At GitHub Universe, Copilot Takes Center Stage, but Questions About the Platform Persist

Helps developers while they are coding with what can be best described as intelligent code auto-complete. You start writing code and ask GitHub Copilot to do its best to grok what you've written and complete the rest.

Provides developers a prompt to simply describe the code that they want Copilot to create — for example, "create a method that adds two numbers together" — or even to resolve more complex tasks such as calling APIs (for which the developer doesn't remember the signature) and service-level agreements.

Creates unit tests for both existing code and for newly generated code. It can explain code, as well as review and document code. In other words, you can look at Copilot as being your first pass of an automated code reviewer and pair programmer.

A natural language chatbot (like ChatGPT — in fact, based on OpenAI's GPT-4) inside the IDE that a developer can query for insights into the code, to provide high-level designs or API information, to generate and document code, and more.

The ability to customize GitHub Copilot on your enterprise custom code, enabling enterprises to fine-tune Copilot's capabilities to observe coding styles, libraries used, APIs, and other development norms to get tailored code suggestions (limited to the new Enterprise tier).

The ability, at the same time, to restrict access to code in your repositories from the tailored customization (Enterprise only).

Offering GitHub Copilot on GitHub.com, as well as a new partnership program that was launched for third-party partner integrations, along with a few already active such as Databricks, New Relic, Postman, Red Hat, and Docker.

Security offerings such as Dependabot, auto-triage, and secrets scanning.

GitHub projects, a workflow/issue tracker with integrated automation.

GitHub Actions on more silicon — GitHub has created runners for macOS M1 and ARM for native Actions support.

Lastly, the really interesting thing: GitHub announced migration tools for Atlassian Bitbucket (a competitor and no surprise) and Microsoft's Azure DevOps (its parent company and a big surprise).

GitHub showcased individual developer metrics but not team or organization metrics. While GitHub talks a good game for providing a platform, all the focus was on individual developers and not on team metrics, platform teams, or enterprise maintainers. And while GitHub's announcement of user interface (UI) enhancements that integrate personal and corporate GitHub accounts in the same UI got rousing applause from developers, this may cause consternation with corporate security teams.

The relationship with Microsoft appears more complicated than it should. We'd like to see GitHub address the confusing relationship that it has with Microsoft. If GitHub has tools to migrate users from Azure DevOps (ADO), what does that say about the future of ADO? It makes Azure DevOps users rightfully wonder.

There's a strong focus on the creation side of coding but not the other steps. GitHub Copilot is currently addressing developers' needs mainly in the coding step of the software development lifecycle (SDLC). But true productivity gains will come with TuringBots addressing all steps of the lifecycle. Generic announcements were made about integrating Microsoft functional testing tools — for example, to implement the tester TuringBot. Although we are aware that Microsoft testing tools do not shine as market leaders, we are eager to see how effectively and quickly GitHub executes on infusing TuringBots throughout the entire SDLC and not just in code generation and security.

There's a question of to what degree enterprises can customize Copilot to match their specific needs. We also wonder how well clients can fine-tune GitHub Copilot. GitHub addresses fine-tuning with an existing code base but could become more aware of enterprises' own specificities or for specific verticals, for example, before starting a greenfield project leveraging architecture patterns, specifications, policies, and standards that are not just code abstractions.