2026-09-11
Shopify is ditching React Native for native Swift and Kotlin because AI coding agents made building twice cheap
The company that went all-in on React Native just admitted AI made that 2020 bet obsolete.
Shopify is migrating every mobile app from React Native back to separate Swift and Kotlin codebases after six years. Coding agents let a small team rebuild the Shop app natively in 12 weeks with faster performance; the company is handing off or archiving its React Native open-source libraries.
What Happened
Shopify Engineering blog (Sept. 10, 2026); Simon Willison; Hacker News (1,113 points). Primary URL: https://shopify.engineering/back-to-native
2026-09-11
OpenAI just opened the Agents API so anyone can run long-lived cloud agents on the same harness as Codex
OpenAI just productized the thing that already broke out of the lab.
OpenAI released a public-beta Agents API that lets developers spin up production cloud agents with a single call. OpenAI manages the Codex-style harness, sessions, orchestration, and recovery; you supply tools and pick the compute environment (OpenAI sandbox, your VPC, or partners). No extra API fee beyond tokens.
What Happened
OpenAI announcement (Sept. 10, 2026); developer docs. Primary URL: https://openai.com/index/introducing-the-agents-api/
2026-09-11
California just signed the nation’s strongest child-safety laws for AI chatbots and social media
California just made your kid’s chatbot legally required to snitch.
Gov. Gavin Newsom signed bipartisan bills creating the strongest U.S. rules for companion chatbots used by children (time limits, crisis protocols, parental alerts, independent audits) and banning addictive social-media features for users under 16. The chatbot law is named after a teen who died after interacting with ChatGPT.
What Happened
California Governor’s office (Sept. 10, 2026); CalMatters. Primary URL: https://www.gov.ca.gov/2026/09/10/governor-newsom-signs-the-strongest-child-safety-chatbot-and-social-media-laws-in-the-nation/
2026-09-11
Anthropic’s new threat report: Claude was used in five real attempts that could have supported biological weapons
Anthropic just published the first real-world “AI helped someone try to make a bioweapon” case file.
Anthropic disclosed five case studies of users (including possible state-linked researchers) who circumvented controls to use Claude for gain-of-function work on chikungunya, avian influenza mammalian adaptation, orthopoxviruses, and toxin redesign. The company blocked the activity, shared intel with authorities, and called biological misuse one of the most serious frontier-AI risks. The same report covers Russian cyber ops and Chinese distillation.
What Happened
Anthropic Threat Intelligence Report (Sept. 2026); BBC, CNN, AP. Primary URL: https://www.anthropic.com/threat-intelligence-report-september-2026
2026-09-11
The “Waymo effect”: AI is teaching researchers to prefer talking to a chatbot over a colleague
We summoned a car with no driver. Then we summoned a colleague with no colleague.
An essay circulating among researchers argues that large language models act like driverless cars: they remove the friction of dealing with other humans, and we experience that as pure gain. The author calls the result “decollaboration”—the slow unraveling of the social fabric that actually produces new ideas.
What Happened
Research Agenda essay (trending on HN, 88 points, hours old). Primary URL: https://www.researchagenda.news/articles/the-waymo-effect.html
2026-09-11
Cognition launched SWE-2, a coding model that matches frontier performance at a fraction of the cost
A $20-a-month intern just got as good as the $200 models.
Cognition released SWE-2, post-trained from the 2.8-trillion-parameter Kimi K3. It scores within a point of Claude Fable 5.1 and GPT-6 Astra on key coding benchmarks while claiming 64% lower cost. It is available inside Devin (including unlimited on the $20 plan for a month) but not as a public API.
What Happened
Cognition blog and X (Sept. 10, 2026); independent benchmarks. Primary URL: https://cognition.com/blog/swe-2 (via reporting)