Do not index
"Can AI just write our posts faster?" That is the question agency owners ask me most, and it is the wrong question. AI does make production faster. It does not make the work better. At most agencies, faster output on a thin idea only distributes the thinness at scale. The constraint in a content business doing $200k to $2M in revenue was never how quickly anyone could type. It was knowing what was worth saying that week.
The research now says the same thing out loud. A Warc and TikTok study of 400 marketers rated AI highly useful for production speed and far less useful for original copywriting or scripting, and found that the brands winning are the ones learning fastest from their communities rather than publishing the most. TikTok's Global Head of Creative and Brand Ads, Andy Yang, called the winning capability "cultural intelligence," the ability to read what a community values in the moment it is forming. You cannot buy that from a tool. You build it by paying close attention to one specific audience for a long stretch of time, and there is no compressed version of that.
Where AI actually compresses a content process
The way I keep this straight is what I call the Judgment Split. Every content process divides cleanly into mechanics and judgment, and the two behave nothing alike under automation.
Mechanics is transcription, formatting, restructuring a rough draft into a readable shape, cutting one long piece into five shorter ones, proofreading, tagging, scheduling. Judgment is deciding which idea earns a post this week, what a founder actually believes as opposed to what sounds respectable in their industry, which angle makes a buyer stop scrolling, and what to kill before it ships. AI is genuinely excellent at the first category and unreliable in the second, because the second requires knowing things that are not in the draft.
LinkedIn appears to have landed in the same place. The platform is retiring its own AI writing assistant in favour of a proofreading tool, which is a company drawing a line between AI that edits a human voice and AI that replaces it. When the platform that benefits most from content volume starts narrowing what its writing tools are allowed to do, that is worth reading as a signal about what the feed will reward.
In my own pipeline, AI touches most of the steps and almost none of the decisions. It cleans transcripts, proposes structures, catches the sentence I wrote twice. It does not choose the angle, and it does not decide whether a client's opinion is actually theirs. The moment it starts doing either, the work drifts toward the average of everything written about that topic, which is precisely the position you are being paid to avoid.
Why volume feels like strategy
Volume is legible. You can count posts, put the number in a report, and show a client that money bought something. Learning is illegible. Nobody can put "we finally understood what this audience is tired of hearing" on a monthly deliverables list, even though that single realization is what makes the next twenty posts work.
That legibility gap is why agencies buy production speed when they have an insight problem. A team publishing 40 posts a month with no new understanding of its audience is not ahead of a team publishing 12 with one. It is just louder. Under the Judgment Split, the honest question about any AI purchase is which column it lives in. If it compresses mechanics, buy it. If it promises to handle judgment, it is selling you the part of the job you were hired for.
This matters most for agency owners between $200k and $2M in revenue and for ghostwriters charging $5k to $30k per month per client, because in both cases the entire margin depends on the client believing the work sounds like them and not like everyone else in their category. It matters for founders running personal-brand content inside a 3 person team, where one person is writing, approving and publishing, and the temptation to fill the calendar with generated filler is strongest at exactly the moment the calendar is hardest to fill.
This is not for you if you sell content by the unit. If your offer is 20 posts a month at a fixed price and the client measures you on delivery, then production speed is the correct purchase and everything above will read as an expensive detour. Skip this if your positioning is volume. The Judgment Split only pays when the client is buying a point of view, and running it properly costs you hours that a per-unit price does not cover.
For everyone else, the practical version is a review layer rather than a bigger calendar. The agencies that hold retainers past month six tend to be running a quality control system that catches voice drift before a client feels it, which is a different discipline from publishing more. That layer is where judgment gets applied on purpose instead of accidentally, and it is the thing AI cannot staff for you.
The trajectory here is not subtle. Production is heading toward zero cost, which means production stops being a business. Every agency in your category will have the same generation speed within a year, and the client will know it, because they will have the same tools on their own desktop. What will still be scarce is the person who has watched one audience closely enough to know which of ten defensible ideas is the one worth publishing this month. If your operation is built to produce, you are building on a floor that is falling. If it is built to decide, the falling floor is working in your favor, because it removes every competitor whose only advantage was being fast.
