Issue / 2026-05-28
A daily board of tools, apps, and references that Hacker News readers pulled into view on 2026-05-28. Each row keeps the original HN thread close to the claim.
What surfaced that day
Anthropic’s latest model; commenters called it a modest but tangible improvement and liked the adaptive-thinking toggle.
Commenters said Opus 4.8 was a “modest but tangible improvement,” praised turning off adaptive thinking, and another called the ultracode benchmark result the best so far.
Armin Ronacher’s Postgres workflow library; commenters called it lightweight and worth comparing to other workflow engines.
One comment introduced absurd as a durable-workflows implementation for Postgres; a reply said it is “very nice” and lightweight enough for a coding agent to keep in its head.
Disposable cloud sandboxes for agent work; one commenter recommended it as a useful agent UX for reducing permission churn.
A commenter said running Claude in a disposable sandbox like exe.dev is more efficient, and described exe.dev as a “new cloud provider with some very useful agent UX.”
A local Claude sandbox helper repo for running coding agents locally.
The commenter said they built dclaude for local agent runs and that it “gets my job done” when they need to run the coding agent locally.
A 60-second game about AI-agent permission fatigue that commenters immediately found fun and a bit too real.
Top comments called it “amazing” and “fun,” then used it to discuss real agent permission hygiene and sandboxing patterns.
Router firmware for Wi‑Fi roaming setups; commenters recommended OpenWRT-flashed hardware and debated the best roaming topology.
A commenter said they still like the TP-Link E610 flashed to OpenWRT better, while others compared roaming behavior across AP/channel setups.
A multiplayer online rave project; commenters called it cool and pointed straight to the GitHub repo.
One commenter said it was cool to see and compared it to past VR/DJ projects; another posted the GitHub repository link and highlighted the MIT license.
A race-sim harness for AI agents; commenters wanted a baseline solver and compared it to Gazebo and Isaac.
Commenters asked for a starter baseline, said it looked cool, and compared it to Gazebo and Nvidia Isaac as alternative simulation stacks.