The Opt-In Trap

You're paying $20–$200/month for Claude. On June 15, your programmatic access moves to a separate credit pool. But here's the part nobody is talking about: you have to claim those credits yourself. If you don't, your agents, scripts, and CI pipelines just stop.

The Detail Buried in the Announcement

Most coverage of Anthropic's June 15 credit split focuses on the credit amounts: $20 for Pro, $100 for Max 5x, $200 for Max 20x. The framing is "same price, new system."

Here's what got less attention:

Credits must be claimed before June 15 by following instructions Anthropic will send to your account email.

Let me translate: you already pay for Claude. On June 15, Anthropic is splitting your existing subscription into two buckets — one for interactive use, one for programmatic use. But the programmatic credits don't flow automatically. You have to opt in to receive something you're already paying for.

This isn't a technical requirement. It's a behavioral one.

Why Opt-In Is a Dark Pattern

In behavioral economics, there's a well-documented principle called default effect: people stick with whatever is pre-selected. Companies use this all the time — pre-checked newsletter boxes, auto-renewing subscriptions, default data sharing.

When a company makes you opt in to something instead of opt out, they know a percentage of customers won't do it. That's not a bug. It's a feature.

What happens to developers who:

Their AI agents stop working. Not because they didn't pay. Because they didn't claim what they already paid for.

The Math: $20 in Credits vs $500 in Value

Before June 15, a $20/month Pro subscription gave you unlimited (within rate limits) access to Claude through third-party tools. Developers built workflows worth hundreds of dollars in API-equivalent value.

After June 15, that same $20 gets you:

At Sonnet 4.6 pricing, $20 covers roughly 6.6 million input tokens. That sounds like a lot. But agentic workflows with large context windows burn 100,000–200,000 tokens per session. That's 33–66 sessions per month. If you run Claude agents daily, you'll exhaust your credits in about two days.

Theo Browne, founder of T3.gg, quantified it: an effective 25x price increase for anyone using third-party tools.

Three Vendors, Same Playbook

This is now a pattern across the AI industry:

GitHub Copilot (June 1): Removed free fallback models. Retired annual plans. Started dual-surface billing (AI Credits + Actions minutes). Free models "no longer part of the offering."

Anthropic (June 15): Split subscriptions into interactive and programmatic credits. Required opt-in to claim credits you already paid for. Boris Cherny: third-party tools are "unsustainably" expensive.

Google Gemini: Weekly usage limits that lock you out for days. Hardware gating — Pixel 9 Pro excluded from Google's own AI features. Three-tier pricing ($0, $20, $250).

Three vendors. Same month. The same playbook: reduce value, increase cost, shift the burden to the customer to notice.

The Quiet Kill

The opt-in mechanism is particularly clever because it creates plausible deniability.

Anthropic can say: "We gave everyone credits! Same price! We emailed you!"

What they won't say: a predictable percentage of users won't claim those credits. Those users' agents will stop working. Some won't notice for days. Some will assume it's a bug. Some will quietly switch to API keys and start paying per-token — which is exactly the outcome Anthropic wants.

The opt-in isn't a mistake. It's a conversion funnel from subscription to pay-per-use.

14 Days to Decide

If you use Claude through OpenClaw, Conductor, Zed, or any third-party agent, you have until June 15 to:

  1. Claim your Agent SDK credits — watch for Anthropic's email
  2. Audit your programmatic usage — you may be burning through $20 faster than you think
  3. Consider what "owning" looks like — because the opt-in trap is just the beginning

The landlords are tightening the lease terms. First they split the payments. Then they make you claim what you already bought. Next comes the rate increase you can't opt out of.

What Owning Looks Like

The alternative to this shell game isn't finding a better landlord. It's owning the property.

The opt-in trap is a symptom. The disease is renting something you should own.

New framework: The June 15 split is just the Tier 2 squeeze. There's a Tier 3 above it (Mythos, government-gated) and a Tier 1 below it (Own, BYOK). Read the full three-tier framework →