Now That Production Is Free, Taste Is the Moat
Pangram Labs' analysis of 57,000 LinkedIn posts found 30 percent of comments from April to June were entirely AI-generated. Higgsfield, per the FT, was valued at $5.4 billion.
Higgsfield, an AI video company most GTM leaders hadn't heard of a month ago, was valued at $5.4 billion this week, per the FT, with Goldman Sachs and Intel on the cap table and a round PYMNTS puts at $400 million. The same day, Alibaba shipped a frontier-adjacent model that runs on a laptop. Read those together and a quiet assumption underneath every content budget dies: the assumption that production capacity is scarce.
Video was the last expensive format. Copy went first, then images, then decks. Now a $5.4 billion valuation says investors believe video production is about to cost what copywriting costs, which is approximately nothing, on models that increasingly run on hardware your team already owns. Every GTM organization is about to be able to make everything.
Which is the problem. When everyone can make everything, making things stops being a strategy. The differentiator moves upstream to the judgment about what to make, what to kill, and what never to ship. The unfashionable word for that judgment is taste, and it is about to become the scarcest line item in GTM.
Watch how the work actually splits when production goes to zero. Lenny's Newsletter this week profiled Yana Welinder, a solo founder who launched a fashion brand with no engineers: ChatGPT generated the imagery, Codex wrote the CAD files, AI drafted the vendor outreach, AI built the store. What did she contribute? Hand-drawn sketches and what the piece calls fashion prompts, detailed specs covering silhouette, volume and fabric behavior. The machine executed everything. She decided everything. Her entire remaining job is taste, expressed precisely enough to be executable.
That division of labor is the template for every content team within two years. The volume work goes to the machine. The human work concentrates into two functions: the spec (taste going in) and the kill decision (taste coming out). Teams staffed for production volume are staffed for the part that just became free.
The market is already pricing the absence of taste. Pangram Labs' analysis of 57,000 LinkedIn posts found 30 percent of comments posted between April and June were entirely AI-generated. That is what zero-cost production without judgment looks like at scale: a third of a professional network's conversation, generated by nobody, saying nothing. The slop economy is a taste vacuum, and every reader and every ranking algorithm is now being trained to route around it.
And the audience is voting, measurably, for the opposite. When Clay co-founder Varun Anand posted the company's AI writing policy, a document whose core rule is that a human must stand behind every sentence, it drew 6,400 reactions against his typical few hundred, roughly twenty times his baseline. A policy about restraint outperformed two years of product announcements. That's not a soft signal. Engagement data is the one place taste shows up as a number, and the number says the market is starving for evidence that somebody, anybody, is still exercising judgment.
Distribution is repricing in the same direction. Digiday reports brands and agencies now find creators by asking AI assistants, and one creator's citation audit found his LinkedIn posts and podcast appearances out-cited his own website. Assistants cite what earned attention, and attention accrues to distinctiveness. A feed full of competent, identical AI content makes the distinctive thing the only thing retrieval systems can distinguish.
There is also a compounding asset here that capacity never was. Research surfaced this week found 61 percent of AI assistant responses cite editorial or earned media rather than brand-owned pages. Models were trained on human judgments of what deserved attention, and they reproduce those judgments when they choose citations. Which means taste is retrieval infrastructure, not just a quality bar. The work distinctive enough to earn coverage and conversation gets cited by the machines your buyers now ask first, and the competent-but-generic work, however voluminous, doesn't exist to them. Capacity depreciates the day it ships. Taste compounds into the citation set.
The org chart is responding faster than the budget line. The State of Brand reported this week that the hottest new job in marketing is editor in chief, a hire whose entire function is deciding what doesn't ship. Clay wrote restraint into policy with an AI writing rule that every sentence must be stood behind. Gamma's Grant Lee rebuilt his product around what he called the magical first 30 seconds, which is a taste judgment about user experience expressed as an engineering priority, and rode word of mouth to $100M ARR with 50 people on the back of it. The pattern in all three: someone senior owns the no.
"Hire for taste" sounds like the least operational advice in GTM. Taste doesn't interview well, doesn't measure in a dashboard, and every mediocre marketer believes they have it. Fair. But at the organizational level, taste is a set of visible artifacts, not a mystical property. A written spec standard, like Welinder's fashion prompts. Someone actually tracking a kill rate, since a team that ships everything it makes has no filter. And one named owner of the no, which is what the editor-in-chief hire actually is. You can't interview for taste, but you can absolutely audit whether a team has built the machinery that expresses it. Most content teams have none of it, because for twenty years the constraint was making enough, and the machinery they built was for yes.
Is taste just brand, renamed to sound scarce? No, the difference is operational. Brand is the downstream residue, what the market concludes about you from what you shipped. Taste is the upstream filter, the decisions that determine what ships at all. You can buy brand help from any of the four holdcos currently pitching identical outcomes language. You can't outsource the filter, because the filter IS the company's point of view applied at the level of individual artifacts, and a rented point of view is a contradiction. Brand is what taste looks like from outside, a year later.
What a GTM leader does with this now. Recount your content team by function, not headcount: how many people produce versus how many decide, and what happens to that ratio when production drops to near-free. Write the kill criteria before the volume arrives, because a team that meets AI-scale production capacity without a filter will ship AI-scale output, and the 30 percent stat shows where that ends. Then instrument the filter: kill rate is a real metric, and a content operation killing nothing is either superhuman or unfiltered. Newsrooms have always killed a meaningful share of what gets pitched; a GTM content team with AI-scale production should expect to kill more than any newsroom ever did, because the cost of making a candidate just stopped being a reason to ship it. And put one name on editorial judgment, with authority to kill work, before your competitors' editor-in-chief job posts finish teaching the market what the role is worth.
For two decades GTM has been a capacity discipline: more content, more touches, more pipeline coverage. The capacity is now infinite and therefore worthless. Higgsfield's investors just paid $5.4 billion for the machine that makes the stuff. The returns will go to whoever knows which stuff not to make.
