From: redpointai
Introduction to GitHub and Copilot
GitHub is the world’s largest source code host and a central part of the developer ecosystem [00:00:00]. It developed GitHub Copilot, the first AI coding assistant, which is now used by over 1.3 million developers and more than 50,000 enterprise subscribers [00:00:06].
Leadership Perspective: Thomas Dohmke, CEO of GitHub
Thomas Dohmke, CEO of GitHub, describes his journey from growing up in East Germany to becoming a founder acquired by Microsoft, and eventually leading GitHub [00:00:49]. He views the coolest part of his role in the age of AI as seeing people use Copilot and experiencing its “magical” qualities, helping them solve problems in natural language [00:01:13]. His long-standing experience in coding, through the rise of the internet and mobile, now positions him to enable a billion developers through AI [00:01:56].
The Impact of ChatGPT
The launch of ChatGPT in November 2022 marked a significant shift, turning initial skepticism about AI into widespread demand for its adoption in enterprise strategies [00:02:44]. This consumerization of AI technology made its power immediately understandable to a broad audience, prompting companies to seek its integration into their operations [00:03:15].
The Evolution and Impact of GitHub Copilot
Keeping Developers in Flow
Initially, Copilot focused on auto-completion, which, powered by large language models, allows developers to generate multiple lines of code, render test cases, or connect to cloud services without leaving their editor [00:04:22]. This significantly helps maintain “creative flow” by eliminating the need to switch between the editor and a browser for internet searches and copy-pasting [00:05:01].
Natural Language as a Development Tool
The addition of chat functionality and the use of natural language are crucial, as humans think in their native languages, not in code or deterministic machine language [00:05:35]. This democratizes software development, allowing individuals, including children, to express ideas and build games in languages like Brazilian Portuguese [00:06:12].
The Future Role of the Software Developer
Despite AI’s advancements, the role of a software developer in five years is expected to look “very similar to today” [00:07:22]. The change will be gradual, given the vast amount of existing code and the continuous creation of new codebases [00:07:27].
AI will:
- Simplify Maintenance: Make it easier to navigate complex, decades-old codebases, including those still running COBOL on mainframes [00:07:46].
- Improve Onboarding: New team members can quickly learn about codebases, stacks, and institutional knowledge by asking Copilot questions in natural language [00:08:21].
- Automate Tedious Tasks: Copilot will handle “not so fun stuff” like managing backlogs, addressing enterprise adoption blockers, and ensuring fundamentals like availability, security, and compliance [00:08:47]. Features like Copilot Autofix will help burn down security vulnerabilities and accessibility issues [00:09:38]. This frees developers to focus on innovation [00:09:49].
AI Agents and Their Function
Copilot itself is considered an agent, akin to a co-pilot flying a plane [00:10:06]. Developers are already accustomed to autonomous processes like CI/CD. AI agents will automate tasks like fixing security issues, much like dishwashers automate cleaning, allowing developers to focus on implementing new features [00:11:13].
Impact Across Industries
While some industries like translation may see replacement of human roles due to AI [00:11:48], in software development, AI will primarily “up level” capabilities, allowing developers to move up the technological stack, similar to past shifts from VMs to containers [00:12:13]. This includes advancements in accessibility, where AI devices might render the world perfectly for users with vision impairments, reducing the need for developers to manually implement accessibility features [00:31:10].
The Learning Journey for New Developers
AI serves as a higher abstraction layer over code, making it more accessible to get started with coding [00:14:43]. Children can use natural language to generate code for games [00:14:50]. However, when issues arise, developers still need to understand the underlying code to debug and fix problems [00:15:06]. AI’s “superpower” for new learners is its ability to unblock them quickly by explaining code or suggesting fixes, a stark contrast to older methods like consulting books or hoping for an “aha” moment [00:15:22].
Impact on the Open Source Ecosystem
AI is expected to make open source even more dominant [00:18:02]. It increases the usage of open source libraries by automatically suggesting them, as these systems are trained on vast amounts of internet information, reflecting what real developers would use [00:18:35].
It will also:
- Ease Maintenance: Help maintainers with tasks like burning down security issues and summarizing discussions [00:19:11].
- Increase Contributors: Lower the barrier to contribution by helping developers understand codebases and adhere to project structures, reducing the fear of pull request rejections [00:19:29].
There’s a potential for AI to favor higher-quality or more prevalent open-source projects, leading to consolidation rather than duplication of efforts [00:21:00]. While AI models can suggest common frameworks like React or Next.js, they also leverage the context of a developer’s existing project to provide relevant answers, promoting consistency [00:22:50].
Customization and Extensibility with Copilot
Fine-tuning for Enterprises
GitHub is releasing fine-tuning for Copilot, a highly requested feature, especially by enterprise customers [00:24:11]. This allows companies to train Copilot on their old codebases and internal libraries, which standard models wouldn’t know about [00:24:25]. Users can select repositories in their GitHub organization, and Copilot will automatically use the fine-tuned model for tailored results in their IDEs [00:24:53].
Extensions for Broader Integration
Extensions are a crucial part of Copilot’s future. They enable integration with various external services and the highly customized platform engineering stacks used by professional software development teams [00:27:01]. This allows developers to interact with their entire stack (e.g., feature flags, cloud services) through Copilot chat, eliminating the need to navigate multiple portals [00:27:40]. Priority areas for extensions include accessibility, security, compliance, and general platform engineering tools [00:46:58]. The mix of extensions and agents represents the “next wave” of Copilot, moving beyond mere code completion to scanning for vulnerabilities, reviewing code, and understanding entire codebases [00:48:06].
Challenges and Surprises in Copilot’s Development
Initially, there was skepticism about AI’s ability to write functional code [00:52:56]. The fact that Copilot could generate code with correct syntax, despite lacking a compiler, was surprising [00:53:19]. The initial “goodness” of Copilot, while surprising, was partly due to developers’ existing habit of modifying imperfect code snippets found online [00:25:50]. Copilot improves upon this by using contextual information (variable names, existing code) to provide more relevant suggestions [00:26:26]. The unexpected rapid pace of AI’s development and mainstream adoption, moving from “Horizon 3” research to immediate relevance, was also a major surprise [00:29:05].
The Future Pace of AI and Innovation
The pace of AI development is expected to accelerate, as AI itself is now being used to build AI systems [00:30:34]. This will lead to faster resolution of security and accessibility backlogs, allowing more focus on innovation [00:30:49]. Innovation is anticipated across the entire stack, including chip companies designing GPUs, improved models, responsible AI practices, and new applications [00:32:46]. No part of the development stack is expected to be untouched by AI’s efficiency and productivity gains [00:34:44].
Market Dynamics and Competition
The AI market is not a “winner takes all” scenario [00:35:38]. Companies, like Microsoft and GitHub, must continuously reinvent themselves to remain relevant [00:35:54]. Competition is healthy, fostering continuous innovation [00:36:27]. Enterprises will likely use multiple AI models for different use cases (e.g., code generation, security, documentation, testing), akin to using various SaaS services in their IT stack [00:39:09].
Advice for Founders
For founders going up against incumbents, Dohmke advises:
- Play the Long Game: Focus on the big picture rather than small, short-term battles [00:43:17].
- Continuous Reconfiguration: Companies must constantly adapt and evolve; there is no stable state or single “product market fit” [00:44:20]. This means teams and organizations must change along with growth [00:45:01].
- Say “No” Constantly: Focus is critical for startups, and it remains essential as a company grows. Leaders must decline most ideas to maintain strategic focus [00:45:12].
Quickfire Insights
- Overhyped/Underhyped: AGI and its fear are overhyped; AI’s impact on biotech and climate change is underhyped [00:51:19].
- AI Concerns: Dohmke expresses no personal fear of AI, even as a parent [00:51:35]. He believes AI will help protect against deepfakes and even manage screen time [00:51:59].
- Most Impressive Incumbent: Nvidia [00:52:44].
- Surprises with Copilot: Initially didn’t believe AI could generate code, but it did. Similarly, didn’t think chat would work, but ChatGPT proved it could [00:52:56].
- Exciting AI Startups: Mentions RVE and his investment in Edge AI, along with undisclosed biotech startups [00:53:50].
- Hypothetical AI Application: If not at GitHub, he would build the “next GitHub” [00:54:09].
Post-Podcast Reflections
The conversation highlights differing views on the future of developers. While Dohmke believes natural language empowers developers, the hosts suggest the number of developers needed in a company may decrease due to AI automating routine tasks [00:55:10]. However, AI will also lower the barrier to building software, potentially leading to more “micro businesses” and an unleashing of human creativity [00:57:39]. The complexity of Copilot’s architecture, using multiple AI models, underscores the likelihood of a “multimodal world” where different models serve specific use cases based on cost, performance, and deployment needs [00:58:31].
The challenge of modernizing legacy code, particularly COBOL systems, presents both a unique challenge and opportunity for AI to translate or reimplement these critical infrastructures [01:01:40].