Use Your Claude Subscription for Coding Agents
Anthropic draws a line between interactive and programmatic Claude usage. Your Pro or Max subscription covers the interactive side - sessions where a human is driving `claude` in a terminal. Per Anthropic's documentation: "Starting June 15, 2026, Claude Agent SDK and claude -p usage no longer counts toward your Claude plan's usage limits." That sentence runs both directions - interactive terminal usage still uses your plan; Agent SDK calls now draw from a separate pool.
For most developers using terminal-based agent managers, nothing changes on June 15. The tools that matter are the ones routing Claude through the API invisibly - primarily IDE-based tools like Cursor and Windsurf, plus any Agent-SDK automations or claude -p scripts in CI. Those tools will hit a new monthly credit cap ($20 on Pro, $100 on Max 5x, $200 on Max 20x) before falling to pay-as-you-go API rates.
tools that use your subscription
These tools spawn the local claude CLI as an interactive subprocess - indistinguishable from a human running claude in a terminal. Unaffected by the June 15 change.
- → Pane - cross-platform terminal-native agent manager, open source
- → Conductor - disputed. Landing page implies subscription passthrough ("uses Claude Code however you're already logged in"), but Theo Browne reports it's affected. Verify before relying on it.
- → Claude Squad - Go-based terminal manager, launches
claudeas interactive subprocess in tmux - → Vibe Kanban - "spawns the official Claude Code CLI as a subprocess, so Claude Code picks up all existing configs"
- → Emdash - Electron app spawning CLI agents as subprocess in worktrees, YC W26
- → Superset - "launches Claude Code" in isolated worktrees, interactive subprocess
- → Recon - tmux-native dashboard, uses
claude --model sonnetdirectly
tools that don't use your subscription
These tools route Claude through Anthropic's API on their backend, or invokeclaude -p headlessly. After June 15, 2026, they draw from the new programmatic credit pool before falling to pay-as-you-go API rates.
- → Cursor - routes Claude through Anthropic API, no Pro/Max passthrough; BYOK key hits the new cap
- → Windsurf - per-token API billing on their backend; BYOK same situation
- → OpenClaw - Agent SDK-based, named in Anthropic's documentation
- → Claude Code GitHub Actions - Agent SDK-based, programmatic by design
- → custom
claude -pscripts - headless mode = programmatic = new credit pool - → T3 Code - reported affected by its author Theo Browne
- → Zed - AI features call Claude through the API, not the CLI
- → Jean - reported among programmatic Claude wrappers (source)
how to verify
The fastest check: open the tool and look at what's running. If you see a liveclaude CLI session in a terminal - the same prompt you'd get from typing claude yourself - you're on your subscription. The tool is transparent.
If the tool has an API key field in settings, or invokes Claude invisibly without showing you a terminal, it's routing through the API. That's fine for some workflows - but after June 15 it means you're spending from the new programmatic pool, not your plan's interactive allowance.
further reading
For the full per-tool breakdown with source links and verdicts, see the complete guide to which AI coding tools are affected by the June 15 change.
If you're deciding between an IDE-based tool and a terminal-wrapper, the Pane vs Cursor / Windsurf comparison walks through the subscription model difference with a concrete cost example.
Pane is one of the terminal-wrapper agent managers listed above - open source, cross-platform, works with your existing Pro or Max subscription.