Blog

Research and updates from Aspect Research.

Aspect Code is live. One command, every repo.

Aspect Code

Aspect Code generates and maintains AI coding context automatically. Inspired by autoresearch, built for every agent.

Asa ShepardMarch 31

2 min read

Oracle Tuning Results and the Path to Adaptive Context Selection

Aspect Code

Results from running the iterative Oracle tuning procedure on SWE-bench instances, what the data shows about guided context at scale, and where Aspect Code research goes next.

Asa ShepardMarch 16

5 min read

A New Direction: CLI-First and Iterative Context Refinement

Aspect Code

Aspect Code is stepping back from the VS Code extension in favor of a simpler CLI approach built around AGENTS.md and ongoing research into iterative context refinement.

Asa ShepardFebruary 24

1 min read

Making AI Reliable on Real Codebases: The KB + Feedback Loop

Aspect Code

Generic AI instructions plateau. I ran benchmarks on a FastAPI template and djangopackages to show how a repo-specific knowledge base plus a feedback loop transforms AI from a liability into a time-saver.

Asa ShepardJanuary 3

9 min read

Benchmarking AI Code Generation – From –24 to +25 Net Tests with Codebase Context

Aspect Code

I tested LLMs with and without structured repo knowledge on FastAPI and Django. Here's how adding context changed the results.

Asa ShepardDecember 10

8 min read

Aspect Code: Giving AI Assistants the Context They're Missing

Aspect Code

AI coding tools break tests and ignore your architecture. Aspect Code gives them structured context so they write code that actually fits.

Asa ShepardNovember 30

8 min read