What you'll learn at the workshop

The full-day workshop titled Mastering AI Development & Building AI Apps with GitHub Copilot takes place on July 31, 2026, as part of Visual Studio Live! @ Microsoft HQ. It will be led by two prominent Microsoft figures — Kayla Cinnamon (Senior AI Developer Tools Advocate) and James Montemagno (Principal Lead Program Manager, Developer Community).

"This is the deep dive we've been wanting ourselves. We'll show how to bend Copilot to your needs — with agents, hooks, MCP servers, and SDK — instead of just accepting the defaults," the instructors summarized in an interview for Visual Studio Magazine. The workshop is aimed at developers who already use Copilot but suspect they're only tapping a fraction of its potential.

From suggester to developer platform

When Copilot launched in 2021, it was essentially smart code completion. Today, the landscape is completely different. Copilot is present across the entire developer workflow — from task planning through coding and code review to production deployment. And that's exactly what the workshop is about: how to cover the full lifecycle of AI application development.

In the planning phase, agents and prompts help break down large tasks into manageable pieces. During coding, inline suggestions, chat, CLI, and the cloud agent come into play. Code review is handled by Copilot Code Review and skills that enforce team standards. And at deployment, the GitHub Copilot SDK takes center stage, allowing you to embed the same AI capabilities directly into your own applications.

Agents, hooks, and MCP: What these actually mean

For the average Czech developer, these terms may sound abstract, but the principle is simple:

• Agents — autonomous helpers that carry out multi-step tasks without your constant supervision. For example, "fix all TypeScript errors in the project and create a pull request."
• Hooks — triggers that react to specific events. You can set up Copilot to automatically check code against company rules before every commit.
• MCP servers (Model Context Protocol) — a standardized interface that supplies Copilot with context from external systems — documentation, API specifications, database schemas. In other words, thanks to them, Copilot understands your specific project, not just programming in general.
• Plugins and extensions — expand Copilot's capabilities with integrations into other tools like Jira, Linear, or your own internal systems.

A key theme of the workshop is precisely context. While basic Copilot only sees the current file, with MCP servers and plugins it "sees" your entire ecosystem — and its suggestions become significantly more accurate.

How to test AI features when you don't know the exact output

One of the biggest pain points of AI-driven development is testing. How do you test something that inherently generates a slightly different output each time? Kayla and James offer a practical approach: create an evaluation set — a curated collection of inputs with expected behavior (not necessarily exact output). Run it with every prompt or model change, much like unit tests.

For subjective outputs (such as documentation quality), human review comes into play. And in production, telemetry helps catch regressions on real data early. "Hooks are great here — you can plug validation directly into the Copilot workflow," they add.

The first step you can do in ten minutes

The biggest "quick win" the instructors recommend: write your own agent or repository-level instructions file. It takes a few minutes, the file lives directly in your repo, and instantly makes Copilot smarter — it knows your conventions, team standards, and codebase specifics. "It's the most effective thing you can do, and most people haven't done it yet," they say.

The elephant in the room: the new credit system

The workshop arrives at a time when GitHub is facing strong criticism for the shift to usage-based billing, which launched on June 1, 2026. Instead of a fixed number of premium requests, developers now receive a monthly allotment of AI credits, with consumption calculated based on tokens.

The result? According to community reports, many developers burned through a substantial portion of their monthly credits within the first few days. One developer on the Pro+ plan (3,500 credits per month for $39) lost 50% within two days of regular work. Another on the Max plan (7,000 credits for $100) consumed everything in under 24 hours.

For context: Copilot Pro costs $19 (1,500 credits), Pro+ $39 (3,500 credits), Max $100 (7,000 credits). After depletion, you can purchase additional credits at an extra cost. BYOK (Bring Your Own Key), meaning the ability to use your own API keys from OpenAI, Anthropic, or Google, remains an alternative — and since VS Code version 1.122, it also works without signing into GitHub.

What this means for Czech developers

For Czech companies and freelancers, Copilot still makes sense — especially thanks to Czech localization of VS Code and Czech language support in chat. With the new credit system, however, strategic thinking is required: it pays to learn the techniques the workshop teaches — particularly writing instruction files and effective prompting — so you don't burn credits unnecessarily.

An alternative for those unhappy with the credit model is BYOK mode. Simply get your own API key from OpenAI (GPT-5.5), Anthropic (Claude), or Google (Gemini) and use it directly in VS Code. This gives you full control over costs and lets you choose the model that best fits your budget.

Workshop registration is open until June 19, 2026 with a $300 Early Bird discount. A standard ticket for the full five-day VSLive! conference costs $2,399, or $2,099 with the Early Bird discount.