Skip to content

Light version: pages open quickly, without the 3D graphics. You are seeing the light version because the 3D graphics would be slow on this device. The content is the same.

Turn on 3D

Preparing the graphics. This can take a moment on the first visit.

Henry Lõiv

Web design & development

Artificial Intelligence 7 min read

Why does Claude Code hit usage limits so quickly?

Understand Claude Code usage limits, long context and tool overhead. Use clearer tasks, useful handoffs and focused checks to reduce wasted work.

A short request can trigger a substantial amount of work in Claude Code. The assistant may read files, run commands, inspect their output and try again before replying. Your typed message is only one part of that activity.

Before changing plans, identify the limit you have reached and the work consuming it. The most useful first step is to inspect usage, narrow the next task and compare what happens over one bounded piece of work.

Identify the limit before trying to fix it

Claude distinguishes usage limits over a period of time from the amount that fits in a conversation. Anthropic currently describes five-hour session allowances and weekly limits for paid plans; usage across Claude's product surfaces can draw from the same allowance. The displayed account information is the place to check your applicable limits and reset times. Usage and length limits.

What you seeWhat it means for your next step
A plan usage limit and reset timeInspect the allowance and other activity on the account. Starting a new conversation does not create a new subscription allowance.
A long conversation or compaction noticeReview the amount of context carried into the next task. This is a different issue from the billing period.
An API rate-limit errorCheck the API organisation or provider's limit and the actual error. A subscription reset may be unrelated.
A dollar estimate in the sessionEstablish whether it is a usage estimate or an actual billed charge.

Inside Claude Code, start with:

/usage
/status

The usage screen helps you inspect consumption; the status screen identifies the active session setup. Command details depend on the installed release. Use the current command reference if your interface differs.

If the concern is an unexpected invoice, use the separate guide to Claude Code API charges with a subscription. Changing how you manage context does not resolve an account using the wrong billing route.

Why one request can consume so much

Claude Code works through a tool loop. “Fix the checkout” can lead to searches, file reads, edits, tests and further model requests. The amount of work depends on what the assistant encounters, not just how many words you type. How Claude Code works.

A vague task also leaves the stopping point unclear. Consider the difference between “improve the website” and “find why the contact form's confirmation disappears after navigation”. The second request gives the investigation a boundary and a visible outcome.

Long sessions add another factor. Earlier messages and tool results can continue contributing to later requests. Anthropic's cost guide also identifies cache misses and scheduled activity as reasons consumption can rise unexpectedly. A cache can reduce the cost of reused context; it does not make that context free. Claude Code costs.

This is why a short follow-up after a long debugging session can be more expensive than the same question in a clean, relevant context.

Narrow the investigation without hiding important context

Suppose a WordPress service card has a broken price label. Asking the assistant to inspect the whole project, improve the design and modernise the code creates several unrelated jobs.

A more useful brief is:

The service card shows a blank price for the free inspection.
Expected: a zero price displays as Free; a missing price is hidden.

Start with the service-card template and the price-formatting code.
Trace additional files only if the evidence points to them.
Keep the design and unrelated pages unchanged.

After fixing it, check paid, free and missing-price examples.
Report the files changed and those three outcomes.

This brief still lets the assistant follow the evidence. It also defines when the work is complete. Use the same approach for a failing form, an overflow bug or an incorrect translation: name the observed behaviour, expected behaviour and smallest useful verification.

Anthropic's best-practice guidance recommends specific context and a way to verify work. The point is to reduce avoidable searching and rework while retaining the checks that establish correctness.

Keep command output relevant

A build log can contain thousands of lines while the useful error occupies a handful. When investigating, save the full log locally and bring the relevant failure into the conversation. If more context is needed, inspect the surrounding section.

For example, a failed test report should identify the failing case, expected result, actual result and relevant trace. Repeatedly dumping every successful test into the conversation adds material without answering a new question.

Do not suppress errors simply to make the transcript smaller. Keep the complete evidence available and reduce the amount repeatedly sent to the model. A concise summary should point to the saved output when it matters.

Review instructions and connected tools

Project instructions should describe stable facts: how to run the application, important constraints and where the relevant systems live. A long history of old tasks can make each new investigation harder to orient.

Claude Code supports project instructions and more targeted rules. Keep necessary instructions; move detailed material to an appropriate reference and make it easy to find when needed. Project memory and instructions.

Connected tools can add useful capabilities, but their definitions and returned data also affect context. Inspect which MCP servers the task actually needs. A small CSS fix may need the browser and repository, while a CRM migration has different requirements. MCP configuration.

Use the tools needed to verify the result. Disabling the browser just to reduce usage can be a false saving if the task is a visual bug that remains unchecked.

Start a new task with a useful handoff

When the work changes substantially, consider a fresh session with a compact record of the current state. Include what is finished, the files involved, the remaining problem and the checks already run.

For example:

The service price bug is fixed and checked for paid, free and missing values.
The next task is mobile spacing in the same component.
Relevant files: the service-card template and its stylesheet.
Preserve the price behaviour. Check the card at 390 px and 1440 px.

Compaction and clearing serve different purposes. Compaction keeps a condensed continuation; clearing starts a fresh conversation context. Neither replenishes a spent plan allowance. For the detailed distinction, see Claude Code /compact versus /clear.

Before switching, preserve decisions that would otherwise need rediscovering. Starting from nothing after every message can create its own repeated work.

Measure one change before upgrading

Choose a typical task and record the starting usage display, model, relevant settings and active sessions. Finish the task, then record the ending display and review effort. Avoid other account activity during the comparison if you want to isolate that task as far as possible.

On the next comparable task, change one thing: a tighter brief, less irrelevant output or a better handoff. Treat the result as a practical observation. Different task difficulty, caching and account activity can affect it, so a single run is not a reliable savings percentage.

If your normal workload still reaches the allowance after this cleanup, a higher plan or a separately budgeted API workflow may be worth comparing. Check its actual terms and spending controls first. More capacity is useful when it supports necessary work; it does not repair an unclear task or an endless correction loop.

The goal is a completed, verified change with less wasted work. Keep the tests that matter and make the assistant's next decision easier to make.

Putting AI to work on your website?

Choosing a model is one part of a working feature. If your business needs a support assistant, content workflow or connection to an existing system, I can help define and build the interface, integrations and checks it needs.

Explore AI integrations for websites and applications⁠

Written by Henry Lõiv

Keep reading

All articles⁠
  1. Best AI agents in 2026: Grok Bot vs Dots and EU availability Compare AI agents by the work they can handle and whether you can use them in Estonia or the EU. Artificial Intelligence 5 min read
  2. AI Agent, Chatbot or Automation: What Does Your Business Need? A chatbot is an interface; an agent is one way to handle the work behind it. Follow a fictional customer enquiry to choose a practical system for your business. Artificial Intelligence 8 min read
  3. Claude Code vs Codex for Frontend Development: A Practical Comparison A documentation-based comparison with a runnable frontend exercise and a blank scorecard. No measured winner is claimed. Artificial Intelligence 9 min read

Hi!

I really love the sea, yachts, sunsets and starry skies.

Through this little 3D world, I can share a small part of myself with you.

Thank you for opening this bottle and reading my letter.

Henry

Let’s talk about your project

Tell me about your project

Tell me what you’d like to build or improve. You don’t need a finished brief — we’ll work out the scope and next steps together.

01 How can I help?
02 How many pages do you need?
03 Which extras do you need? Choose all that apply
04 When does it need to be ready?
05 Project details
06 Your contact details

Enquiry received

Thank you!

Your enquiry has been received. I usually reply by email within 2 working days.