The startup announced itself with a TechCrunch feature and a seed round of twelve million dollars. "Building the Future of Autonomous Software Development." Their website had a dark gradient background, animated particles, and the word "revolutionary" in three different sections. They claimed to have built a software factory that could produce production-ready applications from natural language descriptions.
Jay investigated. Not because Justin asked him to—Justin hadn't even mentioned the article—but because Jay was curious by nature and because the claims were specific enough to be verifiable.
The startup had a GitHub organization. Jay clicked through to the repositories. There were seven, most of them empty or containing only a README. The main repo had forty-three stars, thirty-one of which Jay suspected came from the founders' friends and family. The commit history told a story: bursts of activity followed by weeks of silence, a pattern characteristic of demo-driven development rather than continuous iteration.
The code itself was three Python scripts and a cron job.
Jay stared at the screen. He went back to the TechCrunch article to make sure he hadn't misread the funding number. Twelve million dollars. He went back to the repo. Three Python scripts. One called generate.py, one called test.py, one called deploy.py. The cron job ran them in sequence every six hours.
generate.py sent a prompt to an LLM API and wrote the response to a file. test.py ran pytest on whatever generate.py had produced. deploy.py copied the file to an S3 bucket if the tests passed. There were no scenarios, no satisfaction metrics, no digital twins, no specification language, no containment, no policy enforcement. There was a prompt, a prayer, and a cron schedule.
Jay showed Navan. Navan read through the three scripts in silence, then leaned back.
"Twelve million dollars," Navan said.
"Twelve million dollars," Jay confirmed.
"For three scripts and a cron job."
"And an animated particle background."
They told Justin. Justin's reaction was not what Jay expected. He didn't laugh, didn't dismiss, didn't criticize. He looked thoughtful.
"This is going to keep happening," Justin said. "The term 'software factory' is going to get diluted. Everyone with an LLM wrapper is going to claim they've built one. The distinction between a real factory—with specifications, scenarios, satisfaction metrics, containment, digital twins—and a script that sends prompts to an API is going to get blurred."
"So what do we do?" Jay asked.
"We keep publishing. We keep open-sourcing. We make the methodology visible so that people can compare. When someone claims to have a factory, anyone can look at our public specs and ask: where are your scenarios? What's your satisfaction metric? How are your agents contained?"
"Transparency as differentiation," Navan said.
"Transparency as defense." Justin paused. "The cron job will fail eventually. The methodology won't."
Jay bookmarked the startup's repo. Not to monitor a competitor. To remember what a factory wasn't.
Three Python scripts and a cron job. Twelve million dollars. The animated particle background is doing a lot of heavy lifting. This is the AI startup landscape in one story.