Keep pulling the thread on Burke Holland.
The cost for a user to consume a billion tokens per month on some AI coding platforms is approximately $200, while the platform's underlying cost is estimated to be around $25,000, indicating significant subsidization.
Burke Holland predicts that the current high level of subsidization for AI model usage costs will not be sustainable indefinitely.
GitHub Copilot allows for the orchestration of multiple different large language models, such as Gemini, Codex 5.3, and Opus 4.6, within a single execution run.
While AI agents have demonstrated the capability to build complex software like a web browser, the process of making that software production-ready and shippable remains a significant, unsolved challenge.
Burke Holland argues that AI is currently impacting tech jobs not by replacing workers directly, but by causing companies to divert budget from headcount to GPU infrastructure investment.
Burke Holland predicts that instead of AI replacing developers, it will lead to a future where "everybody is a developer at some level."
The cost and effort to create software have decreased so dramatically that incumbent software companies face a high and increasing risk of having their products re-implemented by competitors or individuals.
GitHub has announced new agentic workflows for GitHub Copilot that include features like memory, deep research capabilities, and the ability to deploy a "fleet" of agents via the CLI.
The GitHub Copilot CLI has exited preview and is now generally available (GA), making it ready for production use by enterprise customers.
In a blog post on January 5th, Burke Holland predicted that the AI model Opus 4.5 would "change everything."
The Sonnet 3.5 model was characterized as being "eager" to please, which often resulted in the generation of sloppy, spaghetti-like code.
Using GitHub Copilot with the Opus 4.5 model, Burke Holland was able to build a native Windows tool for resizing windows in a single attempt.