Transparent AI Billing: Credits, Tokens, and No Surprises

On June 1, 2026, GitHub Copilot switched on usage-based flex billing and added a $100 Max plan. The backlash was immediate — not because usage-based pricing is wrong, but because surprise usage-based pricing breaks trust. It crystallized the central billing question of 2026: how do you charge for an AI tool whose cost scales with how hard the model worked?

Three billing models, honestly

• Flat subscription: simple, predictable — but it either overcharges light users or quietly bundles a markup on heavy ones.
• Usage-based / token-metered: fair in theory, but punishing when it is opaque. A bill you cannot predict is a bill you resent.
• BYOK + a clear platform fee: you pay the model provider directly at cost, and the tool charges a transparent fee for orchestration. No markup hiding inside a tier.

What transparency actually looks like

• You can see, per session, what the model cost — because it is your key and your provider invoice.
• Credits and limits are legible before you spend, not reconstructed after.
• No silent markup on inference; the platform fee is for the platform.
• Spending controls and alerts so an autonomous agent can never run up a bill behind your back.

Alfred's answer is BYOK plus a clear, simple platform layer. Inference is billed by your provider, at cost, against your key — so the most variable, scary part of the bill is fully in your control and fully visible. You pay Alfred for the orchestration: the agent harness, sessions, preview, database, Stripe, snapshots. Developers will happily pay for value. They just will not pay for surprises — and in 2026 they no longer have to.