More problems

Summary

The post outlines research problems for the next phase of AI programming, centered on making coding systems predict actions, perform larger edits, understand codebases, and debug software. Proposed directions include low-latency next-action models for keystrokes, locations, files, terminal commands, and edits; background execution for “perfect edits” and pseudocode compilation; and multi-file changes supported by stronger tool use, reasoning, and agent trajectories. For context, retrieval alone is described as insufficient for codebase questions requiring multiple hops, while useful context may span documentation, source, history, UI pixels, logs, and messages. The post recommends combining retrieval, recurrence, and long-context attention, alongside specialized rerankers, multi-hop embedders, caching, and codebase-level retrieval. For reliability, it proposes calibrated bug detection, always-on AI linting, runtime-state tracking through cursor/debug, and improved debugger tool use, while noting trade-offs involving latency, false positives, infrastructure, and user experience.

Takeaways

1. 1

   Copilot++ is being extended toward next-action prediction for locations, files, terminal commands, and edits, including sequences of low-entropy changes such as repeated tab presses.

2. 2

   The post argues that difficult codebase questions require multi-hop retrieval and proposes training multi-hop embedders, improving rerankers, using prefix caching, and researching codebase-level retrieval.

3. 3

   Always-on AI linting would need to be cheaper and faster than AI review, run several times a minute, and use a lower false-positive rate; debugging could add runtime variable-state information through cursor/debug.