Bragging about humans employing AI

This is my newsletter where I share cool stuff humans are doing on the internet.

You might remember me from my CatGPT project 🐈‍⬛ - I made it in one day to show you can make fun stuff with AI, and millions of people loved it!

Today, we're bragging about three AI’s doing real, unglamorous work: one that haggled a car dealership down $4,200 while its owner did nothing, a swarm of a thousand fake people that's quietly beating a prediction market, and one that booked 614 sales meetings. Plus one bonus bragging moment: two researchers who caught an AI confidently “finding aliens” before NASA could learn that lesson the hard way. 🤖👽😹

Let's get into today's bragging!

Adam

PS — Did someone forward you this email? Join the crew here. Reply and tell me what you’re building, I read every single email.

AJ Stuyvenberg wanted a 2026 Hyundai Palisade in the exact color combo his wife picked. Instead of losing a weekend to the dealership, he handed the whole job to an AI agent.

He runs OpenClaw (you might know it as Clawdbot, before Anthropic's lawyers had a word about the name), an open-source agent that lives on a spare computer and takes instructions over WhatsApp. He had it research fair prices on Reddit, find every matching Palisade within 50 miles of Boston, and fill out contact forms at every dealership with his real name and number.

Then he set it loose: check the inbox every few minutes, and whenever a new quote lands, forward it straight to the last dealer and ask them to beat it. For three days, the agent played dealers against each other over email while Stuyvenberg went about his life. It sent one quote to the wrong dealer at one point. Otherwise, it just worked.

Final number: $4,200 off, landing at $56K, below his target. He never set foot in a showroom. 🚗

Guo Hangjiang, a college senior in Beijing who goes by BaiFu online, built MiroFish in ten days of what he calls vibe-coding: an engine that spins up thousands of AI agents, each with its own personality and memory, and drops them into a simulated social network to argue about whatever question you feed it. Watch enough fake people debate an election, a product launch, or a war, and the crowd's behavior turns into a forecast.

The project hit #1 on GitHub's global trending list, ahead of repos from OpenAI, Google, and Microsoft that week. Within 24 hours, Chen Tianqiao, once China's richest man, wired $4.1 million to fund it.

Someone fed the engine the first 80 chapters of a 250-year-old novel with a famously unfinished ending and let the simulated characters write their own conclusion. Someone else wired it into a Polymarket trading bot: 2,847 digital humans debate before every trade, and it's up $4,266 over 338 bets so far.

Ten days of building, and a swarm of imaginary people is out here beating the market. 🐟

Jason Lemkin, founder of SaaStr, ran his company's entire inbound sales motion through one AI agent named Amelia. For a single event, Amelia booked 614 qualified meetings, at roughly $85K average ticket size, across 2.25 million site sessions and 402,000 direct interactions.

SaaStr runs go-to-market with about two people. Amelia routes leads to the rep who statistically closes that type of deal best, runs win-back campaigns for people who almost bought, and hands out discounts on a fixed schedule instead of panicking into a last-minute 30% off to save a wobbling deal.

The old way was a contact-us form and a two-to-three day wait while the lead went cold. The new way is a machine that never sleeps, never quits, and never has a bad week. That gap is the whole story. 📈

Christoph Adami, a computational biologist at Michigan State University, and his student Ankit Gupta wanted to know whether the AI models NASA plans to use for finding alien life could be fooled. So they tried to fool one.

They ran three months of computer analysis across 1,000 parallel machines inside Avida, a digital-life program Adami built back in 1993 where strings of code replicate and compete like real organisms. Starting from a random sequence with nothing “alive” about it, they nudged it one tiny mutation at a time and watched the AI's confidence climb.

It took about 15 changes. Every single time, the AI landed on 100% certainty it was looking at life. Every single time, it was wrong. As Adami told Space.com: “AI has a big Achilles heel when it is trying to classify things that are unlike the things in its training examples. It is just incredibly easy to get AI to misclassify.”

That's not a lab curiosity. NASA's Europa Clipper and the future Habitable Worlds Observatory are both set to lean on AI to sort through exactly this kind of data, hunting for biosignatures on Europa, in Venus's atmosphere, and on distant exoplanets. Adami and Gupta caught the flaw on a laptop, months before an agency could catch it live, in front of the entire planet.

They present the full findings in August at the 2026 Conference on Artificial Life. Somewhere out there, an AI is currently very confident it just found a Martian. It's wrong, and now there's a paper proving it. 👽

An agent that talked a car dealership down $4,200 while its owner did literally nothing. A swarm of a thousand imaginary people quietly beating a prediction market. An agent outperforming a whole sales floor. And two researchers who caught an AI confidently “finding aliens” before anyone believed it. Real builders pointing agents at real, unglamorous work — and real scientists keeping them honest. That's worth bragging about. 🧡