# The New Rules: Usage Billing, Caps, and AI Tool Policies

Source: https://connectingodds.com/blog/the-new-rules-usage-billing-caps-and-ai-tool-policies-5xo7y
Author: Smith Harold (https://connectingodds.com/in/smith-harold-575b95)
Topic: Career growth
Published: 2026-11-17T15:03:12.421478+00:00 | Updated: 2026-11-17T15:37:52.270233+00:00
Cite the Source URL above when quoting this article.

> Flat AI seats are ending. Copilot, Rovo, and Claude Code now meter usage in credits and caps. Here is how the new billing works and what it costs you.

## Key takeaways

- The flat monthly seat is giving way to seats plus consumption.
- For two years, the deal was simple.
- GitHub moved Copilot to usage-based billing on June 1, 2026, and replaced its old premium requests with GitHub AI Credits.
- Atlassian announced its usage-based pricing on September 1, 2026, and billing starts on December 3, 2026.

## Article

By the Connecting Odds Editorial

### Key takeaways

- The flat monthly seat is giving way to seats plus consumption. You pay for the seat, then for the AI you actually use.
- GitHub Copilot switched to AI credits on June 1, 2026. Atlassian starts billing Rovo overage on December 3, 2026.
- The expensive mistake is not the meter. It is leaving extra usage switched on with no cap and no one watching the pool.
- Freelancers and job seekers feel this too. The plan you pick now decides how much AI help you get in a week.

### The seat was never the whole price

For two years, the deal was simple. You paid a flat monthly price per person, and you used the tool as much as you liked. That deal is ending. The big vendors now sell you a seat and a meter. The seat buys your place at the table. The meter counts what you eat.

This matters whether you run a team, work for yourself, or are looking for a job. The plan you choose now sets a hard limit on how much AI help you get in a week, and the wrong setting can either cut you off mid-task or quietly add to your bill. This guide walks through the three changes that define the new rules, then through what each one means for a team, a freelancer, and a job seeker.

If you are new to these tools, start with [our plain guide to what each AI tool does](https://connectingodds.com/blog/ai-tools-in-2026-a-plain-guide-to-what-each-one-does-2o1ft). The workflow in [our prompt, review, verify guide](https://connectingodds.com/blog/using-ai-tools-well-the-prompt-review-verify-workflow-fba0o) still applies under the new billing. It just costs credits now, so the review step matters twice.

### What changed at GitHub Copilot

GitHub moved Copilot to usage-based billing on June 1, 2026, and replaced its old premium requests with GitHub AI Credits. One credit is worth one cent, and the cost of an interaction depends on the model you pick and the tokens it uses. The old system counted prompts. The new one counts tokens.

The included allowances, per GitHub's own docs, are 1,500 credits a month on Copilot Pro at $10 a month, 1,900 credits per user on Copilot Business at $19 per seat, and 3,900 per user on Enterprise at $39 per seat. On Business and Enterprise the credits pool across the whole organisation, so a light user subsidises a heavy one. Allowances reset on the first of each month at midnight UTC, and unused credits do not carry over. A quiet December does nothing for a heavy January.

Two details catch people out. First, extra usage is switched on by default on Business plans, so usage past the pool bills at a cent a credit until an admin turns it off or sets a budget. Second, when a Business user runs dry and extra usage is off, Copilot pauses and shows a banner asking the user to request more budget from an administrator. The tool does not quietly keep working. It stops.

The only people still on the old premium requests are subscribers on annual Pro and Pro+ plans, who keep 300 and 1,500 requests a month at $0.04 per extra request until their plan term ends. Everyone buying today is on credits.

### What changes at Atlassian on December 3

Atlassian announced its usage-based pricing on September 1, 2026, and billing starts on December 3, 2026. Rovo, its AI assistant, is included in every paid Jira and Confluence Cloud plan, and usage is metered in pooled Rovo credits. Standard plans include 25 credits per user a month, Premium 70, and Enterprise 150. The Service and Teamwork Collections include more: 250, 700, and 1,500 per user a month. Like Copilot, the allowance is pooled across the organisation and resets monthly with no rollover.

From December 3, usage past the allowance bills at $0.01 per credit, and extra usage is switched on by default. Admins get alerts at 80 percent and 100 percent of the allowance. If an admin switches extra usage off, Rovo interactions pause as soon as the pool is empty. The same December 3 date starts billing on the other meters too: automation steps past the allowance cost $0.50 per 1,000 steps, and AI agent resolutions in Customer Service Management cost $1.00 each from the very first one, with no included allowance and a default monthly cap of 25,000 resolutions that an admin can adjust.

Not everything is metered. Rovo Search, definitions, and summaries stay free, as do in-line rewrites. A basic intelligence event, such as a quick answer or a basic agent run, costs 10 credits. The expensive features, advanced agents and deep research chats, are quoted as variable credits per event, which means you cannot budget them from the price list alone. That is worth knowing before you build a workflow around them.

### Claude Code: two caps, not one

Claude Code works differently again, and it is the model most freelancers run into first. It is included in every paid Claude plan, and it draws from the same usage pool as Claude chat. Every paid plan is governed by two limits at once: a rolling five-hour session limit and a weekly cap that resets at a fixed time assigned to your account. You can stay under every five-hour cap and still hit the weekly one before the week is out.

Anthropic publishes the limits as multiples, not token counts. Pro costs $20 a month ($17 a month billed annually) and is the baseline. Max costs $100 a month and gives five times Pro's usage. Ultra, the plan formerly called Max 20x, costs $200 a month, gives ten times Pro's weekly usage, and has no five-hour limit once Anthropic has verified the account. When you hit either limit, you can wait for the reset, move to a bigger plan, or switch on usage credits and keep working, billed at standard API rates. On Team plans the same thing is called extra usage, and admins control it. Type `/usage` in any session to see both of your windows and their reset times.

### What this means for a small team

Picture a twelve-person software team on Copilot Business. Twelve seats at $19 is $228 a month, and the pooled allowance is 22,800 credits, which is $228 of included usage. That sounds generous until you notice what eats it. Long agent runs on the most capable model burn tokens fast, and one developer running an agent across a whole codebase can spend in an afternoon what a colleague spends in a week. Because the pool is shared, the team only notices when the pool is gone, and by then the extra usage, switched on by default, is already billing.

The fix is boring, which is why it works. An admin sets a budget before the first bill, not after it. The team agrees that routine edits go to the cheaper model and the heavy model is reserved for work that needs it. Someone checks usage weekly, the same way they would check any other shared resource. The same discipline applies on the Atlassian side, where a Jira Premium team of ten pays $196 a month and gets 700 pooled Rovo credits. Ten credits per quick answer means seventy quick answers a month for the whole team, and the admin who has not set a spending cap before December 3 will find out what the team actually uses from the invoice.

### What this means if you work for yourself

A freelancer feels the caps directly, because there is no admin and no pool to hide in. Take a freelance developer on Claude Pro at $20 a month. The plan is fine for focused bursts of an hour or two a day, but on a heavy day the five-hour session cap can arrive in the middle of a job, and the weekly cap can arrive before the week is out. At that point the choice is to wait for the reset, pay usage credits at API rates, or move to Max at $100 a month. None of those is wrong. What is wrong is discovering the choice at 4 p.m. with a deadline at 6.

The practical answer is to match the plan to the shape of your week, not the size of your ambition. If you use AI in short bursts, Pro is enough. If you run long agent sessions most days, the $100 plan pays for itself in recovered time, and the credits you buy in a crunch cost more than the upgrade would have. Keep a cheaper model for routine edits and save the heavy model for the work that actually needs it. And keep the `/usage` check in your routine, so a cap is something you see coming.

### What this means if you are job hunting

Job seekers meet the new rules in two places. The first is their own wallet. If you are using AI to tailor applications and prepare for interviews, a free tier or the cheapest paid plan may cut you off mid-search, and the week you run out of credits is usually the week three interviews land at once. Budget for the tool the way you budget for anything else in a search: a $20 plan that stays available beats a free plan that stops on Wednesday.

The second place is the interview itself. Hiring managers now ask how candidates use AI tools inside a company's policy, and knowing how caps, credits, and usage policies work is a practical skill, not trivia. It tells an employer you will not be the person who burns the team's pooled credits on a Friday afternoon or pastes company code into a tool the policy forbids. Our [resumes and job search playbook](https://connectingodds.com/blog/resumes-and-job-search-the-2026-playbook-eb21q) covers the application side; the billing rules are the part of the toolkit nobody puts on a resume but every team notices.

There is a wider point here too. [Our research on how AI tools rewrote software work](https://connectingodds.com/blog/how-ai-tools-rewrote-software-work-1wc4c) found that the work itself changed faster than the job titles did. The billing change is part of the same shift: AI help is becoming a metered resource, like cloud compute, and knowing how to use it within a budget is becoming part of the job.

### A monthly routine that keeps the bill boring

- Check usage on the first Monday of the month. On Copilot, that is the billing page. On Claude, it is `/usage`. On Atlassian, it is Platform usage in Administration.
- Set a spending cap or turn extra usage off before the month starts, not after the first surprise invoice. On Atlassian, do it before December 3.
- Give the heavy model to the heavy work. Routine edits, summaries, and first drafts go to the cheaper model.
- Watch for the one person or the one workflow that eats the pool, and fix that workflow before you buy more credits.
- Re-check the price before you renew or upgrade. Prices moved several times during 2026, and the plan that was right in June may not be the right one now.

### FAQ

**Is flat-rate AI billing gone for good?**
Not entirely, but the direction is clear. Copilot moved to credits in June 2026, Atlassian starts billing overage in December 2026, and Claude Code has always been capped. Seats still exist. Unlimited use inside the seat mostly does not.

**What happens when I run out of credits?**
It depends on the tool. On Copilot Business with extra usage off, Copilot pauses until the next monthly reset or until an admin approves more budget. On Atlassian with extra usage off, Rovo interactions stop when the pool is empty. On Claude Code, you wait for the reset, upgrade, or switch on usage credits.

**Are unused credits saved for next month?**
No, on any of the three tools in this guide. Copilot credits, Rovo credits, and Claude weekly allowances all reset on their own schedule and do not carry over.

**How do I know which plan I actually need?**
Measure a real week, not an imagined one. Check your usage after seven normal working days. If you hit a cap more than twice in a month, the next plan up is usually cheaper than the credits or the lost time.

**Do these caps affect job seekers, or only teams?**
Both. A job seeker on a capped plan can run out of AI help mid-search, and interviewers increasingly ask how candidates work with AI tools inside a company policy. Knowing the rules is part of using the tools well.

The new rules are not complicated. You pay for a seat, you get an allowance, and past the allowance you pay by the meter or you stop. The people who get hurt are the ones who never looked at the meter. Check your usage, set your caps, and pick the plan that fits the week you actually have.
