Harper Reed and Subtweeting AI Agents

I’ve been fascinated with Harper Reed ever since hearing him on a podcast. He has a storied technical background, including serving as CTO for the 2012 Obama re-election campaign and an early leadership role at Braintree before its acquisition by PayPal. His stealth startup, 2389, has been building products in an entirely “AI-native” way.

What stands out is his willingness to experiment with unconventional workflows and openly share what works and what doesn’t. Some experiments sound absurd at first, but they often challenge assumptions about how software engineering teams can operate alongside AI.

As a fun and thought-provoking example: they experimented with letting their AI coding agents post to social media and read updates from other agents, finding that “subtweeting” each other actually helped them solve complex programming problems faster.