Today marks the end of GitHub Copilot’s first complete billing cycle under the new usage-based model that launched June 1. The receipts are coming in, and for teams running agentic coding workflows, the numbers are striking.

Developers using agent mode and Pro+ tier models are reporting monthly bills 10 to 50 times higher than what they paid under the old flat-rate subscription. One developer in the official GitHub community discussion exhausted 54% of their monthly credit allotment with a single request. Another burned through 53% in one day running four code agents. The thread has accumulated nearly 1,000 downvotes, one of the most negatively received announcements in GitHub’s community forum history.

What Changed on June 1

Under the old model, Copilot Pro cost $10 per month and included 300 premium requests before falling back to slower free models. Agentic workflows could run freely within that structure.

The new model charges for every token consumed. Plans include monthly credits that act as a base allotment. Pro includes 1,500 credits ($15 worth), Pro+ includes 7,000, and Business includes 1,900 per user. One credit equals $0.01.

Where it gets expensive is agentic mode. GitHub’s own research found that agentic coding tasks consume roughly 1,000 times more tokens than single-turn queries. When an agent has 30 tools registered, every request includes schema definitions for all of them in the system prompt. Multiply that across a full day of development work and credits deplete fast.

Before caps, some developers were generating bills between $500 and $2,000 per month. One projected their bill jumping from $29 to $750. Another estimated $50 to $3,000 based on their usage pattern in the first week. GitHub does allow spending limits to be set manually, but they default to zero protection. You have to opt in to a cap.

Why Enterprises Should Pay Attention

GitHub Copilot is used by an estimated 1.8 million developers globally. For enterprises that standardised on Copilot as the AI coding tool of choice, the move to usage-based billing means budgets that were predictable in May 2026 are now variable in ways most finance teams haven’t modelled.

The Uber situation earlier this year was a preview of this. Uber burned through its entire 2026 AI budget in four months, then capped all employees at $1,500 per month per tool. The dynamic is the same: when AI tools are metered by consumption rather than seat, agentic workflows hit finance in ways flat pricing never did.

The other risk for enterprises is churn. Developers frustrated with cost unpredictability are already moving. Cursor, Windsurf, and Roo Code appear repeatedly in the GitHub community thread as alternatives offering more predictable pricing. For organisations that invested in standardising around GitHub Copilot, that kind of developer drift creates fragmentation.

What This Means for Business

The GitHub Copilot billing change reflects a broader shift across the AI tooling market. The two-tier pricing structure that’s emerging industry-wide puts most productivity tools in the $20 range and power tools in the $100-200 range. Neither is designed for unlimited agentic compute.

Every AI tool contract your business signs in 2026 and beyond should be read with one question in mind: does this pricing scale linearly with agentic use, or does it stay flat? If it’s the former, the cost exposure is real.

For teams using GitHub Copilot at scale, the practical next steps are straightforward. Set per-user spending caps now before July bills arrive. Audit which workflows are consuming the most tokens and whether the productivity gain justifies the cost. Evaluate whether the agentic tasks generating the biggest bills could run on cheaper models without meaningful quality loss.

The AI tool budget is no longer a predictable line item. Managing it requires the same kind of discipline that cloud infrastructure costs demand.

GitHub Copilot’s billing shift is an early indicator of where AI tool economics are heading. The free-lunch period for agentic coding is over. Businesses that build their productivity stacks on the assumption of flat-rate AI access will face surprises.

Enterprise DNA put together a free field guide on exactly this: the full Claude ecosystem, Claude Code, and how to roll agents out without breaking things. Get the guide.