Ai

Claude has replaced technical debt with cognitive debt. We can ship more code faster, but we understand less. As system complexity grows, understanding shrinks. It's not yet clear this trade improves productivity when measured over months instead of days. It might even be making me stupider by letting me outsource my thinking while convincing me that I'm faster.

Infrastructure always feels expensive right before it becomes invisible. Tokens are no different — stop optimizing and start building.

Claude's Age of Ultron is here - practical tips on using lots of Claudes for fun and profit.

At Hiro, we've fully embraced LLM coding tools, to move quickly while maintaining an incredibly high quality bar. Everyone has access to GitHub and can write production code. This capability comes with its own set of challenges - AI slop, subtle bugs, broader architectural issues. We've been intentional about changing our culture and practices to accelerate AI-driven development. Here are some of the things that are working for us.

Using a classification methodology outlined in the GPTs are GPTs paper, we build a framework to look at the impact of LLMs on various aspects of the economy. We dive into a few interesting areas based on the initial classification.

More AI, more random, less focused

Reasoning about the valuation of OpenAI, Anthropic, and other Foundational Model companies.

Growth strategy, fast trains, slack and deep work, the business of AI.