# AWS releases AWS Emulator, a free local copy of AWS for coding agents to test against > Source: > Published: 2026-11-12 17:02:39+00:00 # AWS releases AWS Emulator, a free local copy of AWS for coding agents to test against Agents wrote and tested all of its code, AWS says, and the repo lists what still doesn't match real AWS. Four more stories inside. Today leads with AWS. Its emulator drew the biggest coding-tool thread on Hacker News in the past day, past 900 points. If your agent writes code for AWS, this changes where it can test. Also today: Google launched a model that scores choices instead of writing text. Zed's Delta now runs coding agents in the cloud. Anthropic expects to price its IPO on Nov 18. And United Airlines banned personal AI agents from booking its flights. ## What changed this week ### AWS releases AWS Emulator, a free local copy of AWS for coding agents to test against AWS released AWS Emulator on Nov 11, a program that answers AWS API calls on your own machine. Code can be tested with no AWS account and no service fee. It's built for coding agents, which write code, run it and read the result. Now that loop works for cloud code without touching real AWS. AWS architect Swami Sivasubramanian, writing on CTO Werner Vogels' blog, says it covers "more than 200 AWS services" and that "All of AWS Emulator’s code was written and tested by agents." Developer Simon Willison counted "29,684 commits in the first three weeks of the project." It's in the public domain, and an AWS skill file lets your coding agent set it up. On macOS and Windows it needs Docker. The repo keeps a "things-that-do-not-work" folder of behaviors that don't match AWS yet. Read it before you trust a passing test. ### Google launches Gemini Decision 1, a model that scores choices instead of writing text Google launched the Gemini Decisions API on Nov 11. Its model, Gemini Decision 1, doesn't write text. It scores "up to 1,000 independent questions" about one input at once, for "tool selection, policy checks, model routing": the quick calls where an agent picks a tool or model. Input costs $0.10 per million tokens. Output is free. Simon Willison calls it "almost identical to Jev," a rival model he says still costs less. Google's own benchmark says "nearly 5x cheaper" than Flash-Lite. Try it where your agent picks tools. ### Zed's Delta now runs coding agents in the cloud, free while it builds paid plans Delta, the coding-agent app from the makers of the Zed editor, added Cloud Runners on Nov 11. Until now "the agent stopped working when your laptop (or you) did." Pick Cloud and the agent runs on "a Zed-hosted Linux sandbox," so long tasks keep going and you can check in from your phone. Up to six run at once, on your own API keys. They're free on Zed plans except Student, for now. Try one on a slow test suite. ### Anthropic will set its IPO price Nov 18, with trading expected Nov 19 Anthropic's IPO site, which its X account linked on Thursday, says the share price is "expected to be set on November 18th (EST)," and shares should "begin publicly trading on November 19th" on Nasdaq as ANT. The range is $70 to $78. Third-quarter revenue was $17.0B, more than 12x a year earlier. The net loss was $23.3B. Most of it is a $21.9 billion non-cash charge that Anthropic ties to its rising valuation. The site names no change to Claude's prices. ### United Airlines bans personal AI agents from booking its flights United Airlines said Nov 11 it will block "unauthorized, third-party personal AI agents or assistants" from booking, per CBS News. Its reasons: an agent holding your login "can expose sensitive information," and one checking you in could break federal rules that travelers confirm in person what's in their bags. CBS names Meta's Muse and OpenAI's dots, assistants that act for you. Southwest made its flights viewable in ChatGPT instead. Building a travel agent? Skip United. ## Also worth your time - [Tests as data: when AI writes the code, developers become its customers, by Thoughtworks' Matteo Vaccari](https://martinfowler.com/articles/tests-as-data.html?ref=thenewway.ai) - [Ramp's November AI Index: 44.8% of businesses on Ramp now pay for Anthropic, against 40.6% for OpenAI](https://econlab.substack.com/p/ai-index-nov-2026) Know someone who'd want this in their inbox? Forward it — that's how this grows. And if we got something wrong, or you think we buried the real story today, hit reply. A person reads every one. *The New Way is written with AI. It gathers the day's stories, checks them against their sources and drafts every summary. A person decides what runs and reviews every issue before we hit send.*