Daily Brief

Fast answers are easy to get. Answers you can trust with your money are harder. Francesco Galletta and Felippe Felisola Caso build a financial-analyst agent harness live, using one real problem: paying for a wedding. They show the five design principles they use for AI you can trust: shared domain, consistency, evidence, accountability and verification. They also explain why calculations belong in code instead of in the model's head. Note: the speakers work at Monzo but are not presenting on Monzo's behalf. 🎟️ Join us at the next one: AIE CODE Summit → https://aie.codes Speakers Francesco G

Your AI agent wrote a confident stock report saying 116% of analysts agree. Nothing in your checks caught it. This is the workshop for fixing that. Laurie Voss (Head of Developer Relations at Arize AI, co-founder of npm) builds a full evals loop on a Claude-powered financial analyst agent: trace it with Arize Phoenix, read the traces, write code evals and LLM-as-a-judge evals, test whether the judge can be trusted, prove a prompt fix with experiments, upgrade from Claude Haiku 4.5 to 5.5 with evidence, and gate it all in CI with npm test. Speaker: Laurie Voss, Head of Developer Relations, Ar

Your agent says it's done. Is it? Nick Nisi and Zack Proser from WorkOS show how to build verification loops that let coding agents check their own work. They cover clear goals, CLI health checks, browser snapshots, LLM judges and evals, so a failure drives the next fix instead of landing on your desk. Includes the live "Midnight Snack Club" demo app and the repo to follow along. Chapters: - 00:00 Why "done" isn't done - 03:17 What it takes to trust agent output - 09:34 Give agents goals they can check - 17:31 Closing the feedback loop - 28:41 Making everything feedback-loopable - 36:29 The m

A hands-on workshop on evaluating AI applications where mistakes are expensive. Annabell Schäfer and Doneyli De Jesus from Langfuse use a banking assistant to walk through the whole AI engineering loop. They cover tracing with PII masking, building datasets, turning domain-expert comments into failure modes, LLM-as-judge and code evaluators, and CI that blocks a prompt change when scores drop. Chapters: - 00:00 Why evals are hard - 05:01 The AI engineering loop - 12:47 Demo: a banking assistant with tracing and PII masking - 20:35 Building datasets - 25:18 Expert review and failure modes - 30

Insurance claims are full of handwriting, faxes, duplicates and 8GB PDFs. Denys Linkov of Wisedocs explains why that unstructured data is both the biggest barrier and the biggest opportunity for AI in insurance. He covers cleaning the data, analysis within a single document, and cross-claim insights, along with the benchmark Wisedocs built with Artificial Analysis to test how models handle medical records. Chapters: - 00:00 Intro - 01:43 Insurance's technology transformation - 09:34 How a claim actually works - 15:50 Step 1: clean data - 20:28 Step 2: analysis within documents - 23:37 Benchma

Say what you want, not how to get it. The OpenProse team shows how to write multi-agent workflows as declarative contracts in plain markdown, which any harness (Claude Code, Codex and others) can compile and run. Demos include event-driven jobs triggered by webhooks and pull requests, a "grumpy senior" PR reviewer, and an end-to-end bank model-documentation workflow. They also discuss composing contracts, pinning a winning run into a reusable workflow, and how this compares to DSPy. Chapters: - 00:00 From prompt engineering to declaring intent - 05:40 What OpenProse is - 09:58 Agents, skills

Full episode: https://www.youtube.com/watch?v=f6cjAEr08qk Me on twitter: https://x.com/dwarkesh_sp