# When Content Becomes Free: Odd Lots on How AI Is Upending Social Media Work

Source: Realpha Blog (blog.getrealpha.com)
Original article and charts: https://blog.getrealpha.com/en/blog/oddlots-2026-09-24-ai-is-upending-the-lives-of-people-who-do-social-m/

> From the September 24, 2026 Odd Lots live show in Los Angeles: Taylor Lorenz and Rachel Karten on what AI actually did to social and marketing work — bosses outsourcing judgment to models, audiences who say they hate AI content but watch it anyway, and how-to creators losing ground. Listening notes and reflections; educational, not investment advice.

Published: 2026-09-24
Locale: en
Tags: AI, creator economy, social media, brand marketing, Odd Lots


![A restaurant cook wearing smart glasses works the stove, the camera reaching past him into an empty dining room](/covers/oddlots-2026-09-24-ai-is-upending-the-lives-of-people-who-do-social-m-cover.png)

> Once Zhuang Zhou dreamt he was a butterfly, a butterfly flitting about, happy with himself and doing as he pleased. He did not know he was Zhou. Suddenly he woke, and there he was, solid and unmistakably Zhou. He did not know whether Zhou had dreamt he was a butterfly, or a butterfly was dreaming he was Zhou.
>
> —— Zhuangzi, "Discussion on Making All Things Equal" (Warring States period; translation mine)

On September 24, 2026, at the first-ever live Odd Lots show in Los Angeles, hosts Tracy Alloway and Joe Weisenthal sat down with Taylor Lorenz and Rachel Karten to talk about what AI has done to people who do social media for a living. Karten had interviewed roughly a hundred marketers; the recurring scene was a boss dropping a subordinate's meme into Claude and asking whether it was funny, and she quoted one of them saying it was demoralizing as hell. The other number came from Instagram: of the twenty most-viewed Reels of all time, two or three are AI-generated. The episode is about what happens to people who live off content once making it costs almost nothing — drawn from the US social and marketing industry, so anyone in a different trade has to re-check it against their own.

## What the Episode Is About

Odd Lots usually covers bond markets, supply chains, commodities. This time they were in Los Angeles, so the topic became the influencer economy. Both guests work inside it: Taylor Lorenz writes Usermag, a newsletter on tech and internet culture; Rachel Karten writes Link in Bio, read by social teams at companies, which she describes as the trade publication for people who do this job.

Joe's opening question was blunt: what happens to creators when making content becomes free? I came away thinking the answer here is messier than the "AI replaces humans" headline — what's getting knocked out isn't what most people assume, and the part that hurts practitioners most has little to do with the technology.

## The Main Points

**The boss with AI brain is the industry's sorest spot right now.** Karten wrote a piece called "Does Your Boss Have AI Brain?" based on conversations with about a hundred marketers. The pattern: the boss demands AI, the boss uses AI, and the boss hands his own gut instinct over to the model. She heard about a social manager who worked hard on a funny meme, sent it up, and watched it get run through Claude with the question "is this funny." Another marketer found that his boss only approved ideas that came out of Copilot, so he started mocking his own ideas up inside the Copilot interface to get them signed off. Her takeaway wasn't to ban AI from marketing teams. It was that the rollout has no shape to it — back at Condé Nast, onboarding a project tool like Asana took months of training, while AI arrived with nobody teaching anything and no guardrails.

**Consumers read your social feed as a proxy for how much care goes into the product.** Karten's point is that people expect a brand to put creative effort in, and the effort visible on social gets read as the effort behind the product. See your favorite chain posting AI imagery and the next thought is: what other shortcuts are they taking?

**What people resent is being tricked, not AI itself.** Lorenz noted that YouTubers are reading AI-written scripts and the comments still say "amazing analysis, great script." The same person can pile onto one video for using AI and then, one scroll later, say they love the story about an orange falling in love with a carrot — that's Fruit Love Island, AI-generated. Karten's read: a meme is low stakes and nobody cares about its craft, but people care about the craft of a company they buy from.

**The category actually getting knocked out is how-to.** Lorenz pointed out that media got democratized through how-to content — the earliest YouTube videos were makeup tutorials and fix-this-yourself clips. AI is a better authority for exactly that; asking a model about your broken dishwasher beats hunting for a video. Joe added that his mother proudly told him she'd used ChatGPT to fix her toilet. Karten's reply stuck with me most: in the old world she'd have called a plumber, and that was one real interaction with another person.

![Two panels side by side: on the left, a question travels through a row of video thumbnails before reaching the answer, a long path; on the right, the same question reaches the answer in one step, a short path.](/figures/howto-path-collapse-en.svg)

**Monetizing confusion is a short-term win and a long-term mess.** Karten looked up the twenty most-viewed Reels ever; two or three are AI-generated. One was dressed up as behind-the-scenes footage, pretending to show how a wave gets filmed for a movie and what it would do to houses. She called it borrowing the proof-of-reality framing to confuse people, and said the short-term benefit comes with bad long-term results. An AI account can post 50 times a day, pulling the slot machine until something hits.

**The accounts that grow are running serialized shows.** Karten sees one common thread in what's growing now: familiarity, as if each account were producing its own TV program. Her example was a restaurant cook who wears Meta glasses and records himself cooking, one episode a day, up to roughly 500,000 followers in two or three months. Lorenz added: don't try to be a lifestyle influencer in 2026, nobody cares and there's a lot of anger toward those people. The episode also cited Instagram head Adam Mosseri, who has said the social part of social media now lives in DMs — sending a friend a meme that reminds you of them counts as connection more than posting an update about your life.

![Fifty small dots in a grid with only one lit up, beside a single carefully made dot on its own.](/figures/fifty-posts-one-hit-en.svg)

## Going Further

### "Something I made got vetoed by a person who never looked at it"

A lot of people will feel that one in the chest. You spend three hours on an idea; your manager spends thirty seconds pasting it into a chat box and comes back with "AI thinks it's mediocre." The painful part isn't whether the model judged well. It's that responsibility for the judgment evaporated — you can't argue with the model, and you can't argue with someone who has given his judgment away.

![A chain running left to right: your pitch goes to the manager, the manager feeds it to a model, the model returns one line of comment; below, a dotted path trying to run back is crossed out at both the manager and the model, so it never reaches the start.](/figures/judgment-outsourced-broken-loop-en.svg)

Listening to that, I thought about the same shape in investing. How many of my positions did I reason my way into? How many came from a KOL saying he liked it, a research note with a price target, a model that spat out a nice score? The difference stays invisible while things go well and shows up the day the stock halves. Reasoning you built yourself, you can check: here was my thesis, does it still hold, do I stay or go. Borrowed conviction leaves you with one move — go ask that person again, and he may not answer this time.

![Two overlapping lines on a timeline split after a dashed marker labeled the day it halves: the upper line rises, meaning the reason still stands; the lower line falls, meaning there is no one left to ask.](/figures/own-reasoning-vs-borrowed-en.svg)

That Condé Nast comparison is the most useful line in the episode. Months of training to roll out a task tracker, and nothing at all to roll out something that makes judgments for you. Applied to my own work: if AI helps with investing homework, I need to decide in advance which part I hand over and which part I keep. What I do is hand over gathering and translating, and keep "what does this number mean." Maybe that line is in the wrong place, but I drew it.

### They say they hate it and watch it anyway — which one do I believe?

The contradiction in the episode is funny: a comment section full of anger about AI imagery, and the same people forwarding an AI fruit soap opera to a friend. The resolution Lorenz and Karten offer is stakes — low-stakes entertainment doesn't get judged on craft, decisions involving money do.

![A line sloping from upper left to lower right, with the horizontal axis showing what the thing costs you and the vertical axis showing how much AI the audience tolerates; memes sit highest at the left and products you pay for sit lowest at the right.](/figures/ai-tolerance-by-stake-en.svg)

That layering transfers directly to reading market sentiment. Surveys, social volume, comment tone measure what people say. Fund flows, volume, positioning changes measure what people do. When the two diverge, neither is lying — they answer different questions. Words answer "who do I want to be," money answers "what am I afraid of and greedy for right now." So when a survey says eighty percent are bearish, the follow-up question is whether that eighty percent has moved its book.

![Two lines over time: the one labeled talk drops to eighty percent bearish, the one labeled money stays nearly flat, and the gap between them is marked as a divergence zone.](/figures/mouth-versus-money-en.svg)

Same with companies. The earnings call is the mouth; capex, inventory, receivables are the hands. If management says the theme this year is discipline while capex jumps forty percent, go with capex.

### Whatever can be copied is losing value

The hardest inference in the episode comes from Lorenz: how-to creators are going to struggle, because models are faster and better at telling you how to do the thing, tailored to your particular dishwasher. What survives is what isn't easily replicable — commentary with a point of view, a person whose presence is the product.

I read that as one sentence: the value of a skill depends on whether it can be copied, not on how many years it took you to acquire. That's a rough thing to hear about a career, and there's no dodging it. Gathering data, translating, writing boilerplate copy, producing fixed-format analysis — all in the copyable pile. What's left? The episode gives a concrete answer in that cook with the glasses shooting an episode a day: his content can't be copied because it's what he actually does every day.

![A row of bars rising from left to right, with easily copied skills short on the left and hard to copy ones tall on the right, the height being the value that remains.](/figures/replicable-value-decay-en.svg)

One restraint here. This is what two people inside social and marketing saw in September 2026, and the episode offers no data on how much how-to creator income has actually fallen. The argument assumes models keep getting cheap and good, and Lorenz herself admits she's tried every AI editing tool and none of them can even cut a podcast. So it's an observation with a direction, not yet a conclusion you can bet on.

## Where to Look Next

- The Bloomberg Odd Lots episode page and show notes for September 24, 2026, recorded live in Los Angeles
- Rachel Karten's newsletter Link in Bio, covering social teams and platform format changes; "Does Your Boss Have AI Brain?" ran there
- Taylor Lorenz's newsletter Usermag, on internet culture and tech policy
- For the platform's own framing, look up Adam Mosseri's remarks on DMs and connection at Bloomberg Screen Time

## One Thing to Take With You

**The moment you hand judgment over, you stop accumulating it.** The tool saves you time and hides the cost downstream — the step you skipped was the step that would have made you sharper next time. That boss didn't get dumber. He just skipped one rep every time, and two years from now his feel for what's funny will still be sitting on the day he started outsourcing it.

![Two lines start from the same point: one keeps climbing, the other goes flat after a dashed marker labeled the day you started outsourcing, and the height gap on the right is the practice you missed.](/figures/judgment-stops-compounding-en.svg)

Here's something I've tried, and you could run it this week. Before you ask AI "is this any good, is this right," write three lines first — your answer, your reasoning, and the part you're least sure about. Then ask. Then look at the gap: did the model see an angle you missed, or did it just say what you already knew in prettier words?

You don't need investing to do this. Picking a restaurant, making a case to your kid, weighing a job offer, deciding whether a message a friend forwarded is credible — all fine. What matters is that the three lines are yours, and that you go back and check them.