“Integrate a capable coding model into a developer's primary workspace, and the IDE transforms from a place where you write code into a platform where you direct an agent, observe its reasoning, manage its focus, and decide when its output is ready for shipment.” This core philosophy defined the inaugural JetBrains x Codex Hackathon. Across roughly 40 submissions over a single weekend, teams explored what it truly means to build with AI natively inside the IDE, rather than merely bolting it on. The six finalists presented some of the most compelling solutions.
🥇 First Place: Hyperreasoning – Aditya Mangalampalli
Many coding agents rely on a single model call, hoping for the best outcome. As Aditya explains, “LLMs often spend a lot of time thinking in circles.” Hyperreasoning replaces this single-shot approach with a method akin to a search algorithm: the system drafts several potential approaches to a task. A learned controller then decides which paths to expand, which to prune, and which to verify against tests. Compiler errors and failing tests provide crucial feedback, influencing how the controller weighs its options.
Within the IDE, a dedicated tool window provides a live visualization of this search process, allowing developers to see the paths the controller explored before settling on a solution. The project posits that a smaller local model, when wrapped in this type of verified search loop, can compete effectively with much larger frontier models at significantly lower costs. The IDE, in this context, serves as the environment where AI reasoning becomes transparent and steerable, rather than an opaque black box that simply outputs code.
🥈 Second Place: Scopecreep – Bhavik Sheoran, Kenneth Ross, Roman Javadyan, Joon Im
Hardware bring-up is typically a complex, multi-tool endeavor: a schematic viewer in one window, vendor applications for oscilloscopes and power supplies in others, a terminal communicating with the device, and a spreadsheet for collecting results. Scopecreep consolidates this entire workflow into a single JetBrains tool window. By providing a circuit schematic, an AI agent can autonomously conduct board testing – intelligently selecting signals to measure, capturing readings, and generating a comprehensive report.
A notable design decision lies in how the agent determines...
