Are You Using Your Claude Subscription to Its Full Potential?

If you pay for a Claude subscription — Pro at $20 a month, or Max at $100 to $200 — here's an uncomfortable question: how much of it are you actually using? Your capacity arrives in five-hour windows that reset around the clock, whether you're at the keyboard or asleep. Use a window or lose it. Nothing rolls over.

Most of us use Claude Code the same way: type a prompt, watch it work, repeat, roughly 9 to 5. That leaves entire windows untouched every day. This post covers the three upgrades that change the math — plugins, loops, and putting your subscription on the night shift — with specific tools you can install today.

The math your subscription does while you sleep

A quick refresher on how limits actually work. Every plan has a five-hour session window plus a weekly cap, and every Claude surface — claude.ai chat, the desktop app, Claude Code — draws from one shared pool. Anthropic doesn't publish exact token quotas; plans are multipliers, with Max giving you 5x or 20x Pro's usage.

Here's the part that matters: windows reset on the clock, not on your schedule. If you work 9 to 6, you touch maybe two windows a day. The overnight windows, the early-morning windows, most of the weekend — capacity you've already paid for, expiring quietly in the background.

You can't bank it. But you can fill it, and the rest of this post is about how.

Plugins: the upgrade most subscribers never install

Before filling more hours, make the hours you already use count for more. That's what plugins are for. A Claude Code plugin is a one-command install (/plugin) that bundles slash commands, agents, skills, hooks, and MCP servers — job-specific instructions and tools that make Claude dramatically better at a particular kind of work.

A few that earn their spot:

• Superpowers — enforces real engineering discipline: test-driven development, a debugging methodology, planning before code. It's the difference between an intern and a colleague.
• Language server plugins — connect Claude to LSP servers for Python, TypeScript, Rust and more, so it sees live diagnostics instead of guessing at errors.
• Context7 — pulls current library documentation at request time, patching the knowledge-cutoff problem when you're on a fast-moving framework.
• Frontend Design — gives Claude a concrete aesthetic direction so your UIs stop looking like every other AI-generated interface.

That's four of a catalog of thousands, including official Anthropic plugins and partner plugins from GitHub, Playwright, Supabase, Sentry, and Stripe. Finding the good ones is genuinely a discovery problem now — it's the reason I built PlugMyPlugin (full disclosure: this blog lives there), which shows you what each plugin is actually made of and what it needs before you install it.

"My job is to write loops"

At YC Startup School 2026, fresh off the Opus 5 launch, Claude Code's creator Boris Cherny said the line everyone clipped:

"I don't prompt Claude anymore. I write loops and the loops do the work. My job is to write loops."

A loop is not a clever prompt. It's a task plus a way to verify the result, so the model can check its own work and keep going instead of stopping to ask you. "Make the tests pass" is a loop. "Fix the flaky CI job and prove it with ten green runs" is a loop. "Refactor this module" is not — there's nothing to verify, so it stalls or drifts.

Cherny's point from the same talk: give the model the task and a way to verify the output, and it'll just go. He described Opus 5 running in auto mode for "days, weeks, months" without external scaffolding. The bottleneck stopped being the model a while ago. It's how much verifiable work we hand it.

Put your Claude subscription on the night shift

Now combine the two ideas. You have expiring overnight capacity, and you have a way to define work that doesn't need you in the room. Claude Code can run recurring loops on an interval and schedule agents to run on a cron schedule — which means your subscription can work hours you don't.

Off-hours jobs that work well, because each one has built-in verification:

• Nightly code review — review the day's diff and open annotated findings before you wake up.
• Test-coverage patrol — find the least-covered module, write tests, run them, commit only what passes.
• Dependency triage — check for updates, read the changelogs, apply the safe ones, and report the risky ones.
• Docs drift — compare README and docs against the actual code and queue fixes for anything stale.

The pattern in every case: the loop produces something checkable — passing tests, a diff, a report — and you review the output over coffee instead of babysitting the process.

A starter playbook for this week

1. Install two plugins that match your stack — one discipline plugin (like Superpowers) and one language server. Five minutes.
2. Write one loop. Pick a task with a verifiable finish line — a failing test suite, a lint backlog — and hand it over with the success condition spelled out.
3. Schedule one off-hours job. Start with nightly code review; it's low-risk and immediately useful.
4. Check your usage after a week. Same subscription price, more shipped work — that's the whole trade.

The subscription isn't the ceiling

The gap between "types prompts during work hours" and "runs verified loops around the clock" is enormous, and it costs nothing extra — it's the same Claude subscription either way. Plugins raise the quality of every window you use. Loops let windows run without you. Scheduling spends the windows you used to waste.

Start with one plugin and one loop tonight. If you want to see what's out there first, browse the plugin directory — every listing shows what a plugin actually does and what it plugs into before you commit.