GitHub Copilot code review will start consuming GitHub Actions minutes

Original Article ↗ Hacker News Discussion ↗

Billing change & immediate reactions

  • Many were unaware Copilot code review had been using Actions for free; some stopped subscriptions when code review began triggering Actions and slowing reviews.
  • Users object to non-Actions activity consuming Actions minutes and to code review now charging both AI credits and Actions minutes, seen by several as double billing and cost obfuscation.
  • For at least one 300k LoC repo, a Copilot review uses ~5–10 minutes of Actions time, raising cost concerns for teams with many PRs.

GitHub strategy, goodwill, and “rug pull” concerns

  • Multiple commenters frame this as part of a broader trend: AI features initially subsidized to build dependency, then repriced sharply upward (“bait and switch” / “rug pull”).
  • Some think this is simply charging closer to true cost; others emphasize GitHub/Microsoft’s size and argue that a de‑facto critical infrastructure provider should behave more like a public utility.
  • Perceived “enshittification” of GitHub is a strong theme, with people recalling earlier simplicity and goodwill.

AI economics & pricing debates

  • Extensive argument over whether inference is currently profitable:
    • One side: inference alone is likely profitable; training and capex are the real money sink; API token prices may already include margin.
    • Other side: as long as training and subsidies aren’t covered, current prices are still below true economic cost.
  • Subscriptions are widely seen as heavily subsidized versus equivalent API usage.
  • Many expect further price hikes as funding tightens and providers try to recoup training and infrastructure costs.

Reliability, Actions, and CI/CD alternatives

  • GitHub Actions is criticized for reliability (measured availability under 99% and incidents being undercounted), complexity, and security issues.
  • Some still find Actions “good enough” and convenient; others report more outages than with self‑hosted CI.
  • Solo and small-team developers discuss moving to self‑hosted or cheaper CI (GitLab runners, Forgejo, Woodpecker, Drone, TeamCity, Gitea, home hardware), but note trade‑offs in idle cost, parallelism, and complexity.

Data, competition, and local models

  • Concerns about sending code to foreign AI providers versus US‑based ones; perceived IP and regulatory risks differ by jurisdiction.
  • Open‑weight and local models are discussed as a future escape hatch, but many say current local models are notably weaker and often not cheaper when hardware and electricity are included.