Starting today, GitHub Copilot no longer has a flat monthly fee. As of June 1, the coding assistant flips to usage-based billing: monthly AI Credits at $0.01 each, with token consumption governing how fast you burn through them. Heavy users who’ve been on flat rate will get a reckoning. Code completions are still free; everything else runs on the meter. It’s a structural change, and its timing is not incidental.
The context: Anthropic closed a $65 billion Series H on May 28, putting its post-money valuation at $965 billion — narrowly past OpenAI’s $852 billion and making it the most valuable private AI company in the world. The engine isn’t a scientific breakthrough or a new architecture. It’s Claude Code. The developer CLI launched as an add-on and grew into Anthropic’s primary revenue driver, pushing the company’s annual run rate to $47 billion and its first quarterly operating profit. That a developer tool accounts for most of a nearly-trillion-dollar valuation tells you where value is settling in this stack.
Three days before the funding close, Anthropic shipped Claude Opus 4.8 — 41 days after Opus 4.7, the fastest upgrade cadence in the company’s history. The headline feature is Dynamic Workflows: Claude Code can now spin up to 1,000 parallel subagents for repository-scale migrations, running codebases end-to-end from kickoff to merge. The model is also four times less likely than 4.7 to let bugs slip through unflagged. Anthropic is not slowing down.
Microsoft noticed. Build 2026 opens tomorrow in San Francisco and is expected to include the first public showing of Project Polaris — an in-house mixture-of-experts model being positioned as the future of GitHub Copilot, replacing GPT-4 Turbo by August 2026. The motivation isn’t subtle: Claude Code has overtaken Copilot as the dominant developer AI in multiple recent surveys, and Microsoft can no longer afford to run a pass-through coding business on third-party models while its own tool loses share.
Meanwhile, in a warehouse
The week’s other durable image comes from San Jose, where Figure AI’s robots sorted 101,391 packages in 81 consecutive autonomous hours. What began as a planned 8-hour test ran until there was simply nothing left to run. The livestream peaked at 1.5 million viewers on X; the internet named the three machines Bob, Frank, and Gary. CEO Brett Adcock’s summary: the robots are now at human parity on sorting speed. The unexpected viral arc — millions tuning in to watch machines do a boring job, giving them names and personalities — says something about where we are culturally with automation.
The through-line is hard to miss. The dynamics reshaping developer tooling — automation shifting from experiment to load-bearing infrastructure, competition moving to who controls each tier — are playing out in warehouses on a longer horizon. GitHub changed its billing model today because AI coding is infrastructure now, not a nice-to-have. People watched Bob, Frank, and Gary for 81 hours because some part of the internet still isn’t sure whether to be amazed or unsettled.
The answer, probably, is both.