Keep pulling the thread on Thomas Dohmke.
Thomas Dunk predicts that within the next five years, users will be able to use natural language interfaces to build completely personalized software on their personal devices, similar to J.A.R.V.I.S. in Iron Man.
GitHub has measured productivity gains of 25-28% on end-to-end developer tasks and 55% or higher on specific coding tasks for users of Copilot.
GitHub believes AI agents will be crucial for reducing security backlogs by automatically fixing known software vulnerabilities, improving the security of both open-source and commercial software.
GitHub recently announced an 'agent mode' for Copilot in VS Code, which allows the AI to take actions like installing packages via terminal commands to help implement features.
GitHub's 'Project Padawan' is expected in 2025 and will enable developers to assign a well-defined GitHub issue directly to Copilot, which will then create a draft pull request and work through a plan to resolve it.
As of a few quarters ago, GitHub Copilot was used by 77,000 organizations and had 1.8 million paid users.
Shortly after its preview launch, GitHub Copilot was writing approximately 25% of user code, a figure that later rose to around 50% depending on the programming language.
The full release of OpenAI's 'O3' model is expected to provide the improved reasoning capabilities necessary for advanced AI coding agents to achieve higher scores on benchmarks like VBench.
Thomas Dunk believes AI agents are still far from being able to take a rough, high-level idea and autonomously break it down into small, actionable work chunks without significant human guidance.
Thomas Dunk believes it will be at least a decade before AI agents can reliably transform legacy code, such as 30-40 year old COBOL on mainframes, into modern cloud applications.
Thomas Dunk believes the pricing for AI agents will be based on compute consumption, not value-based pricing equivalent to a human salary, because agents will have infinite supply limited only by GPU capacity.
Thomas Dunk believes the back-and-forth competition between open source models and proprietary closed-source models will accelerate innovation in the entire AI space.