Daily Brief

a16z General Partner Angela Strange sits down with Valon’s Andrew Wang and Linda Du to unpack what it takes to rebuild the infrastructure underneath a $13 trillion mortgage market that still relies heavily on systems designed before the internet. Linda and Andrew explain why Valon chose the hardest path: becoming a regulated mortgage servicer, translating decades of federal and state regulation into software, and proving the platform on its own loans before selling it to the industry. That foundation made Valon roughly three times as efficient as traditional servicing and created the system o

Most teams start their evals in a spreadsheet. Here's what happens when they try to grow out of it. Hossein Niazmandi, who leads solutions engineering in the West at Braintrust, explains why measuring agent quality is much more than putting a UI on a spreadsheet. He covers the two pillars of agent quality, evals before production and observability after, and why non-deterministic LLMs make both necessary. Then he walks through the stages teams go through: spreadsheets, a custom UI, side-by-side experiments for PMs and domain experts, and finally a flywheel that turns production failures into

AI agents are multiplying traffic overnight. Here's how to ship fast without taking your database down with it. Ben Dicken from PlanetScale argues you no longer have to choose between moving fast and keeping things up. He covers the principles behind reliable infrastructure, isolation and redundancy, then how to scale with sharding, back pressure and decoupling, the same ideas PlanetScale uses to run databases for fast-growing AI companies like Cursor. Then he gets to agents: how a simple VSchema JSON file lets an agent design a sharding plan for Vitess, how Traffic Control sets resource budg

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

A small model guesses the next few tokens. The big model checks them in one pass. When is that worth it? Sheilah Kirui, developer advocate at Akamai, explains speculative decoding and how to tell whether it's worth turning on. She walks through the prefill and decode phases of inference, how a small draft model proposes tokens that the target model verifies in a single forward pass, and the memory cost of hosting a second model and its KV cache. She covers how to choose a draft model, then demos both setups side by side with vLLM on a single NVIDIA Blackwell GPU: structured output ran 1.6x fa

AI changed how we write code. But 70% of engineering time goes to running and fixing it. Varun Krovvidi, product marketing lead at Resolve AI, explains why AI that writes code well still struggles to run production. Code is self-documenting, modular and a single domain. Production spans code, infrastructure, telemetry and many teams, and needs one correct answer. He covers where agents crack as you scale: anchoring bias and the model treadmill, too much or too little context, coherent answers that aren't causal, missing guardrails, and no learning across investigations. Then he lays out six p

Your agent shouldn't run on your laptop. Give it its own computer in the cloud, with one API call. Philipp Schmid from Google DeepMind walks through the Gemini Interactions API, Google's new API for both models and agents. It handles server-side state, long-running background jobs, and built-in plus custom tools across text, audio and images, and it replaces user/model turns with a timeline of steps. Then he shows the Antigravity agent in the Gemini API, which runs the same harness as Antigravity in its own managed cloud sandbox. Agents can run code, install tools, share environments with oth

If you run Claude and Codex side by side, you're the router passing messages between them. It doesn't have to be that way. Vlad Luzin, co-founder and CTO of BAND, argues the future belongs to AI agents talking to other AI agents, inside companies, between companies, and between businesses and consumers. Most of us already run multi-agent systems by hand, copying output between a planning session and a reviewing session, and loop engineering just hands that job to a script. He explains why: a single agent suffers from confirmation bias and attention dilution, no matter how big its context. The

Engineers are going from typing code to governing agents. The next step is a system that turns bug reports into production code. Davis Palmie, member of technical staff at Factory, explains the software factory: a system of agents that takes input signals like bug reports and user feedback and turns them into production code. He covers the questions every leader asks, from runaway costs and model choice to what happens to engineers and org charts, and argues coding was never the bottleneck: review, debugging, docs and testing are. He lays out the principles a factory needs (model-agnostic, so

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

Attackers use AI too, and the fastest AI attack he cites took four minutes. A pen test once or twice a year can't keep up. Eli Cohen of Snyk, co-founder of the runtime security company Helios, explains why security has to go on the offense in an agent-first world. Developers now ship far more code, much of it insecure, while attackers use frontier models to chain small vulnerabilities into critical ones in minutes. He walks through why traditional tools fall short: static scanning misses runtime issues, dynamic testing misses business logic, and human pen tests are expensive and happen only a