Daily Brief

Lincoln Murr, head of AI product at Coinbase, explains agentic finance: agents that pay with money and agents that grow it. He covers the path from AgentKit, an early experiment giving an AI a wallet, to x402, an open standard built on the long-unused HTTP 402 "payment required" code. x402 now handles hundreds of millions of agent transactions a month. Other topics are microtransactions as a replacement for the ad economy, agents running trades on simple instructions, and what two years of building taught Coinbase. Trust is earned step by step, and the AI is the enabler, not the product. He en

Every investor wants an edge: knowing first, seeing their blind spots, and acting with conviction. Anekha Sokhal, founder and CEO of Moshi, argues the most valuable context isn't in a database but in the investor's head. She explains how to build AI that does the data grind of a quant team while leaving the decision to the human. Chapters: - 00:00 Intro - 04:03 What edge really means - 06:06 What AI changed in finance - 08:26 One-person pods and the next generation - 10:28 Split the grind from the decision - 12:31 A quant team for every investor Anekha Sokhal — https://hellomoshi.ai Join us

Banks don't buy AI software until it survives their security and compliance review. Al Chen shares what it took to deploy inside five major banks without a year-long bottleneck. He covers the build pipeline, CVE remediation with agents writing and testing exploits, backup and restore verification, PII leaking through LLM judges, and why zero leakage and data sovereignty are never free. Chapters: - 00:00 Intro - 02:15 What banks check - 04:17 Fixing CVEs with agents - 08:30 Proving backup and restore - 10:33 PII leaking through LLM judges - 14:48 Making customer problems engineering problems

Robotics never got the internet's gift of free training data, so someone has to manufacture it. Josh Gao, founder and CEO of Mecka AI, runs one of the largest robot data factories, and he explains what that actually involves: custom baseball caps and tactile gloves because nothing off the shelf captures force, 100-gigabit upload lines to move hundreds of petabytes, and custom quality gates. Josh lays out the training recipe the industry is converging on (roughly a million hours of egocentric human data, tens of thousands of hours of robot data, and a few hours of in-domain teleop ) and why tac

Monzo's in-house agent started as an emoji you could react with in Slack. Now it authors over 10% of all merged PRs across Monzo's tech teams and runs more than 1,800 tasks a day. Felipe and Francesco explain why it worked so fast. Their answer is the "lots of little multipliers" Monzo already had: an MCP gateway into internal tools and knowledge, a sandboxed agent orchestration system built for a regulated bank, and a platform that turns the agent itself into a building block. That building block now handles incidents, on-call data quality and requests from non-engineers. Chapters: - 00:00 I

Buying GPUs gives you potential. The operator decides how much useful work you actually get out of them. Oliver Kennon looks at compute as a commodity market: on the same B200 chip, the serving stack alone can deliver about 50% more output. Cluster security is part of whether intelligence is "useful". SemiAnalysis's ClusterMAX rates operators, and rental price and production cost are different numbers. Your spot on the cost curve depends on your competitors, not just you. He closes on what it takes to build credible compute grades and futures. Chapters: - 00:00 Intro - 01:05 Compute is a comm

A faster model only helps if the finished job gets faster. Morgan Suo of MiniMax uses one real research task to test the changes inside the M3.1 Flash model: an analyst checking whether circuit-board suppliers are turning AI demand into cash. The changes are sparse attention on long inputs, a 4-bit KV cache, low-precision experts, speculative decoding, server load and reasoning budgets. Each one saves compute, and each can quietly cost accuracy or wait time. She then shows how to judge a model switch: check sources and calculations, count the full cost of a usable report, and make the result r

Gareth Yoder leads the cash program at Vanguard. He walks through modernizing a cash ledger that takes up to 30 million calls a day, running data, producer and consumer migrations in parallel. The turning point was exposing the legacy code base to AI tools through an MCP server. Research that took a mainframe engineer seven days dropped to 15 minutes. He closes with five lessons: make legacy code AI-queryable, treat migration docs as product, reimagine instead of lift-and-shift, adopt AI incrementally, and never skip observability. Chapters: - 00:00 Intro - 01:10 What a cash ledger is - 01:52

At GEICO, coding got fast but everything around it (decisions, sign-offs, context) stayed slow. Ford Prior's fix was a multiplayer, AI-native war room: the right decision makers, one driver, and fleets of agents wired into GEICO's knowledge, tools and code. His bet was that one day and $1,000 in tokens could map a legacy claims domain end to end and ship one real change. He ran it three times across rentals, damages and salvage. He covers FleetKit, the markdown rulebook that steers the agent fleet, plus the worksheets and diagrams it produced and what he'd change next. Chapters: - 00:00 Inner

What would you get if Bloomberg, Claude, and Netflix had a baby? Moshi founder and CEO Anekha Sokhal joins swyx at AI Engineer NYC to discuss building AI tools for traders: connecting market data, handling conflicting sources, and turning insights into recommendations people actually use. They also discuss the pressure on AI engineers to become the whole team, and how to make the most of conference conversations. Moshi: https://hellomoshi.ai/ AI Engineer: https://ai.engineer/ Latent Space: https://www.latent.space/ Chapters: 00:00 AI Engineer NYC and the one-person team 00:54 Meet Anekha So

GPU compute is bought and sold like a commodity, but delivering it is a lot messier than trading a contract. Evan Conrad of SF Compute joins swyx at AI Engineer New York to explain the market they’re building—and why he thinks it needs an independent operator. We cover: • Why SF Compute had to build cloud operations to deliver the GPUs it trades • How interruptions, storage and hardware generations affect compute prices • Why batch and live inference need different contract terms • Physical versus cash settlement, and the risk of trading a price disconnected from actual GPUs • Why NVIDIA and

How do you organize an engineering team when AI agents can do more of the implementation? Alex Lieberman and CJ Hess of Tenex join swyx at AI Engineer New York to discuss paying for outcomes, the MetaHarness process, enterprise AI adoption, and where AI-built software can—and cannot—replace SaaS. They explain how Tenex keeps plans, project context and decisions in version control, discuss what gets in the way of deploying agents in companies, and debate the economics of rebuilding software. They also consider the consequences for workers and New York’s AI ecosystem. Chapters: 00:00 Cold open