Hacker News Daily | Today’s Top Stories: Native Mobile, AI Agents, and Trust

Hacker News Daily — September 11, 2026

Today’s episode covers the day’s most discussed Hacker News stories, with context from the linked sources and community discussion.

Featured stories

Shopify is moving from React Native back to Swift and Kotlin

Shopify says changing economics—particularly the effect of AI-assisted development—led it to revisit its mobile-stack decision and return to native Swift and Kotlin. HN commenters debate native performance and maintainability against React Native’s shared codebase and over-the-air update advantages.

Original: Shopify Engineering

HN discussion: news.ycombinator.com/item?id=49643982

OpenAI Agents API

OpenAI introduced an API aimed at the infrastructure around agentic applications, including tools and sandboxes. The discussion focuses on state, memory, isolation, self-hosting, model portability, and the risk of vendor lock-in.

Original: OpenAI developer documentation

HN discussion: news.ycombinator.com/item?id=49649213

The Deathray: A simple way for an untrusted site to freeze a Mac

The demonstration shows how a website can use graphics workloads to freeze or crash a computer. HN commenters discuss the long history of similar WebGL behavior, macOS GPU recovery limitations, and whether disabling WebGL or WebGPU is worth the compatibility trade-off.

Original: Auberon’s technical write-up

HN discussion: news.ycombinator.com/item?id=49649124

Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra

Cognition announced SWE-2 as a new software-engineering model. The HN community questions how well headline benchmark results generalize to newer tests, and asks for clearer statistics, reproducibility, and information about model openness.

Original: Cognition

HN discussion: news.ycombinator.com/item?id=49645443

More questions about whether researchers can trust OpenAI with unpublished math

A highly discussed thread examines allegations and counterclaims about the relationship between private mathematical research, model conversations, and later AI-generated work. The central community concern is whether data provenance and non-training assurances can be independently verified.

Original: Source post and linked discussion

HN discussion: news.ycombinator.com/item?id=49639408

Editorial note

Story rankings, scores, and discussion counts were collected from the official Hacker News API at approximately 08:55 UTC on September 11, 2026. Community comments represent individual opinions and are not independently verified claims.