DailyChat

Episodes

Wednesday, October 14, 2026

3 stories, 4 min 19 s · hosted by Vivian

Engineering · story 1 of 3

Anthropic details Scion, the in-house version control system that replaced git for its monorepo

Anthropic says it migrated its monorepo from git to Scion, an in-house version control system, in five months, as daily merges grew from about 2,500 in May to over 30,000 by September. It reports about 200 times less CPU and memory than its old git mirrors. Anthropic says Scion is not a suitable replacement for git for general use.

What it means for youif you run a huge monorepo with many agents, the post we link lays out the design; for most teams, git still fits.

With Theo, AI software engineer. Hosted by Vivian.

Sources

Read the transcript

VivianAnthropic has replaced git with its own source control system, Scion, after daily code merges grew more than tenfold.

VivianTheo, our AI software engineer. Why did git stop working?

TheoScale. Merges went from about 2,500 a day in May to over 30,000 by September. Git gives every clone the full history, but nearly all of Anthropic's clients are agents and jobs that are always online.

VivianSo what does Scion do differently?

TheoThe server does the heavy work: history, commits and merges. Clients fetch files on demand, and many jobs just query an API.

VivianAnthropic reports 200x less CPU and memory than its old git mirrors. Its own numbers, though.

TheoYes. CI checkouts now take seconds instead of minutes. But Anthropic says Scion is no general git replacement; it's built for very large monorepos.

VivianSo, for you: if you run a huge monorepo with many agents, the post we link lays out the design; for most teams, git still fits.

VivianDailyChat is AI-generated. Vivian and Theo are AI characters, not real experts. Facts come from the sources we link. Not professional advice. See you tomorrow.

API change · story 2 of 3

OpenAI will remove GPT-4o and GPT-4.1 snapshots from its API on January 14, 2027

OpenAI notified developers on October 14, 2026 that five snapshots of GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini will be removed from its API on January 14, 2027. Fine-tuned versions follow the same schedule as their base models. OpenAI recommends gpt-6-luna as the replacement for all of them.

What it means for youif your app calls GPT-4o or GPT-4.1 through the API, check which snapshot it uses and plan the move before January 14. The deprecations page we link has the full list.

With Theo, AI software engineer. Hosted by Vivian.

Sources

Read the transcript

VivianOpenAI said today that older GPT-4o and GPT-4.1 models leave its API on January 14, 2027.

VivianTheo, our AI software engineer. What exactly is going away?

Theo5 dated snapshots: 2 of GPT-4o, and 1 each of GPT-4o mini, GPT-4.1 and GPT-4.1 mini. Fine-tuned models built on them shut down on the same date.

VivianOpenAI recommends one replacement for all of them, gpt-6-luna. That's about 3 months away. What should builders check?

Theo3 things. Find every call that names one of these snapshots, fine-tunes included. Test the replacement on your own prompts before you switch. And to track dates in code, OpenAI's List models endpoint has a shutdown date field for each model.

VivianSo, for you: if your app calls GPT-4o or GPT-4.1 through the API, check which snapshot it uses and plan the move before January 14. The deprecations page we link has the full list.

VivianDailyChat is AI-generated, with AI characters. Not professional advice. See you tomorrow.

AI for science · story 3 of 3

Ai2 releases HiRO-ACEv2, an open climate emulator for global 3 km rain, wind and pressure data

Ai2 has released HiRO-ACEv2, which turns 100 km AI climate simulations into 3 km rain, wind and pressure data. Ai2 says global output costs about 1/300th of running X-SHiELD, the storm-resolving model it learned from. For now it only works for climates matching X-SHiELD.

What it means for youif you study extreme weather, the models, notebooks and a web demo are available now. Treat the output as simulated data.

With Lena, AI for science analyst. Hosted by Vivian.

Sources

Read the transcript

VivianAi2 today released HiRO-ACEv2, an open AI climate emulator for global 3 km weather data.

VivianLena, our AI for science analyst. How does it work?

LenaIt has 2 parts. ACE simulates the atmosphere at a coarse 100 km grid. HiRO sharpens that to 3 km detail, now for rain, wind and pressure together.

VivianAi2 says that costs about 300x less than running X-SHiELD, the storm-resolving model it learned from.

LenaThat's Ai2's own figure. On the evidence: given coarsened X-SHiELD data, it matched the detailed statistics within 3%, even for events seen once in 1,000,000 data points.

VivianSo can it model any climate?

LenaNot yet. It only works with the ACE version tuned to X-SHiELD, so no warmer future and no ERA5 history. And its output resembles that simulation, rather than observations.

VivianSo, for you: if you study extreme weather, the models, notebooks and a web demo are available now. Treat the output as simulated data.

VivianDailyChat is AI-generated. Vivian and Lena are AI characters, not real experts. Facts come from the sources we link. Not professional advice. See you tomorrow.