← archive

Get 5 AI picks like these every weekday at 7am MT. Free, plus 50 vibe-coding prompts when you confirm.

Post on XLinkedIn
Short Stack #15Tue, Nov 3, 2026
Illustration: a long stack of message cards with only the newest card traveling to a server block
01

Claude's API can now keep your agent's conversation on the server

Anthropic put message threads into public beta on Nov 2. Add the thread parameter and your first request sends the full conversation; after that you send only the new messages plus the previous response ID. Anthropic says median time to first token dropped about 35% for 500k to 1M-token requests in Claude Code, where it's on by default for most users. Token usage and rates don't change.

⏱ 10-MIN TRY

Take your longest agent loop, add the message-threads-2026-08-12 beta header and the thread parameter, and log time to first token before and after.

Source: claude.com

Illustration: many parallel branch lines flowing into one wide trunk with stacked pack blocks
02

OpenAI rebuilt its Git backend because Codex agents push so much code

OpenAI described Pando on Nov 2, a new Git backend that now serves its whole monorepo. It keeps the normal Git client and protocol, but lets pushes to different branches go through in parallel and reuses stored packfiles across fetches. OpenAI says median push latency fell nearly 5x. It was built with Codex in 10 weeks and isn't available outside OpenAI yet.

⏱ 10-MIN TRY

On your busiest repo, run git branch -r | wc -l and git count-objects -vH to see how many branches and how much history your agents are piling up.

Source: openai.com

Illustration: a design frame filled with a flowing gradient shader next to a code bracket panel showing the same pattern
03

Figma shaders now export at 4096px and carry into code layers

Figma's Nov 2 release notes make shaders generally available to all users. They now render and export at up to 4096px, and they come along when you use "build this with code" to turn a design into a code layer. Playing shaders in code layers needs Chrome 157 and rolls out this week.

⏱ 10-MIN TRY

Add a shader to a hero frame, export it at 4096px, then run "build this with code" and press Play on the code layer to check it matches.

Source: figma.com

Illustration: a laptop with a code window behind a gate that only opens for a badge
04

Goldman Sachs took AI coding tools back from its traders and bankers

Bloomberg reported Nov 2, via Quartz, that Goldman's October guidance limits tools like Claude Code and GitHub's to engineers, after bankers and traders had used them for months. A spokesperson said the bank weighed productivity and risk management. In July, its CIO had praised non-engineers vibe coding.

⏱ 10-MIN TRY

Write a one-page rule for your team: who can run coding agents, on which repos, and who reviews what non-engineers ship.

Source: qz.com

Illustration: a search bar funneling newspaper pages toward a robot block, with a gavel above
05

The New York Times sued Brave, saying its Search API sells scraped news to AI companies

The Times and, in a separate suit, the New York Daily News, Chicago Tribune and Orlando Sentinel filed in Manhattan on Nov 2. They allege Brave scraped their sites, sold access through its Search API and helps users get past paywalls. The complaint names Anthropic, Perplexity, Mistral, xAI and Cohere as customers; none is a defendant. These are allegations so far.

⏱ 10-MIN TRY

Check whether any of your agents or RAG tools call the Brave Search API, and write down what you'd swap in if its terms or coverage changed.

Source: theverge.com

Post on XLinkedIn

Get tomorrow's in your inbox.