The world of software development is undergoing a fundamental transformation. What started as a simple code autocomplete has, in recent months, evolved into complex systems capable of independently planning, writing, and testing entire modules. It is this shift from passive models to active AI agents that forced GitHub to intervene in response to a capacity crisis.
Why is Copilot "hitting a wall"? The phenomenon of agentic AI
To understand the situation, it is necessary to explain the difference between a standard model and an agentic system. While a typical LLM (Large Language Model) responds to your query with a single output, agentic AI operates in a continuous loop. The agent sets a goal, breaks it down into steps, calls tools (e.g., terminal or browser), checks the results, and attempts to correct errors if they occur.
According to reports from InfoWorld, these processes lead to so-called parallel and long-running sessions. Instead of one short query, the agent generates hundreds of operations in the background to complete a complex task. This creates enormous pressure on the GPU infrastructure and computational capacity that GitHub originally did not account for with Copilot.
Analyst Charlie Dai from Forrester stated that this shift makes the workload much less predictable. This is a critical problem for cloud providers who must guarantee low-latency service availability.
Changes in plans and restrictions on model access
GitHub has taken several drastic steps to address the situation. First, it suspended registrations for Copilot Pro, Pro+, and Student plans. The goal is to ensure that existing users have sufficient performance and do not experience outages caused by overload.
In addition to restricting new users, there is a restructuring of available models within existing subscriptions:
- Pro Plan: Significant restrictions are being introduced here. Opus series models will no longer be available in this tier.
- Pro+ Plan: This tier is now positioned as the primary choice for demanding users. It offers more than five times the limits compared to the Pro version. The Opus 4.7 model will remain available here, while versions 4.5 and 4.6 will be gradually removed.
It is important for developers to monitor how these changes affect their workflow. GitHub has promised that usage limits will now be directly visible in VS Code and Copilot CLI, allowing for better work planning.
Comparison with competitors: Is GitHub not alone?
This trend is not unique to Microsoft and GitHub. Anthropic also had to take similar steps. As reported in a report on X, Anthropic recently adjusted how time limits work for the Claude model during peak times to handle user demand without a complete system collapse.
If we compare the capabilities of agentic systems within the most common models, the situation looks as follows:
| Model / Family | Agentic capabilities | Main advantage |
|---|---|---|
| GitHub Copilot (Opus 4.7) | Very high | Direct integration in IDE and workflow |
| Claude (Anthropic) | High | Excellent logical reasoning and reasoning |
| GPT-5 / OpenAI | High | Broad ecosystem integration |
Practical impact for Czech developers and companies
What does this mean for a Czech programmer or a smaller development company in the Czech Republic?
- Pricing policy: GitHub Copilot is also available for the Czech market. Prices typically range around 10 USD (approx. 230 CZK) for the individual version and higher amounts for Pro/Pro+ versions (approximately 20-50 USD). It is necessary to consider that for truly intensive agentic work, investment in higher tiers will be required.
- Availability: The tool is fully available in Czech in terms of interface (via VS Code), but interactions with AI itself primarily take place in English, which is standard for the global developer scene.
- EU regulation: Given that GitHub falls under Microsoft, it must comply with the strict rules of the EU AI Act. Capacity limitations may also be a way to ensure that systems operate in accordance with the reliability and security requirements that the EU places on high-risk systems.
If you are an existing user of the Pro or Pro+ plan and the new terms do not suit you, GitHub offers the option to request a refund. This option is open to users who contact support between April 20 and May 20, 2026. If you decide on a refund, you will not be charged for usage in April either.
Can I purchase Copilot Pro now, even if registration is suspended?
No, GitHub has currently stopped new registrations for Copilot Pro, Pro+, and Student plans. If you already have a subscription, your services should work, but with new limits.
How do I know if I'm close to the limit for agentic AI?
GitHub has implemented limit tracking directly into your development environment. You will see them directly in the VS Code editor or in the Copilot CLI, so you can switch to a different model or plan in time.
Is Copilot Pro+ available for Czech companies with a lower budget?
The Pro+ plan is designed for users with high computational demands. For small Czech companies or individuals, it is still more economical to start with the individual version if they do not require continuous runs of autonomous agents.