Webinar Series: Usage-Based Billing with Confidence

Now that GitHub Copilot has moved to a usage-based billing model, we know many users are thinking carefully about what this change means for themselves, their teams, their budgets, and the way they support developers at scale.

To help you navigate this shift, we're hosting Usage-Based Billing with Confidence, a two-part workshop series covering both the billing platform controls available to manage spend and the agent quality and token optimization strategies that drive efficient usage.

Sessions are scheduled across multiple time zones to be globally inclusive, and on-demand recordings will be available for anyone who registers.

Workshop 1: GitHub Billing Platform Controls

How to monitor and manage Copilot spend without slowing innovation. You'll learn how to:

• Understand how token-based, usage-driven billing works
• Plan budgets across enterprise, team, and individual levels
• Apply governance through layered budget controls and guardrails
• Monitor usage in real time and prevent unexpected spend
• You'll leave with a practical framework for managing GitHub Copilot usage — so your teams can move fast while you stay in control of costs.

Workshop 2: Agent Quality & Token Optimization

Token cost is only the surface problem — the real challenge is agent quality. If your agent misses, your token strategy has already failed. This session digs into:

• How LLMs, harnesses, and the context window affect agent behavior
• The relationship between token usage and output quality
• Controls you can employ to improve agent quality, optimize token usage, and reduce spend

Who this is for:

Whether you're an engineering leader, platform team member, administrator, or someone helping your organization manage Copilot adoption, these workshops are meant to help you feel more informed and confident.

Register here:

• Usage Based Billing with Confidence: GitHub Billing Platform Controls
• Usage Based Billing with Confidence: Agent Quality & Token Optimization

We hope you'll join us, bring your questions, and leave with a clearer understanding of how to set your organization up for success.

[EuSouVoce]

Workshops are not the answer....

[ghlef]

عن طريق الخطأ

[mqhe2007]

You seem to be saying: It's you idiots who don't know how to use it.

[SteveZMTstudios]

However, while you were reducing the quota, you were also adding a full-screen Agent mode to VSCode.

[mickeyd37-beep]

So, I signed up for the standard monthly subscription, provided my pay information, now I'm NOT subscribed? WTH?

[Aminb3]

So, I signed up for the standard monthly subscription, provided my pay information, now I'm NOT subscribed? WTH?

[Soobeq]

The whole point of Copilot was precisely that no one had to worry about token usage, etc. You keep claiming that we don’t know how to use agents, that our queries are suboptimal, and that we can’t manage our budget. Well, your only advantage over others was that none of us wasted time on that. But as usual, when you have something good, you manage to completely ruin it. You really aren’t the only ones, and in this case, you aren’t competitive at all. Developers will now have to flip a coin with every prompt to see how many tokens Copilot demands.
All you had to do was add a tool that would inform users that this prompt is using too many resources and the wait time for a response will be longer, while keeping the previous billing model. But you preferred to get your hands on the cash.

[ForgeMechanic]

I think the core issue for me isn't the price of AI Credits. It's that I don't really understand the unit I'm being billed for.

As developers, we're used to billing models that expose measurable units. In Azure I can see vCPU hours, storage consumed, transactions, executions, requests, bandwidth, etc. Even if the pricing model is complicated, I can usually build a mental model of what drives cost and make informed decisions. With AI Credits, I don't have that understanding yet.

A few questions that would help:

1. What actually defines a credit?
2. Is it based on tokens, requests, agent iterations, tool calls, or some combination?
3. Are failed requests charged?
4. Are retries charged?
5. Are reasoning tokens weighted differently?
6. Do 1,000 small requests cost more or less than one large request with the same token volume?
7. How do different models affect credit consumption?
8. Is there a predictable relationship between model usage and credits that developers can use for planning and optimization?

To use an Azure comparison, imagine if Virtual Machines were labeled only as "Low Cost", "Medium Cost", and "High Cost" without exposing the underlying pricing units. You could still use the service, but it would be pretty difficult to make cost-effective decisions.

I don't need an exact prediction for every request. I just need enough transparency to understand what actually drives credit consumption so I can use the platform intentionally and avoid surprises.

[ForgeMechanic]

Also, I really don't need a webinar to tell me how billing works. I need it in writing, because that's how my management will expect it.

[richardchen874-sys]

I think the core issue for me isn't the price of AI Credits. It's that I don't really understand the unit I'm being billed for.

As developers, we're used to billing models that expose measurable units. In Azure I can see vCPU hours, storage consumed, transactions, executions, requests, bandwidth, etc. Even if the pricing model is complicated, I can usually build a mental model of what drives cost and make informed decisions. With AI Credits, I don't have that understanding yet.

A few questions that would help:

1. What actually defines a credit?
2. Is it based on tokens, requests, agent iterations, tool calls, or some combination?
3. Are failed requests charged?
4. Are retries charged?
5. Are reasoning tokens weighted differently?
6. Do 1,000 small requests cost more or less than one large request with the same token volume?
7. How do different models affect credit consumption?
8. Is there a predictable relationship between model usage and credits that developers can use for planning and optimization?

To use an Azure comparison, imagine if Virtual Machines were labeled only as "Low Cost", "Medium Cost", and "High Cost" without exposing the underlying pricing units. You could still use the service, but it would be pretty difficult to make cost-effective decisions.

I don't need an exact prediction for every request. I just need enough transparency to understand what actually drives credit consumption so I can use the platform intentionally and avoid surprises.

This is exactly the missing piece for many developers and teams.

The problem is not only usage-based billing itself. Developers can work with usage-based pricing when the unit is understandable and measurable.

With AI agents, the cost driver is harder to reason about because one visible request may include:

• input tokens
• output tokens
• reasoning tokens
• tool calls
• agent iterations
• retries
• hidden repo context
• model-specific multipliers
• failed or partially failed runs

Without a clear mapping between those drivers and credits, it is hard for teams to plan usage or explain costs to management.

I agree that written documentation matters more than a webinar here. Teams need something they can reference later: what is counted, how retries are billed, how models differ, what controls exist, and how to estimate cost before a run starts.

For agentic coding, I think the most useful metric would be cost per completed task, not just credits consumed per prompt.