Before the Linotype, a newspaper page was composed by hand. A compositor stood at a case of type, picking letters one by one into a composing stick, and every word passed through his fingers twice: once as he set it, and once as he justified the line. Mistakes still reached the page, but many never did, because the work was slow enough that a second reading came free with the first. When Ottmar Mergenthaler's machine arrived in the 1880s and began casting whole lines of type from a keyboard, composition stopped being the bottleneck. Papers that wanted clean copy had to invent something the old process had not needed: a separate desk whose only job was to read. The copy desk was not a luxury of the modern newsroom. It was the bill that came due when speed removed the unpaid editor who had lived inside the delay.
Marketing is sending that bill now.
The Free Catch Is Gone
For most of the last twenty years, building a campaign took long enough that several people touched it before anything went live. An offer sat in a brief, then in a deck, then in a ticket, then in a draft, and somewhere in that chain someone noticed that the sixty-day guarantee had been retired in the spring, or that the audience list still included the segment legal had barred, or that the product claim on the landing page no longer matched the one in the catalog. The catch was not a process anyone designed. It was a side effect of calendar and headcount. Scott Brinker, writing this week in the language marketing operations teams have started to borrow, calls the gap that opens when that side effect disappears the alignment gap: the distance between what the customer is trying to do, what the company knows and wants, and what the systems can actually represent and act on.
AI collapses the middle of that chain. A team can now generate a nurture sequence, a set of ads and a landing page in the time it used to take to schedule the kickoff. Bynder's 2026 State of Digital Asset Management report found that 97% of marketers say AI has changed content operations, and Pantheon's State of the Web found that 79% of marketing leaders already treat manual review of AI output as an obstacle to wider adoption. The second number is the tell. Review feels like friction to people who have just been handed speed, which is exactly how a compositor in 1890 might have felt about the new desk that kept sending his galleys back.
Context Is Not a Prompt
The usual answer is to write a better prompt, or to paste more documents into the agent. Brinker’s point, and the one worth stealing, is that this misunderstands where the failure lives. An agent can only work with the context it is given: customer history, product facts, brand rules, approval rights, pricing, the list of claims legal will still stand behind. Most companies already have that information somewhere. What they do not have is a single, current, owned version of it that every system reads from. A messaging change that lives in a slide deck and never reaches the agent, the email tool, the ad account and the site is not a messaging change. It is three slightly different companies speaking at once.
The same pattern shows up when separate agents optimize their own channels. One improves an email sequence against open and click rates. Another reallocates paid spend against last-click conversions. A third rewrites onsite copy against form fills. Each can be doing exactly what it was told to do while the customer receives three versions of the offer, three tones and three next steps. Before AI, the calendar forced those teams into the same room often enough that the contradictions had a chance to surface. Empty the calendar and the contradictions ship.
The economics are familiar from any process that hides quality inside latency. When a factory line moved slowly enough that a foreman could walk it twice an hour, defects were cheap to catch and expensive to miss. Speed up the line without adding inspection and the same defects leave the building at the new rate. Marketing's production queue was that slow line. The people in it were not hired as inspectors, but inspection is what the queue bought. Take the queue away and inspection has to become a job, with an owner, a cadence and a budget, or it simply stops happening.
You Have to Staff the Desk
None of this argues against the speed. The Linotype was a genuine advance, and so is a marketing team that can move in hours instead of weeks. The argument is about what the advance costs when nobody pays for the function it displaced. A company that treats review as an obstacle to adoption is choosing, usually without noticing, to run without a copy desk. The mistakes that used to die in a ticket queue will now die in front of a customer, or they will not die at all.
The operating move is unglamorous and concrete. Decide which facts and rules the agents are allowed to act on, name an owner for each, and put a date on when they were last checked. Set risk tiers so that a subject line can ship with light review and a pricing claim cannot ship without a named person. Make the update path shorter than the generation path: if a product rule changes on Tuesday, every system that speaks for the brand should know by Wednesday, not by the next quarterly cleanup. The teams that do this will look slower than the ones that do not, in the same way a newsroom with a copy desk looked slower than one that sent Linotype galleys straight to the stone. They will also be the ones whose customers still trust what they read.
Speed removed the unpaid editor. The desk has to be hired on purpose.
