Human, verified

The BRIGHT Method, July 19, 2026

Subject line: Prove you're human

Brand New reviewed Granola on Monday and Stack Overflow on Wednesday, and I've been chewing on the pairing ever since. An AI notetaker that rebranded to look as un-AI as possible. And the site AI has spent three years quietly emptying out, rebranded around the value of human knowledge. Opposite ends of the same wave, and both walked into the week making the same claim: there are humans in here.

Everyone in tech is about to make some version of that claim, if they aren't already. So here's the rule I think these two launches teach, and I'll spend the rest of this issue defending it: "human" is the most valuable claim in branding right now, and it only works if the audience it's aimed at can check it in under a minute.

The AI company that refuses to look like one

Ragged Edge rebuilt Granola's identity around the actual handwriting of Sam Stephenson, one of the founders. The logo is loose and a little lumpy, a word scribbled down while someone else is talking. Let me be honest about the craft: if a contractor sent me that mark, those weights, that baseline, I'd send it back with notes. Here the wobble is doing the work. The whole category ships in one kit (geometric sans, gradient orb, that purple), and Granola's pitch is that the software stays out of the way so you can stay present with the person across the table. So a mark that looks like a person made it in a hurry is the pitch, sitting in the one spot nobody skips.

What makes it strategy rather than styling is how fast the claim verifies. Download the app, take one meeting, your half-typed notes come back whole. Ten minutes, checked. And the numbers went the way numbers go when the label matches the tin: it trended on X on day one, biggest download day the the company ever had, Ramp's spend data ranked them the second-fastest-growing software brand a month later, and within the quarter they'd raised $125 million at $1.5 billion. Ragged Edge will be dining out on this one for years. Deservedly.

Stack Overflow got the harder brief

No wobbly logo saves you when the market itself has moved. Monthly questions peaked above 200,000 back in 2014. By late last year they were under 50,000, and January's trade coverage clocked a December that came in 78 percent below the one before it. Developers left for a chat window trained on Stack Overflow's own answers, which is about as cruel as market forces get.

So what do you do if you're the agency? Koto didn't repaint, credit to them. They reframed the company from a Q&A destination into the source of record for verified technical knowledge. 83 million questions and answers, asked, tested and corrected by actual people in public, becuase that corpus is the one thing the chat windows can't generate for themselves. The system is called "always in build," the stacked bars keep their equity, and Stack Sans, the custom typeface, sits free on Google Fonts. That last one is my favourite decision in the whole project: the brand's voice now lives in every developer's font menu, whether they ever visit the site again or not.

(There's also a generative asset tool built with Claude that keeps the system on rails. A machine tending the brand about human knowledge. I clocked the irony, and I'd still have made the same call, a claim only stays believable if something boring keeps it consistent.)

Now, the comment sections. Veteran developers ran the check, opened the site, saw the question counts, and called the community framing a eulogy. And they're right! For them. But they stopped being the buyer a while ago. This brand launched at Microsoft Ignite, an enterprise stage, and revenue reportedly grew 17 percent to about $115 million last year on data licensing to AI labs and the enterprise knowledge product, all while the public site shrank. For a CIO buying trusted internal knowledge, or a lab paying for clean human-checked training data, "83 million verified answers" still cashes. A rebrand can't bring demand back. What it can do is aim what's left of the company at the audience that still values it, and I think that's exactly what this one is doing, whatever the comments say.

This claim is already on your homepage

If you sell B2B software, you've probably shipped a version of it yourself. Human in the loop. Built by practitioners. Real engineers answer support. I've written those lines for clients, and I can tell you buyers now check every single one. They open the support chat and time the reply. They count heads on LinkedIn. They ask the product a hard question and watch to see whether a person ever turns up. (The support-chat one is brutal. I've run it myself before signing with a vendor.) A claim that fails the check does worse than nothing, it hands the buyer proof that your surfaces say things the company can't back. And once they've caught one, they quietly re-price all the others.

The so-what for founders

Run the check before a buyer does. Open your homepage and your deck and list every sentence that promises a human: who built it, who reviews it, who answers when it breaks. Next to each one, write down the audience it's aimed at and the 30-second check that audience would actually run. You'll get one of three outcomes. It passes, so make it more specific, swap "expert support" for the name and face of the engineer who answers. It fails, so pull it today. Or it passes for a different audience than the one reading that surface, which is the Stack Overflow situation, and that's a repositioning. Fine, but do it on purpose, and move the claim into the room where it cashes.

If I could steal one move from both launches, it's specificity. Granola put a founder's actual handwriting where a wordmark usually goes. Stack Overflow counts its answers in public. A number and a name survive checks that adjectives never will.

Reply with the human claim you're least sure about and I'll tell you whether it survives. The work is at bybrightstudios.com.