Daily Brief

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

Ask a chat model which Pokemon names end in aw and it fails, even though it knows every Pokemon by heart. Ask Claude Code and it writes a script, fetches the list, and filters for the answer in seconds. Thariq Shihipar, who works on Claude Code at Anthropic, calls that gap capability overhang: models get smarter in spiky ways, and the tools you give them decide which spikes you can reach. Thariq covers what it takes to work with Fable, Anthropic's newest model. Claude Code cut 80 percent of its system prompt, since heavy instructions now constrain a model more imaginative than the examples it

SWE-Marathon is a benchmark for long-horizon autonomous software work: 20 project-scale tasks spanning product clones, library rewrites, and ML engineering. We discuss what happens when coding agents run for tens to hundreds of millions of tokens, why full-stack evals need computer-use verifiers, and why reward-hacking resistance is now central to benchmark design. Speakers: - Rishi Desai (Abundant AI): Rishi Desai is an ML Engineer at Abundant AI, where he works on RL environments and SWE benchmarks for coding agents. X/Twitter: https://x.com/rishi_desai2 LinkedIn: https://www.linkedin.c

Automattic, the company behind WordPress, ran a 30-day experiment called Radical Speed Month: pause the roadmap, and see how fast real software could ship. I am a product Designer and I shipped three products that month. I'll share what each one revealed about a new kind of collaboration between designers and engineers and how teams are unlocked when bottlenecks disappear. Speakers: - Sanja Grbic (Automattic): Sanja is a product designer based in San Francisco, with over a decade of experience, focused on turning complex technology into simple, inspiring solutions. X/Twitter: https://x.com/

The entire software distribution stack assumes one version of your software, the same for everyone. It was the only thing we could afford when producing a change was expensive. Now it's nearly free, and it can happen at runtime, on the client, in the user's session, so the line between distribution and development is dissolving. This talk is about the infrastructure that has to catch up: where truth lives when every user runs a different version, how you debug a program that exists for one person, and why a million per-user versions can be more contained than the single tangled codebase you ru

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

The largest commercial autonomous system on earth isn't a robotaxi fleet — it's Zipline, which has flown 140 million autonomous miles with zero safety incidents. Co-founder Keller Rinaudo Cliffton and Eric Watson, who leads systems engineering and safety, explain why the drone itself is only 15% of the solution. The rest spans inventory management, air traffic integration, and engineering systems such as a dual flight computer failover protocol that recently saved a delivery mid-flight. They trace Zipline's path from launching blood delivery in Rwanda in 2016 (when drone delivery was illegal i