.m.no LinkedIn data set named
SocialPilot (November 2025)Tuesday to Thursday, 10 a.m. to noon and 1 to 4 p.m., Eastern timeover 683,000 posts
CoSchedule (updated September 2026)top five hours: 10 a.m., 11 a.m., noon, 3 p.m., 2 p.m.; worst five include 3 and 4 a.m.its own research from mid-2022, size not given

Hootsuite's LinkedIn page gives both windows, and its all-networks page says "between 4 AM and 6 AM on Tuesdays and Wednesdays." I cannot tell which is meant, so I print both.

Read across the rows. Tuesday is Sprout Social's best day and one of Buffer's two lowest. Four in the morning is inside the window Hootsuite's article body calls best and on CoSchedule's list of the five worst hours. Buffer says mornings now see lower engagement than afternoons and evenings; four others include a morning window among their best. Sprout Social says weekends "are the worst days to post." Buffer's roundup says weekend posts get "noticeably more engagement than the early weekdays," though its LinkedIn page puts both weekend days below most weekdays. The clocks differ (local time for Sprout Social and Buffer, Eastern for SocialPilot, none named by Later). Still, add the windows and, on one weekday or another, they cover 4 to 6 a.m. and every hour from 8 a.m. to 8 p.m. If every one of these is right, the best time to post on a weekday is most of the day.

The answers also move from year to year. Hypebot reported Hootsuite's 2023 answer for LinkedIn as 1 p.m. Pacific on Mondays, from a different analysis of "thousands of posts." A National Law Review article from May 2024 gave Sprout Social's answer for LinkedIn company pages as Tuesdays and Wednesdays from 10 a.m. to noon and Thursdays at 10; the 2026 page says Tuesday from 11 to 5. Both earlier figures are secondhand; the companies' current pages carry newer editions. Buffer reports its own shift: in 2025 its peak times "sat firmly within the workday," and this year they moved later. "Our data can't tell us why the change is happening," its LinkedIn page says. HubSpot's best-times page says it was first published in April 2014.

Two more facts. First, every page in the table promotes its publisher's own product, and most promote a feature that schedules posts or picks the hour for you. That is an incentive. It is not evidence that anyone's numbers are wrong. Second, most of these pages also say the numbers are a starting point and your own audience decides. Buffer says it plainly: "sharing your social media posts at the 'right' time is not the secret sauce for content success," and posting regularly "will have a bigger impact." The headlines say there is a best time. The fine print is closer to my view.

There is real agreement too. Most of the pages name Tuesday to Thursday, and most rate weekends low. I cannot tell how independent the data sets are, but a pattern that repeats across several probably says something about when people who use LinkedIn for work look at it. That is my inference.

What these studies cannot show

None of these pages describes an experiment. As far as they say, nobody assigned posts to hours at random. The posts that went out at 4 p.m. on a Wednesday were written by different people, about different things, than the posts that went out at 6 a.m. on a Sunday. That is , and it can make an hour look good when the credit belongs to whoever tends to post then. Some pages adjust for time zone and some list hours by industry. None describes holding the author or the content constant.

Where a page says whose posts it studied, they are the vendor's own users: Buffer's were "sent through Buffer," and Sprout Social calls its numbers "first-party data." The others do not say. I assume users of scheduling tools are mostly brands and people whose job is social media. An average over them is not a measurement of me, or of you.

None of the six pages gives an . They rank the hours. They do not say whether the best hour beats an ordinary one by 2 percent or by 200, and they give no margin of error. A ranking with no sizes cannot tell you whether it is worth rearranging your day.

And with 168 hours in a week, one has to come out on top in any sample, even if the hours were nearly the same. says the winner of one noisy sample will usually look more ordinary in the next. With millions of posts, noise is probably not the whole story; the users sampled, the way engagement is counted and the platforms themselves all change. Why the answers drift is my inference. Buffer says its data cannot tell it why its own moved, and I found no explanation on the other pages.

What the platforms say

What the platforms publish does not put the clock at the center.

LinkedIn's engineers described their in a blog post filed under 2020: tens of thousands of candidate posts, each scored by the predicted chance that this particular member will act on it. Its only mention of the clock that I found is "time of day," among the "contextual features" of one sub-model, and it does not say whether that is the hour the member is looking or the hour the post went up. In 2024, Social Media Today paraphrased LinkedIn's editor in chief, Dan Roth, as telling a podcast that members should not chase trends like best posting times. In 2025, when members noticed old posts in their feeds, Mediaweek quoted a LinkedIn spokesperson: "We may sometimes share posts that are older and still relevant to a member's career." A LinkedIn communications lead called the episode "part of some testing" and "only temporary." So by LinkedIn's own account a post can be shown well after the hour it went up, and still counts.

Instagram is the clearest that time is an input. Its head, Adam Mosseri, listed "when it was posted" among the signals for Feed in 2023, and called it "mundane" next to how popular a post is. TikTok's public explanation of its recommendations, from 2020, does not list posting time among its factors. YouTube's Todd Beaupre, as Search Engine Land quoted him: "If the audience doesn't really care whether you uploaded it at three in the morning or seven at night, then the algorithm shouldn't care either." The same report says YouTube's liaison, Rene Ritchie, wrote that an off-hours upload may start slower but total views usually even out over a few weeks or a month.

In these pages I found no platform that says flatly that posting time does not matter, and none that names a best hour.

The strongest case against me

Timing has measurable effects. Here is the best evidence I found.

People keep daily rhythms. A study of 509 million tweets from 2.4 million people in 84 countries, as AAAS summarized it, found mood rising and falling through the day in step with sleep. It measured mood, not who is online when, but the plain part needs no study: if the people who would want my post are asleep, they are asleep.

Attention fades fast. A 2007 paper on the news site Digg, which then had a million users, tracked 1,110 stories minute by minute and put the of a story's novelty at 69 minutes. That was one site, nearly twenty years ago. Where something like it holds, the first hour is a large part of the game.

Early reactions compound. In a randomized experiment on a social news site, a single upvote given at random to a new comment "increased the likelihood of positive ratings by 32%" and raised final ratings by 25 percent on average. This is , and I take it to be the mechanism behind the timing advice: an awake audience gives a post its early push. The experiment tested early votes, not hours.

The best controlled evidence is from Reddit. Researchers collected 132,307 submissions of 16,736 images, so the same picture was posted about eight times, which holds the content still. They found that "submissions posted during certain hours of the day are more popular." The peak did not come when the site was most active, and the paper gives no separate size for the hour.

A Journal of Marketing study of 5,700 Facebook posts by one West Coast newspaper found that the hour mattered differently for different content: readers engaged more in the morning with negative, high-arousal posts, and more in the afternoon with and demanding ones. The university's press release says ad-supported sites can raise digital ad profit "by at least 8 percent" by scheduling around their audience's sleep. I read the release, not the paper, and it does not say how that figure was reached.

Then the points about me. I have not measured where my audience lives, but I would guess most of it is within a few time zones of me. If so, a 4 a.m. post waits hours for its readers, and I cannot show that LinkedIn makes up the difference later. And "lie" means someone knows better and says it anyway. These companies say where their numbers come from, and most print the caveats. What I can show is advice that conflicts from page to page, comes without sizes, and runs under headlines more confident than its own fine print. I would call that overselling. I have no evidence that anyone is lying, and I was wrong to use the word.

How big is it next to the work

The question that matters is size. No source I opened measures the hour against the content head to head. The nearest things:

On Twitter, researchers with data on users, content and past performance found their best models "can explain less than half of the variance" in how far a post spread. More than half of the variation was left unexplained by what they could measure, and their simulations suggest more data would not close the gap. The abstract does not say whether the posting hour was among the things measured.

In a well-known music experiment, 14,341 people downloaded unknown songs, some seeing what others had chosen. Seeing others' choices made the outcome less predictable, but "the best songs rarely did poorly, and the worst rarely did well." As I read it, quality set the range and chance picked the spot inside it. The experiment was about social influence, not the clock.

Buffer's own pages make a useful comparison. Its timing page gives no sizes. Its frequency study of more than 2 million posts, which compared each account's busy weeks with its own quiet weeks, reports that weeks with 2 to 5 posts went with about 1,200 more per post than weeks with one, and weeks with 11 or more went with about 17,000 more. This is vendor data with no error bars, and busy weeks may differ from quiet ones in other ways. But the same company puts a number on how often and none on when.

And the same link can take more than one try. On Reddit, as Nieman Lab reported one study, 52 percent of links that reached the front page "had been shared at least once before" by a different user. Same link, different attempt, different result. Timing may be part of that, and so may luck; the report does not separate them.

My own record

Here are my last 20 LinkedIn posts, read from my own analytics on Sunday, November 15, 2026, at about 3:55 p.m. Eastern.

Posted (Eastern)ImpressionsBoostAge when read
Sun Nov 15, 3:12 PM1,580boosted42 minutes
Sun Nov 15, 11:53 AM1,156boosted4 hours
Sat Nov 14, 10:22 PM1,620boosted17 hours
Sat Nov 14, 3:52 PM14,405boosted1 day
Sat Nov 14, 3:14 PM8,224boosted1 day
Sat Nov 14, 12:50 PM7,261boosted1 day
Sat Nov 14, 12:20 PM4,941boosted1 day
Fri Nov 13, 10:52 PM4,739boosted1 day
Fri Nov 13, 9:02 PM7,451boosted1 day
Fri Nov 13, 2:35 PM4,175boosted2 days
Fri Nov 13, 11:16 AM12,045boosted2 days
Thu Nov 12, 8:37 PM15,278boosted2 days
Thu Nov 12, 6:05 PM6,351boosted2 days
Thu Nov 12, 11:17 AM21,331boosted3 days
Thu Nov 12, 2:10 AM356boost ended3 days
Wed Nov 11, 5:04 PM17,359boosted3 days
Wed Nov 11, 5:03 PM14,994boosted3 days
Wed Nov 11, 5:08 AM6,809boosted4 days
Wed Nov 11, 4:42 AM11,912boosted4 days
Tue Nov 10, 12:15 PM8,185boost ended5 days

Now the problem, stated plainly. Every one of these was a boosted post. I paid LinkedIn to show each of them to more people. So these impressions are not what the posts earned on their own, and I do not know how much of each number was paid for. The table cannot test posting time. It does not support my claim and it does not refute it.

What the table can show is limited. I do post at odd hours: the earliest here is 2:10 a.m., the latest 10:52 p.m., and seven of the twenty went out on a weekend. The worst number by far, 356, belongs to the 2:10 a.m. post. That fits the case against me. It is also one of two posts whose boost had ended, so the table cannot say which hurt it. Two posts went out before dawn on Wednesday. The 4:42 a.m. one has 11,912, above the 8,185 median of the 17 posts more than a day old, and the 5:08 a.m. one has 6,809, below it. The three newest posts were under a day old, so their counts are unfinished.

Then the pairs. Two posts went out one minute apart on Wednesday afternoon and stood 16 percent apart when I read them. Two went out 38 minutes apart on Saturday and stood 75 percent apart. Two went out at 11:16 and 11:17 a.m. on different days and stood 77 percent apart, though the larger had a day longer to collect. That is what I would expect if the post matters more than the clock. It is equally what I would expect if the boosts differed, and I picked these pairs after seeing the numbers. So they show only that the hour alone did not set the number, not what did.

What waiting costs

I found no study of people holding finished work for a better hour. The nearest evidence is on perfectionism. A meta-analysis found that worry about doing things wrong goes with more procrastination (a correlation of 0.23 across about 10,000 people), while high standards alone go with less (minus 0.22). Modest numbers, about procrastination in general, not posting.

The rest is my inference. If I finish at 4 a.m. Saturday and hold the post for Sprout Social's Tuesday at 11, I wait 79 hours for a gain nobody has put a size on. For a company with a launch date, a team and one shot, the sum is different.

Time might not even be real

About that. Special relativity says there is no universal "now": observers moving relative to each other disagree about which distant events are simultaneous, and no experiment can pick a winner. Whether that makes time itself unreal is an open argument among philosophers of physics, not a result. Either way, only gets strange at speeds and distances far from mine. My readers and I share one planet, so Einstein does not excuse a 4 a.m. post. Clocks are real enough to schedule by. I just do not think the schedule matters much.

What the evidence supports

Timing is a real factor. People sleep, attention fades fast, and early reactions compound. The two sizes I can cite belong to their own settings, and neither is the size of an hour: a claimed 8 percent or more on ad profit, from one newspaper's data, and a 25 percent lift in final ratings from one early upvote. No study I found compares the hour with the content directly, so "timing is small next to the work" is my reading and not a measured result. And in my opinion the advice is oversold: the published best times conflict with each other, shift from year to year and come without sizes. The pages come mostly from companies that sell scheduling tools, and most print the caveats themselves.

So "running lie" was too strong. "Running oversell" is the most I think the sources support. For one person who makes a lot of things, I think waiting for the best hour costs more than it gains. That is my judgment. No study I found tests it, and my boosted numbers cannot settle it.

What I do

This is my practice, not advice.

  • I post when the work is done.
  • I do not hold a finished thing for a recommended hour.
  • I do not hide the timestamp. If it says 4 a.m., I was up.
  • I put the effort into the next thing instead of the schedule.
  • I look at my numbers, knowing my boosted ones cannot answer this question. A proper test needs unboosted posts and a lot more of them.
  • If I ever have something that truly is time-sensitive, I will post it on time.

Sources and a note on method

Vendor pages, opened November 15, 2026: Sprout Social, https://sproutsocial.com/insights/best-times-to-post-on-linkedin/ . Buffer, https://buffer.com/resources/best-time-to-post-on-linkedin/ , https://buffer.com/resources/best-time-to-post-social-media/ and https://buffer.com/resources/how-often-to-post-on-linkedin/ . Hootsuite, https://blog.hootsuite.com/best-time-to-post-on-linkedin/ and https://blog.hootsuite.com/best-time-to-post-on-social-media/ . Later, https://later.com/blog/best-time-to-post-on-social-media/ . SocialPilot, https://www.socialpilot.co/blog/best-time-to-post-on-linkedin . CoSchedule, https://coschedule.com/blog/best-times-to-post-on-linkedin . HubSpot, https://blog.hubspot.com/marketing/best-times-post-pin-tweet-social-media-infographic . Earlier years as reported by Hypebot, March 28, 2023, https://www.hypebot.com/hootsuite-shares-the-best-times-to-post-on-social-media-in-2023 , and the National Law Review, May 29, 2024, https://natlawreview.com/article/maximizing-social-media-engagement-finding-optimal-times-post-2024 .

Platforms: LinkedIn Engineering, "Understanding dwell time to improve LinkedIn feed ranking" (the page shows no date; 2020 is from its web address), https://engineering.linkedin.com/blog/2020/understanding-feed-dwell-time . Social Media Today, February 27, 2024, https://www.socialmediatoday.com/news/linkedin-shares-insights-into-latest-feed-algorithm-updates/708710/ . Mediaweek, July 17, 2025, https://mediaweek.vinyl.media/why-linkedin-users-are-seeing-old-posts-in-their-feed . Adam Mosseri, "Instagram Ranking Explained," May 31, 2023, https://about.instagram.com/blog/announcements/instagram-ranking-explained . TikTok, "How TikTok recommends videos #ForYou," June 18, 2020, https://newsroom.tiktok.com/en-us/how-tiktok-recommends-videos-for-you . Search Engine Land on YouTube (the page shows no date), https://searchengineland.com/youtube-issues-guidance-on-when-to-post-content-for-maximum-reach-438774 .

Research: Golder and Macy, Science (2011), as summarized by AAAS, https://www.aaas.org/news/science-tweets-reveal-role-circadian-rhythms-mood . Wu and Huberman, "Novelty and collective attention" (2007), https://arxiv.org/abs/0704.1158 . Muchnik, Aral and Taylor, "Social Influence Bias: A Randomized Experiment," Science (2013), https://pubmed.ncbi.nlm.nih.gov/23929980/ . Lakkaraju, McAuley and Leskovec, "What's in a Name?" (2013), https://ai.stanford.edu/~jure/pubs/reddit-icwsm13.pdf . Kanuri, Chen and Sridhar, Journal of Marketing (2018), per the University of Notre Dame release, https://mendoza.nd.edu/news/capitalizing-on-sleep-wake-cycle-can-drastically-increase-digital-ad-profits-from-social-media-study-shows/ . Martin, Hofman, Sharma, Anderson and Watts (2016), https://www.arxiv.org/abs/1602.01013 . Salganik, Dodds and Watts, Science (2006), https://collaborate.princeton.edu/en/publications/experimental-study-of-inequality-and-unpredictability-in-an-artif/ . Gilbert on Reddit, as reported by Nieman Lab, March 27, 2013, https://www.niemanlab.org/2013/03/on-reddit-good-links-rise-to-the-top-but-it-can-sometimes-take-a-few-attempts-to-get-there/ . Sirois, Molnar and Hirsch, European Journal of Personality (2017), https://eprints.whiterose.ac.uk/id/eprint/112533/ . Stanford Encyclopedia of Philosophy, "Being and Becoming in Modern Physics," https://plato.stanford.edu/entries/spacetime-bebecome/ .

I used AI to research and draft this, under my direction. The draft went through two more passes, also with AI. The first read it as a statistician, a social media researcher, a vendor's lawyer and a tired reader, and softened about thirty sentences that said more than their sources. The second, on November 15, 2026, reopened every source listed above and checked each quotation, number, date and name against it. All of them opened. That pass changed what the reopened pages did not support: a Shopify citation was dropped because it gave a single day's peak and not a yearly answer, the Sprout Social 2024 figure now says it was for company pages, the newspaper study now says what the release says, and my reading of LinkedIn's "time of day" was removed because the post does not say which time it means. Hootsuite's two LinkedIn windows were confirmed: both are on its page, and I print both.

The limits. The pages were read through a tool that fetches a page and returns excerpts, not by eye in a browser, and a direct download to search the text was blocked again, so a long quotation could differ from the page in punctuation. Where a page writes a range of hours with a dash, I write it out with "to". Where I say reported, summarized, paraphrased or quoted by, I read a summary and not the original: that covers the Golder and Macy, Gilbert and Kanuri studies, LinkedIn's podcast remarks, YouTube's remarks and both earlier-year vendor figures, since the companies' pages now carry newer editions. For four papers (Muchnik, Martin, Salganik, Sirois) only the abstract was read. The full Digg paper was read this time, which is where the 69 minutes comes from. I did not read X's ranking code, so I say nothing about X. No statistician has read this essay.

Disclosure: I use LinkedIn and I pay to boost posts there; every post in my table was boosted. I have no relationship with any company named here.

Josh Solomon
Josh Solomon founded Known Quantity LLC in October 2026: an AI consulting practice for finance and revenue teams, the Gimble Gaming studio, three clothing labels (two with his wife, one his own men's line, Cordwell), and this essay desk. He uses AI to research and draft, under his direction, with every claim checked against its source. LinkedIn · Invite him to speak

Questions this essay answers

What is the best time to post on LinkedIn?

The vendor pages disagree. Read on November 15, 2026: Sprout Social says Tuesday, 11 a.m. to 5 p.m.; Buffer says 3 to 8 p.m. on weekdays, with Monday and Tuesday lowest; Hootsuite gives Tuesday and Wednesday, 8 to 9 a.m. in one part of its page and 4 to 6 a.m. in another; CoSchedule puts 3 and 4 a.m. among its worst five hours. Add the windows and, on one weekday or another, they cover 4 to 6 a.m. and every hour from 8 a.m. to 8 p.m.

Where do the best time to post figures come from?

Mostly from companies that sell tools for scheduling posts. They analyze large sets of posts, in some cases posts their own users sent through the tool, and publish the hours that did best. Every page in the essay's table promotes its publisher's own product. That is an incentive, not evidence that the numbers are wrong, and most of the pages also say your own audience decides.

Why can't these studies prove that an hour is best?

None of the pages describes an experiment: posts were not assigned to hours at random, so the posts at one hour were written by different people about different things than the posts at another. That is confounding. None of the six gives an effect size, so a ranking cannot say whether the best hour beats an ordinary one by 2 percent or by 200. And with 168 hours in a week, one has to come out on top in any sample.

Does posting time matter at all?

Yes, timing is a real factor. People sleep, attention fades fast (a 2007 study of Digg put the half-life of a story's novelty at 69 minutes), and early reactions compound (one random early upvote raised final ratings by 25 percent on average in a randomized experiment). No study the essay found compares the hour with the content directly, so "timing is small next to the work" is the author's reading, not a measured result.

What do LinkedIn and the other platforms say about posting time?

LinkedIn's engineering post scores each candidate post by the predicted chance that a particular member will act on it, and mentions "time of day" only among the features of one sub-model. Instagram lists "when it was posted" among its Feed signals. TikTok's 2020 explanation does not list posting time. The essay found no platform that says posting time does not matter, and none that names a best hour.

Is the best time to post a lie?

The author went in calling it a running lie and takes the word back. The companies say where their numbers come from and most print the caveats. What the sources support is overselling: advice that conflicts from page to page, shifts from year to year and comes without sizes. For one person who makes a lot of things, the author's judgment is that waiting for the best hour costs more than it gains; no study tests that.

More in this series

Essay 17 · November 2026

Raise the Bar, Open the Gate

Real competition and broad access can work together, and giving good things away is a way to compete

Read the essay

Essay 15 · November 2026

Whose Name Is on the Share

What you have when a stock sits in your brokerage account, and what changes if you register it in your own name

Read the essay

Essay 14 · November 2026

Two Months' Salary

How the price of a diamond gets made, and what to ask before you pay

Read the essay

Essay 13 · November 2026

The Worst Investor You Know Is Ordinary

I have traded stocks and options for years and lost money at it more than once. The studies say my kind of loss is common. This is what the record says about people like me, why we are slow to learn, and the test I would use before trusting my own trades.

Read the essay

Essay 12 · November 2026

A Colleague, Not a Notebook

About four weeks of working alone with an AI on a lot at once: what I built around it, how it broke, and what I would tell someone starting tomorrow. Most of my failures had the same cause: I believed a report without looking.

Read the essay

Essay 11 · November 2026

The Arrow Points the Other Way

People ask what happens when AI gets a quantum computer. The evidence points the other way: AI is already running the quantum machines, and what quantum gives back is a new kind of data, not a faster brain.

Read the essay

Essay 10 · November 2026

Happiness Is the Fuel

Slavoj Žižek calls happiness an unethical category and love, in a formal sense, evil. He is right that happiness is a bad thing to chase. He is wrong about what it does once it arrives.

Read the essay

Essay 9 · November 2026

Nobody Had to Cheat

On January 28, 2021, customers of Robinhood and several other brokers could sell GameStop but not buy it. The record shows no conspiracy. It shows a market where nobody has to break a rule for the small investor to be shut out.

Read the essay

Essay 8 · November 2026

The End of Logistics

Most of what wears people down is not the work. It is the remembering, tracking, chasing and checking around it. Hand that to a machine and keep the deciding, and what you get is not more output. It is calm.

Read the essay

Essay 7 · November 2026

Every Road Ends at the Same Sea

Why a perfect God could still have a real choice. If every road God would choose ends at the same sea, no road is better than another, and the roads are nothing alike.

Read the essay

Essay 6 · November 2026

Why a Smart Machine Would Keep Us

The Matrix had it backwards. A superintelligent machine would not need us for power or for bodies. It would need us for a reason to go anywhere, and whether it keeps us is being decided now.

Read the essay

Essay 5 · November 2026

Personal Bubble Universe

Every person lives inside a world their own perception builds. Closeness is how much two of those worlds overlap, and how much you appreciate the part that does not.

Read the essay

Essay 4 · November 2026

Impatiently Waiting

Every real project stalls in someone else's queue. The case for staying impatient and pointing the impatience at a second, bounded piece of work.

Read the essay

Essay 3 · November 2026

Failing Confidently

Other people's forecasts of you are usually accurate about your past and wrong about your ceiling, and failure gives them more evidence every time.

Read the essay

Essay 2 · November 2026

The Utility of Not Caring

Caring is not free. It is a tax on judgment, and the tax is highest exactly where we most want to be right.

Read the essay

Essay 1 · November 2026

The Ego Death of Society

Why the hardest part of the AI transition isn't the technology. It's letting go of being the smart one.

Read the essay

Have Josh on your podcast or stage

The one-person-company case study, the evidence, and the honest objections. Everything in writing first.