Daily Brief

A single internal app now handles Railway's marketing content, email, social posts, media, creator relationships and launch planning. Mahmoud Abdelwahab built it with coding agents, but getting the features to run was only the beginning. The difference between a quick SaaS replacement and a tool his team wants to open every day lives in the details: background saving, responsive interactions, reversible actions and consistent screens. His starting point is a structured codebase with reusable abstractions, so each new feature follows existing patterns instead of inventing another way to do the

A pipeline stripped the minus signs from loan amounts and turned a net $784,000 into $5.3 million of debt, with multiple models agreeing on the wrong result. Hrishi Olickel uses failures like this to explain why transforming data requires more than a universal schema or a confident agent. A customer means something different to billing, support and compliance. Useful structure depends on the source, the storage constraints, the decision being made and the assumptions a model brings to the task. His alternative is a just in time ontology: build the transformation when the question is known, and

“Fill out a form for Mr. Harper” becomes a process that lasts days: identify the right household, find missing information, contact the client, wait, follow up, obtain a signature and retain the evidence. Karthic Rao uses that wealth management example to show why a fixed flowchart breaks on ambiguity and an improvising agent creates a different reliability problem. At Perpetual, his design puts control flow, evidence and external actions in a deterministic system. Models handle interpretation and judgment at explicit, typed points. A judgment can create a smaller plan, but that plan must pass

More than 9,700 code review runs in 30 days came from a workflow that lets the model read code but never modify it. Eduardo Simon and Javier Casal describe Orbit, Flywire's internal platform for putting model steps inside deterministic engineering workflows. A YAML definition goes through the same merge request review as other code, declares the tools and access each step can use, and runs on the company's existing deployment infrastructure. The predictable work stays in code; a model supplies the judgment. Human review of merge requests remains part of the process as the platform expands acro

Full episode: https://youtu.be/LwQQ7nBCGSs Me on twitter: https://x.com/dwarkesh_sp

a pilot for our new talk show featuring your favorite opinionmakers and leaders in AI! Sponsorship and business inquiries: business@latent.space