GitHub Copilot Just Proved That "Unlimited AI Coding" Was Never Sustainable

GitHub announced sweeping changes to Copilot on April 20. New signups for Pro, Pro+, and student plans are paused. Usage limits are tightened. Pro+ gets 5x Pro's limits. Opus models are gone from Pro entirely, restricted to Pro+ only.

The stated reason is unusually honest: "Agentic workflows have fundamentally changed Copilot's compute demands, with long-running, parallelized sessions consuming far more resources than the original plan structure was built to support."

This is the first major pricing correction driven specifically by agentic coding costs. Not model improvements, not new features. The existing users running agent sessions blew through the compute budget that the pricing model was designed around.

I've been watching this coming. When you give developers access to an agent that can run for hours on a single task, spinning up multiple parallel threads, maintaining context across thousands of tokens, the per-user compute cost isn't in the same universe as autocomplete suggestions. Copilot was priced for tab-completion. Developers are using it for autonomous multi-file refactors.

The Opus restriction is the detail that matters most. Opus 4.7 is the model that handles complex agentic workflows well. Restricting it to Pro+ means the most capable coding agent is now behind a higher paywall. The free tier and Pro tier get the lighter models. This creates a two-tier developer ecosystem where the quality of your AI coding assistant depends directly on your subscription level.

Connect this to the K2.6 story. GitHub is restricting access to the best closed models because the economics don't work. Meanwhile, open-weights models are hitting frontier-level coding benchmarks. The gap between "what you can run yourself" and "what you get from a subscription" is narrowing at the exact moment subscriptions are getting more restrictive.

For builders: cost-per-agent-session is now a real architectural constraint. If you're building tools or workflows that depend on agentic Copilot usage, model your compute costs explicitly. Consider hybrid architectures where routine tasks hit cheaper local models and only complex reasoning tasks route to frontier closed models. The era of flat-rate unlimited AI coding assistance is over. Superset, an open-source IDE for running 10+ AI coding agents in parallel using git worktrees, just crossed 9,000 GitHub stars. The market is already building alternatives.