Early-Signal Detector

tool · first seen 4 Oct 2026

coding agents

Also: AI-IDE, SWE agents, GenXR-Bench, code agents, AI coding agents, coding assistant, LLM coding agents

Systems where language models autonomously read, edit and test code in a repository using tools.

Coding agents are LLM-driven systems that read, edit, run and test code inside a repository through tool calls, as in Claude Code, Codex, OpenCode and GitHub Copilot's agent mode. The sources treat them as an existing substrate rather than a new idea. Research papers study how revisions to SKILL.md instruction files change agent behaviour across 21 models, and how to keep memory state consistent over long-horizon software tasks. Developer repos build tooling around agents: agent-oriented project scaffolds, screen-annotation input, git variants for agents, and mobile harnesses. A cluster of identical "GitHub-Copilot-app" repos from throwaway accounts looks like spam and inflates the count.

The concept is already mainstream, so the useful signal is in its sub-problems. These include instruction-file management, long-horizon memory consistency and version control designed for agents, which will shape how reliably such agents work on real codebases.

Analyst note (established), 17 Nov 2026.

Mentions per week

02819 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Rank 26 on 14 Nov 2026 · 154 mentions in 28 days · 152 independent groups · papers → code → discussion · first seen 6 weeks ago · acceleration -0.17

Earliest sources

Latest sources

Mentioned together with

model context protocolClaude CodeSWE-benchcontext engineeringAI agent safety evaluationGitHub Copilotagent skillsautomated scientific discoveryterminal user interfaceKV cache reuse