Dex Horthy on Ralph, RPI, and escaping the "Dumb Zone"
GitHub buckles under agents, the AI Vampire, and moving your terminal to the cloud.
When the Ralph autonomous loop was born, Dex Horthy was "in the garden," witnessing the spark that set the AI engineering community on fire. Andrew sits down with the HumanLayer founder to discuss how to escape the "Dumb Zone" by applying his strict RPI (Research, Plan, Implement) methodology - a process that forces agents to generate intermediate design artifacts and align on architectural decisions before writing a single line of code. They also break down the brutal economics of agentic coding, recounting how Dex’s team used autonomous loops to clone six sponsor products overnight at a hackathon.
1. GitHub struggles with availability
GitHub had a rough start to last week. Why? Likely because we all rolled out of the weekend with thousands of AI-generated commits ready to go. Ben Lloyd Pearson and I discussed how this outage highlights the fragility of human-centric infrastructure in an agent-first world. When the token machines turn on Monday morning, the current pipes simply aren’t big enough to handle the volume. We are trying to force agent-scale traffic through human-scale pipes, and the internet is breaking because of it.
Read: GitHub appears to be struggling with measly three nines availability
2. 10x the work, 1x the pay
Steve Yegge is back with a reflection on Gastown culture and the risk of burnout. He argues that while agents allow us to do 10x the work, we aren’t getting paid 10x the salary. Instead, we are just feeling the pressure to multi-thread our attention spans until we burn out. It is a necessary look at the human cost of the efficiency we are all chasing. The company captures the value of your increased output, but you are the one paying the cognitive price.
Read: The AI Vampire
3. Scaling down to scale up
New research from Tim Dettmers at Ai2 shows that open coding agents trained on specialized codebases can actually outperform massive foundation models. This is huge for the brownfield problem where generic models fail to understand custom architectures. This research suggests the future involves scaling down to scale up. It opens the door for highly specialized, private agents that actually know your stack better than a frontier model ever could.
Read: My Journey Towards Coding Agents: Building SERA
4. Warp launches Oz
If you are like me, you might already be renting a VPS just to keep your agents from locking up your laptop resources. Zach Lloyd joined the show to announce Oz, a new platform for launching and orchestrating cloud agents. Think of it as “Vercel for agents.” It solves the local resource constraint problem and allows you to share agentic outputs with your team rather than keeping them trapped in your local terminal. It is a major step toward making agentic workflows collaborative rather than isolated.
Read: Introducing Oz: the orchestration platform for cloud agents
Watch: Zach’s previous episode on the future of the terminal. It offers a fascinating look at how the conversation has shifted from local command lines to cloud orchestration in just a few short months.