The Claude Billing Change and How to Reprice Agency Work

Straight up: the Claude billing change that landed on June 15, 2026 is a pricing event for your agency, not a technical footnote. As of that date, programmatic Claude usage - the Agent SDK, headless 'claude -p', Claude Code in GitHub Actions, and third-party agents - no longer counts against your Pro or Max plan limit. It moves to a separate monthly credit you claim, and bills at standard API rates after that (Anthropic's details are at https://support.claude.com/en/articles/15036540-use-the-claude-agent-sdk-with-your-claude-plan).

Why should you care more than a hobbyist does? Because the agents you build for clients run in the background, on cron, all month. That usage used to hide inside your personal subscription for free. Now it is metered. If you do not adjust, that cost quietly comes out of your margin.

Here is the trap most shops are about to walk into. Say you have a Gmail-to-CRM agent on a $2,400/month retainer, running every few minutes on a client's inbox. Until June 15, that agent's compute rode along on your Max plan and felt free. You priced the retainer on your time to build it, not on what it costs to run.

Now multiply that by every client agent you have in production. Each one draws on the new programmatic track. Once your monthly credit is spent, every run after that bills at API rates. That is real money leaving your account every month for work you already quoted as a flat retainer. The cost did not appear - it was always there. June 15 just made it visible.

I spent eight years inside Apple, PlayStation, and Schwab, and the one thing enterprise pricing always got right was separating build cost from run cost. Apply the same discipline here. Three moves:

Notice the pattern: you are not raising prices across the board. You are pricing the build and the run separately, the way every serious infrastructure vendor does. We broke down the deeper logic of pricing the workflow instead of the hours in our writeup on Claude Code workflows that get you paid (/blog/5-claude-code-workflows-that-get-you-paid), and this change makes that split mandatory, not optional.

Before you reprice anything, claim what Anthropic is giving you. Eligible Pro, Max, Team, and Enterprise users get a separate monthly Agent SDK credit that applies only to programmatic usage and refreshes with the billing cycle. Eligible users were emailed instructions before June 15 - find that email.

The detail that bites agencies: the credit is per-user, not pooled. A five-person shop cannot run everything through one account's credit and call it covered. And members on seat-based Enterprise plans with Standard seats are not eligible for the credit at all. Map your seats before you map your prices.

Here is the mechanic that protects your margin: stop treating Claude usage as a cost you absorb to look cheap, and start treating it as infrastructure the client pays for. Name the outcome - 'your inbox triaged and logged to the CRM, every five minutes, forever' - price the build and the workflow as a deliverable, and add a transparent usage line for the compute that runs it.

Clients already pay this way for hosting, for Twilio, for every API their software depends on. An always-on Claude agent is no different. When you frame programmatic usage as the engine under the deliverable, a usage line is obvious and fair - not a surprise. If you want the full scoping-and-packaging playbook, we laid it out in how to price a Claude Code agency project (/blog/how-to-price-a-claude-code-agency-project), and you can see more of how I package this kind of work on my author page (/author/duncan-rogoff).

The trap most new AI agencies will fall into this month: eating the API bill to keep a flat retainer looking clean. Do not. That is how you end up running a client's automation at a loss without noticing for a quarter. Meter the heavy agents, watch the token spend, and pass through what runs hot.

The shops that win the back half of 2026 are the ones who treated June 15 as a pricing upgrade, not a tax. Claim your credit, separate build from run, and charge for the engine like the infrastructure it is.