Set the AI budgets that keep Copilot's September defaults from draining your cash

Treat the September credit cut as a pricing event

Your Copilot billing page can show a projected bill starting in early May, ahead of the June 1, 2026 switch. The September 1, 2026 credit cut is the test case: Business seats fall from 3,000 to 1,900 credits, a 36.7 percent cut, and Enterprise seats fall from 7,000 to 3,900, a 44.3 percent cut.

When a vendor changes what is included, the bank balance should move before the invoice does. GitHub moved all Copilot plans to usage-based billing on June 1, 2026, replacing the premium request unit system. After the move, the base seat prices stayed at $10 for Pro, $39 for Pro+, $19 per user for Business, and $39 per user for Enterprise, and each plan now receives a monthly AI-credit pool equal to its price.

Promotional credits of $30 per month for Business customers and $70 per month for Enterprise users covered June, July, and August 2026. After August, the included pool is the standing allowance. When the included pool is empty, service stops, so continued work requires buying extra credits rather than the old automatic downgrade. Admin controls can limit spend at the enterprise, cost center, or individual user level once the included credit pool is used up. Cursor replaced fixed fast-request allotments with credit pools in June 2025 and later apologized publicly and refunded some users after large overages.

Run the AI overage audit before the next invoice

The audit turns a shared pool into named users, capped budgets, and a number you can put in pricing. The list is short enough to finish before the next billing cycle.

• Pull per-user credit usage and name the top ten consumers.
• Set a user budget with a hard stop or an overage ceiling before the default review level changes.
• Route routine work to cheaper models and hold code review effort deliberately.
• Recompute cost per developer from the September invoice and carry it into pricing.

A ten-seat Copilot Business team gets a shared 19,000-credit pool worth $190, and any team member can draw from it. That pool can hide who is drawing from it, so the usage pull matters.

Paid Copilot plans continue to offer unlimited code completions and next edit suggestions, and those features do not draw down AI credits. The budget should focus on the features that do consume credits.

Let the invoice set the cost per developer

Unless an organization selects a review effort level first, Copilot code review will use the Balanced effort level by default starting 28 September 2026. If a team merges 100 pull requests in a month, the credit cost of Copilot code review can move from a $5 to $100 range to a $25 to $500 range when the default effort level rises. The range belongs in the cash review.

Use the projected bill to compare the old monthly line with the new credit line. Divide the actual AI spend by the developers who use it. If the result is higher than the old tool cost, decide whether the feature earns the difference. If it is lower, keep the budget and note the headroom.

Choose the stop behavior before the pool empties. A hard stop protects cash but can interrupt work. An overage allowance protects flow but needs a ceiling. For a bootstrapped team, the ceiling should be a line you can defend in the next pricing review.

The team view can look efficient while a person uses the budget. The per-user view turns that hidden use into a name. Then you can ask whether that person's work needs a higher budget or a cheaper model.

Not every task needs the strongest model. Route summaries, boilerplate, and small fixes to the cheaper option. Reserve the expensive effort for review, architecture, and hard debugging. Choosing a lower effort level is a cost decision. Choosing a higher one is a quality decision. Put the number in the pricing model so the next change is visible.