GitHub Copilot's new bill is a metered surprise

From 1 June, Copilot's flat fee became the floor, with usage billed on top and a new $100 Max tier for heavy hitters. The backlash was instant and loud — and understandable, because people had built workflows and team budgets around a predictable number that just turned into a variable one. Finance teams booked a subscription line; they woke up to something that now scales with engineering activity. That is a governance shock, even if the economics were always inevitable.

Same month: Windsurf becomes Devin Desktop, Google's Gemini CLI heads to Antigravity, and the tiers reshuffle across the board. This isn't Microsoft being greedy in isolation — it's the entire AI-tools market admitting the flat-rate land-grab phase is over and the unit economics have arrived. Compare the moves:

The through-line is one shared arithmetic, not three coincidences. The 'agentic' workflows vendors pushed you towards in 2024–25 are exactly the ones that cost the most to serve. Surprise.

Here's the part the dunks miss: usage-based billing is arguably fairer — light users subsidise nobody, heavy users pay for what they actually burn, and a meter forces discipline on runaway agent loops that chew tokens for marginal gains. The complaint is really about predictability, not principle. And that complaint is legitimate. A billing model that developers can't forecast is a governance nightmare, and GitHub owes its enterprise customers better spend dashboards before it moves the meter.

Credit where it's due: a meter on agentic workflows is defensible engineering economics. You can't charge the same flat fee for 'suggest next line' and 'refactor the entire auth module across twelve files while reading your CI logs'. Those are not the same compute load. The mistake was letting developers build habits around a price that didn't reflect cost, then switching the model without enough lead time. The anger is at the execution, not the destination — and that's a reasonable place for it to land.

The contrarian take here isn't 'rage-quit Copilot' — that's just reactive. It's this: demand legibility. The vendor that ships a clear, real-time usage dashboard with per-feature cost breakdowns and a hard monthly spend cap wins the enterprise argument, regardless of whether their base price is higher or lower. GitHub has the distribution and the GitHub.com integration depth. The question is whether it earns the continued spend by making its new pricing model governable, or whether it hands the 'CFO-approved' segment to a competitor that does. That's the competition to watch — not who has the cheapest token, but who has the most predictable bill.

So the smart move isn't rage-quitting Copilot — it's treating AI-tool spend like cloud spend: forecast it, cap it, watch the agentic workflows that quietly rack up tokens. And hold vendors accountable for the transparency their new pricing model requires. The companies that win your budget next won't necessarily be the cheapest. They'll be the ones whose bill you can actually explain to your finance team.