Two posts hit the timeline today saying the same thing from different mics. Greg Isenberg published a thread on X (70.9K views) about design.md, the markdown format Google open-sourced for capturing the soul of a design system in a single file: palette, typography, tokens, spacing, animation rules. Mengto, Greg’s podcast guest the same day, walked through the full workflow in a 50-minute session: design.md plus HTML plus skills, all pointing at the same file, all generating consistent output across landing page, app, pitch deck, ads, and promo video. The line Mengto repeated three times: “taste is the real value here.”

The day before, Greg posted the other half of the thesis as a separate piece: only about a thousand companies on Earth doing more than five million in revenue are truly AI-native. Most use ChatGPT at work and call themselves native. “That’s like saying you’re a chef because you own a microwave.”

The connection neither one made explicit: being marketing-native means your brand lives in a file an agent can read in 30 seconds. If it doesn’t, you will generate 50 ad variants this week and they will all sound like they came from the model, not from your brand.

The Easy Half Now Costs Zero

Before the rest of the argument, it helps to be specific about how cheap the easy half of the problem became. Crowdlinker launched designmd.me on April 10: paste any URL, get a complete design.md in seconds. No registration. No payment. Around 12,800 files generated to date. Mengto’s own tools (Aura, New Form, Variant) sit at the same price point. The line from the podcast that captures the air that just moved: “the only moat that we have right now is just we can click a button.”

And as Greg was posting the design.md thread, Anthropic shipped something that moves the back wall. The Anthropic-SpaceX Colossus 1 deal: 300 megawatts of new capacity, more than 220,000 Nvidia GPUs, Claude Code rate limits doubled, default Opus API rate limits 5x’d at twice the previous price. The constraint that strangled marketing teams for two years (how many tokens fit in the quarterly budget) is moving toward zero.

That changes the question on the marketing table. It is no longer “can your AI generate 50 ad variants this week?” but “do you have the taste filter to kill 47 of the 50?” Boris Cherny named the pattern at Anthropic, and Aakash Gupta expanded on it in his May 3 LinkedIn post: building got cheap, deciding what to build did not. The same pattern shows up in marketing the moment an agency hands you 50 ads that don’t look like your brand and nobody on the team has the calibrated taste to kill the 47.

The Hard Half Is Where the Brand Gets Decided

The easy half costs zero and takes 30 seconds. At the end of the process, you have a design.md sitting in a Google Drive nobody opens until the next launch. The hard half has six pieces, and none of them are sold by the file. The five most teams skip:

Taste curation with a calibrated kill rate. Someone with the judgment to kill 80% of the generated assets before noon. Below 50%, you’re not generating enough. Above 95%, your ingestion pipeline is broken or your taste is too narrow. The voice that says “no” needs to be named and available every morning, not in an approval meeting every two weeks.

Brand voice memory across sessions. Today, Anthropic announced three things on its Managed Agents platform: dreaming (the agent reviews past sessions and curates its own memory), outcomes (external rubrics that grade agent output and force iteration), and multi-agent orchestration. Without persistent memory, every new prompt starts from “describe our voice in three words.” With persistent memory, the agent remembers what your product is actually called, the one word that is banned in headlines this quarter, and the structure of the last campaign that cleared the bar two weeks ago.

Cross-channel cohesion evals. A rubric that scores whether the visual DNA holds across the Meta ad creative, the landing page, and the product onboarding screen. Without that rubric, every agency invents its own DNA and the brand dilutes across five campaigns in six months.

Named decision rights. Who approves a campaign without re-aligning across three teams? If the answer is “the committee,” AI marketing speed dies before the first quarter is out.

Audit trail. Six months from now, who can defend why that campaign shipped? If the answer is “I don’t remember, the agent approved it,” you have a governance problem, not a creative problem.

I have watched the same pattern five times since 1990. Back then it was corporate identity binders, printed and locked in the marketing director’s office. The 2000s replaced the binder with a brand bible in Word that nobody opened after kickoff. By the 2010s, the artifact had moved to a design system inside a Figma library, which a human could browse but a machine could not. The 2020s upgraded that to design tokens and Storybook, which became legible to code but stayed opaque to agents. And here we are in 2026, where design.md plus persistent memory finally produces something an agent can not only read but also learn from between sessions. Every one of those cycles played out the same way. The company that could describe its brand on a whiteboard in two minutes paid 1x for the rollout. The one that couldn’t paid 3x and is still maintaining a Frankenstein customization layer. The new thing is the compression. What used to take five years to reveal where the bottleneck landed now reveals itself in six months.

The Operational Number the Chorus Skipped

In the replies to Greg’s thread, among 100+ comments, Simplr Intelligence dropped the operational line nobody else quoted: “feed the same design.md into your ad creative pipeline too. When the post-click vibe matches the pre-click ad, CPA drops noticeably; saw ~18% on Meta in a recent test.” 18%. That is the line that pays for the readable-brand audit.

Most marketing teams have one agency for Meta, another for the landing page, and another for the product. Each shows up with a brief written in isolation, a palette that almost matches but doesn’t, and an interpretation of the voice that nobody bothered to reconcile. When the customer’s three points of contact come from three different DNAs, conversion drops. Not because the talent is weak, but because there is no shared file the three agencies read at the start of every conversation.

A comment from Patrick (@patrickssons) in the same thread shifts the headline: “the real win isn’t looking like a 100M company. It’s that your second hire ships on-brand without three weeks of design reviews. Brand consistency is a hiring multiplier.” That sentence is the institutional version of design.md: an artifact that outlives whichever agency you fire next quarter, whichever designer hands in their notice, and whatever org refactor lands in Q3. If your brand lives in one person’s head and that person walks out on Friday, by Monday you are starting from scratch.

What to Ask in the Next Marketing Meeting

Five questions to put on the table this week, before approving the next quarterly budget:

1. Readability. Does your brand exist in a file an agent can read in 30 seconds, or only in a deck the committee approved three quarters ago and nobody has opened since?
2. Shared DNA. When the Meta ads and the landing page get redesigned, do both teams pull from the same design.md, or does each agency invent its own DNA and then we argue about why conversion dropped?
3. Kill rate. What is the discard rate on AI-generated assets this week? If nobody knows, there is no curation. If it sits above 95%, the pipeline or the taste filter needs adjusting.
4. Memory. Does the brand voice persist across agent sessions, or does every new prompt start from “describe our tone in three words”?
5. Owner. Who is the named owner of design.md, voice.md, and brand.md for the next 90 days, capable of defending why each decision lives where it does?

If your team can’t answer four of the five in under fifteen minutes, you don’t have a creative problem. You have a readable-brand problem. And the gap between the company that fixes it this quarter and the one that pushes it to the next is six to twelve months of cost-per-acquisition advantage.

At IQ Source, AI Maestro runs the internal version of that audit. We map whether the brand actually exists as code an agent can read. We test whether the voice survives across sessions. The cross-channel cohesion evals get scored against real customer-facing surfaces, not slide mockups. And we name an owner of the file for the next 90 days, with the authority to defend each decision in writing. For software companies whose product UI is part of the marketing surface, the Tech Partner engagement extends the same design.md into the moat (defensible competitive advantage) of the product itself, where brand and product experience become inseparable; that is the pattern we already covered in runtime as commodity, workflow as moat. And for teams figuring out how this translates into their actual campaign pipeline, the Marketing service brings the full workflow: design.md, cohesion evals, kill rate, named owner.

The easy half costs zero. The hard half decides your next quarter. If the conversation in your team is still about which ChatGPT plugin to wire into Slack, the thousand AI-native companies Greg is counting are already six steps ahead. The good news is that there aren’t a thousand different winners in every market. There is room for your company to be the first marketing-native player in your vertical in Latin America or anywhere else. The question this week is who, on your team, opens the file first.