Cursor is pushing coding agents out of the editor and into infrastructure. Its new SDK exposes the same runtime and harness behind Cursor itself, targeting CI/CD jobs, cloud execution, and embedded agent workflows inside other products.
HN liked the hack, but the real excitement was infrastructure. Cua’s background macOS driver keeps the cursor, focus, and Space in place while an agent works inside another app.
HN jumped on the trust problem before the string oddity. A case-sensitive <code>HERMES.md</code> in commit history sent Claude Code requests to extra-usage billing, and the thread zeroed in on how invisible routing rules can burn real money.
HN did not spend long on the version number itself. People jumped straight to the practical test: if Zed is calling 1.0, is the fast Rust editor finally good enough to be where humans, Claude Code, and Codex all meet?
HN did not treat this as abstract legal trivia. Once the Claude Code leak became the hook, the thread turned into a practical question for every team shipping AI-assisted software: if the model wrote the bulk of it, what is actually yours?
HN treated Ghostty’s GitHub exit as more than a forge move. What hit people was the subtext: when even a maintainer with deep GitHub history decides the relationship is no longer worth it, reliability and focus stop sounding like background complaints.
This was not just another “local models are bad” rant. The thread blew up because it mixed a blunt reality check with a serious counterargument: some of the pain comes from small models, but a lot of it may come from the harness wrapped around them.
Why it matters: agentic coding stops looking cheap once billing follows actual token burn. GitHub says all Copilot plans switch on June 1, with monthly AI Credits replacing premium requests and a preview bill landing in early May.
Hacker News was drawn less to the travel flex than to the hard limits: battery drain near 1% per minute, uncomfortable thermals, long-context slowdown, and the familiar feeling that local models still need babysitting on real work.
The improvement sounds small until you remember where agent products lose trust: waiting. GitHub says its Copilot cloud agent now starts more than 20% faster, building on a 50% startup improvement shipped in March.
GitHub is no longer talking about routine uptime tuning. In its April 28 update, the company said a 10x capacity plan launched in October 2025 had to be reworked for 30x scale by February 2026, after recent incidents hit 230 repositories and 2,092 pull requests.
Hacker News did not focus on the headline that plan prices stay flat. The thread zeroed in on a simpler point: on April 27, 2026, GitHub admitted that long agentic coding sessions cannot be subsidized forever, and predictable Copilot costs are giving way to token math.