Study reveals language models in coding agents encode future program states up to 25 steps in advance
A coding agent solving a software-engineering task spends dozens of steps reasoning, editing code, and running tests, yet little is known about what the underlying language model internally…
Read the full articleYou might also wanna read
Separating Problem Solving from Code Generation: Evaluating LLMs on Competitive Programming Through Natural-Language Editorials
Large Language Models (LLMs) increasingly succeed on competitive programming problems, yet existing evaluations conflate algorithmic reasoni
arxiv.org·7d agoHow we built a Linear coding agent: the hard parts
Building a production coding agent that lives in Linear. Wrapping Claude Code and Codex as child processes, surviving state loss from archiv
Revisiting local LLMs for coding: A 4-week hands-on assessment
Notes from my Thoughtworks colleagues on AI-assisted software delivery
Revisiting local LLMs for coding: A 4-week hands-on assessment
Notes from my Thoughtworks colleagues on AI-assisted software delivery
Revisiting local LLMs for coding: A 4-week hands-on assessment
Notes from my Thoughtworks colleagues on AI-assisted software delivery
Evaluating AI's Stochastic Model Discovery: A Fresh Perspective
A new framework emerges to measure AI coding agents like Codex and Claude Code. The focus is on variability, cost, and complexity in model d
Beyond AI Vibes: Deterministic Foundations for Agentic Coding
AI coding adoption is high and trust is dropping. A testing pyramid for agents, plus reproducible production context that grounds AI in real
AI's Impact on Software Engineering: Evolution or Replacement?
With vibe-coding, what I find most useful isn’t even writing code but understanding it.
The Verge·10mo ago
Comments
Sign in to join the conversation.
No comments yet. Be the first.