Eduardo Samayoa, Founder, AI Clear
By day seven, I wasn’t building anymore. I was watching.
The AI had taken the brand voice document from day three, cross-referenced it with the market research from day one, and was writing ad copy that referenced the product descriptions it had written itself two days earlier. It was iterating on its own logic. I hadn’t touched it in 36 hours.
That was the moment I understood I wasn’t using a tool. I was operating a system. And I had no idea what it was going to do next.
Let me back up.
Eighteen months ago I set myself a challenge: build a complete, revenue-generating ecommerce brand from scratch in nine days using only AI. No designers. No copywriters. No agency. Just me, a set of prompts, and about $400 in AI costs.
Day one and two was market research and product selection. The AI analysed trending products, competitor pricing, and search volume, then recommended a niche I wouldn’t have picked myself. I almost overrode it. I didn’t.
Days three and four was branding. Name, logo direction, colour palette, brand voice guidelines. All AI-generated, all reviewed by me, none of it written by a human.
Days five and six was the store build. Product descriptions, collection pages, homepage copy, email sequences.
Days seven and eight? Ad creative and campaign setup.
Day nine – launch.
Within 72 hours, the brand had made its first sale. I hadn’t written the product page. I hadn’t written the ad. I hadn’t even chosen the product. Someone had handed over real money for something that, in any meaningful sense, the AI had built.
I felt less like I’d built something and more like I’d orchestrated something. Building implies you know how every piece works. Orchestrating means you set the conditions and trust the system. That distinction matters more than I realised at the time.
The pricing decision that changed everything
I ran this experiment multiple times. Some brands worked. Some didn’t. The ones that didn’t taught me more.
Two brands didn’t convert. One was a market problem. The niche was too small, and I should have overridden the AI’s research when my instinct said the audience wasn’t there. That was my fault. The other was a creative problem – the brand felt generic. The AI had optimised for correct instead of distinctive. That was the AI’s fault. The lesson was the same either way: competent and compelling are not the same thing.
The first thing the AI got right that I would have got wrong was pricing. I would have underpriced to compete. The AI priced based on perceived value and competitor gaps – not margin. It was right. That was the first time I realised I was bringing operator bias to decisions the model didn’t have.
The first thing the AI got wrong was brand voice consistency. The tone on the homepage didn’t match the email sequences, which didn’t match the ad copy. Every surface was technically good. They just didn’t feel like they came from the same company. Models drift. I had to create tighter guardrails and re-run everything. The AI could build a brand. It couldn’t maintain one without guardrails. That’s not a criticism, but a design constraint. And it’s one most operators deploying AI have never thought about.
But the decision that actually changed my life happened later. At store thirteen.
I was running multiple brands simultaneously by that point, each with AI handling different parts of the operation. I was reviewing a pricing decision on one store and realised I couldn’t explain why the AI had priced a product the way it had. Not that it was wrong. I just couldn’t trace the logic.
If I couldn’t explain it – and I was the person who built the system, who had designed every prompt, who had watched every output – what would happen when a regulator, a customer, or a lawyer asked a founder who had simply downloaded an app from the Shopify store and pressed go?
I sat with that question for a while. I’d built something that worked. Revenue was real. The brands were live. By any measure I’d succeeded at what I set out to do.
But I realised I couldn’t tell you, at least not precisely, what my AI had decided, why it had decided it, or what would happen if it decided wrong at scale. I had moved fast and set things in motion that I no longer fully controlled. I wasn’t being reckless, but I had made something I couldn’t fully explain, and had released it into the world.
The gap nobody was filling
I went looking for an independent transparency standard I could apply to my own AI operations. Something that would let me answer the question: what is my AI actually doing, and how do I prove it?
It didn’t exist.
Every AI governance platform I found was selling internal compliance tools to enterprises. Nobody was measuring what companies actually disclose publicly about their AI practices. Nobody was building the independent, outside-in standard that regulators, buyers, and the public actually need.
So I built one. AI Clear is, in effect, a credit score for AI transparency. It rates how transparently companies disclose their AI practices, scored against a published rubric, built entirely on public information and open to anyone at aiclear.org. The rating is outside-in, the way credit is scored from the outside, on the evidence, whether or not the company likes it. No company pays for its grade, and no company can change it.
Why this matters right now
In April, the New York Times profiled Matthew Gallagher, who used a stack of AI tools to build MEDVi, a GLP-1 weight-loss telehealth company, into a business that did $401 million in sales in 2025 with two employees. The profile presented it as proof that one person and AI could build something enormous. Within a day, a second story emerged that the profile had missed. Reporting that followed surfaced hundreds of fake doctor accounts on Facebook, AI-generated deepfake before-and-after photos, and misleading advertising. The FDA had already sent a warning letter for misbranding the compounded drugs. A class action followed.
Whatever Gallagher’s intent, the point for the rest of us is the same. AI made it possible to build and scale all of it fast, and there was no independent way for anyone, the Times included, to check what the company actually disclosed about its AI, its practices, or who was accountable when something went wrong. Fast and transparent are not the same thing, and nothing existed to tell them apart.
This is already changing how companies get evaluated, and the pressure is coming from the market before the law. Enterprise procurement teams are adding AI governance to standard vendor reviews. Institutional investors are folding it into due diligence. Buyers are putting AI accountability on the same checklist as security and privacy. The law is moving the same way: Colorado’s AI Act takes effect January 1, 2027, the EU AI Act is phasing in through 2026 and 2027, and California, Texas, and Illinois have passed their own AI transparency and governance laws, with New York and others close behind.
The question every one of these audiences is asking is the same one I was asking at store thirteen: what AI system does this organisation operate, what decisions is it making, and what happens when something goes wrong?
Most companies cannot answer that question. Not because they’re hiding anything. Because they’ve never had to.
That’s about to change.
AI makes competent decisions. Competent and compelling are not the same thing. And in a world where AI is making decisions about what you pay, what you see, and whether you get the loan, competent isn’t good enough without transparent.
I built twenty businesses before I understood that. You don’t have to.
Eduardo Samayoa is the founder of AI Clear, an independent AI transparency rating system, and previously co-founded Thinkr, an AI-powered ecommerce platform. He is based in San Francisco.
