AI credits

AI credits are the billing primitive that emerges when you charge for AI actions rather than seat licenses. The logic: a seat-based price assumes every user consumes roughly the same amount of AI. That assumption breaks the moment you start running agents, because an agent completing a multi-step task burns far more compute than a developer asking for a single autocomplete suggestion. Credits let the tool charge proportionally to actual consumption.

GitHub's June 2026 switch of all Copilot plans to AI-credit billing made this concrete for a large developer audience, and generated significant backlash when teams discovered that agentic workflows — long debugging sessions, automated code review at scale, background agents running overnight — could cost dramatically more than their previous flat subscription. The move formalized a trend already visible across tools like Cursor, Lovable, and Bolt, which had been running credit-based models for months.

For builders making tool or infra decisions, AI credits introduce a new cost-modeling problem: you need to estimate not just how many users you have, but how 'agentic' their usage is. A team that runs one coding agent per developer per day has a very different credit consumption profile than one that queues up dozens of background agents simultaneously. Budget caps and per-user credit limits have become standard admin controls as a result.