Build 2025: Microsoft flexed its muscles, but too late?

Microsoft Build 2025, which took place May 19–22 in Seattle, was, according to analysis by Michael Parekh, unequivocally an AI conference. Satya Nadella and the new head of the CoreAI division Jay Parikh (former top engineer from Meta) presented ambitious plans: agentic agents across the Copilot ecosystem, open-sourcing the Windows Subsystem for Linux (WSL), and most importantly full support for the Model Context Protocol (MCP) — a standard nicknamed the "USB-C of AI applications."

"It's about speed. How can we learn faster than anyone else?" Parikh said in an interview with Axios. Microsoft also announced that it would host Grok 3 and Grok 3 Mini from Elon Musk's xAI on Azure, and GitHub introduced its own AI coding agent capable of fixing bugs and adding features without developer intervention.

The problem is that all of this comes at a time when GitHub Copilot — Microsoft's key AI coding product — is losing ground.

Copilot can't keep up: suspended registrations and trimmed limits

In April 2026, GitHub suspended new registrations for individual Copilot Pro, Pro+, and Student plans. The reason? Agentic workflows, where AI works independently on long tasks, consume many times more computing power than the original pricing model had anticipated.

"As Copilot's agentic capabilities expand, agents are doing more work and more customers are hitting limits designed to preserve service reliability," GitHub stated in an official blog post. Without further measures, they said "quality of service would degrade for everyone."

In practice, this means that developers using Copilot for agentic tasks are hitting limits much faster than before. GitHub also restricted access to Opus models on lower tiers and introduced separate session limits and weekly limits. Charlie Dai, Forrester analyst, noted: "Pricing structures built on lightweight assistance no longer work. The pressure on GPU capacity, reliability, and unit economics is enormous."

Competition breathing down Microsoft's neck

While Copilot tightens limits, competitors are expanding:

• Claude Code by Anthropic — launched as a standalone agentic tool for developers, it gained popularity thanks to its ability to work in massive codebases and autonomously solve complex tasks. In May 2026, Anthropic launched Agent View — an overview of all running AI agents in one place.
• OpenAI Codex — OpenAI acquired the startup Windsurf for 3 billion dollars in May 2025 and launched Codex, its own agentic coder that works asynchronously. In May 2026, 87% of developers at Singapore-based Sea Limited used Codex at least once a week.
• Cursor — an AI-powered editor combining chat, inline editing, and agentic mode, it has become a popular alternative especially among startup developers. Its "vibe coding" model — programming using natural language — resonates particularly with the younger generation of coders.

According to Business Insider, Claude Code is one of the biggest token consumers in the industry — which paradoxically shows how intensively developers are using it. And Uber's CTO admitted in May 2026 that the company exhausted its annual budget for AI tools, including Claude Code, in just four months.

What does this mean for Czech developers?

For Czech programmers and companies, the situation is a double-edged sword. GitHub Copilot remains free for students and open-source projects, which is an advantage for academia and non-commercial development. However, the restrictions on individual plans impact freelancers and small teams who cannot afford enterprise licenses.

Claude Code and Cursor are available in the Czech Republic without restrictions — the English interface is not a barrier for developers. OpenAI Codex currently works primarily through a web interface and ChatGPT integration, so it is accessible to anyone with a subscription ($20/month for ChatGPT Plus, $200 for Pro).

Practical advice for Czech companies: if your team is hitting Copilot's limits, it's worth testing a combination of multiple tools. While Copilot excels at code completion directly in the IDE, Claude Code and Cursor handle autonomous agentic tasks like refactoring entire modules or writing tests better.

Microsoft isn't giving up — it's betting on the ecosystem

Microsoft has one weapon that the competition lacks: vertical integration. GitHub, Azure, Visual Studio Code, Windows, and Office form an ecosystem into which Copilot fits in a way that neither Cursor nor Claude Code can offer. The newly announced MCP protocol support across Copilot Studio, Dynamics 365, Azure AI Foundry, and Windows 11 has the potential to attract enterprise customers who want AI agents connected to all company systems.

Moreover, the fact that Microsoft hosts even competing models like Grok on Azure shows a pragmatic strategy: "better they run with us than on AWS or Google Cloud." Azure AI Foundry today offers models from OpenAI, Meta, Mistral, DeepSeek, and xAI — and developers can choose which model to use for a particular task.

The question is whether that will be enough. AI coding is shifting from assistance to autonomy — and in this discipline, Microsoft is currently playing catch-up rather than setting the pace.

Numbers that speak volumes

• $3 billion — the amount OpenAI paid to acquire Windsurf (May 2025)
• 87% — share of Sea Limited developers who use OpenAI Codex weekly
• 0 new registrations — status of Copilot Pro, Pro+, and Student plans since April 2026
• 5× higher limits — difference between Copilot Pro and Pro+ under the new restrictions
• 4 months — the time it took Uber to exhaust its annual budget for AI coding tools