coding-agents

A new empirical study finds coding agents overwhelmingly read and write their own instruction files, rarely touch classical documentation, and almost never use docs to recover from failures.

Grok 4.6 is now available in GitHub Copilot for agentic coding, days after SpaceXAI positioned it near GPT-5.6 Sol and Claude Fable 5 on the Artificial Analysis Intelligence Index.

A George Mason University study finds a software task's difficulty for AI coding agents can be predicted from patch and repository structure before the agent ever runs.

A Google study finds AI agents that first write a semi-formal specification before generating tests catch 9.8 percentage points more real bugs than agents that generate tests directly.

A study spanning seven AI models and five coding-agent harnesses finds they fail identically when a required repository fact is unavailable, and fabricate rather than stop.

A RepoComplianceBench study of four frontier coding agents found they open contribution-rule files only 3.5% of the time and never voluntarily withdraw AI-banned contributions.

Mistral rebranded its Le Chat assistant as Vibe, a single agent spanning Work Mode for office tasks and Code Mode for remote coding.

The maker of Devin, an autonomous AI software engineer, closed a Series D round led by Lux Capital, General Catalyst, and 8VC, more than doubling its valuation from $10.2 billion eight months earlier.

GitHub launched a native desktop app for Windows, macOS, and Linux that runs multiple isolated Copilot agent sessions in their own git worktrees and ties them back to issues, pull requests, and merges.

Mistral released Medium 3.5, a dense 128B open-weight model with a 256k context window that consolidates Medium 3.1, Magistral, and Devstral 2 under a Modified MIT license, with a per-query reasoning toggle.

OpenAI's April 16 Codex update adds background computer use on Mac, parallel multi-agent execution, persistent memory, and 111 plugin integrations, escalating the agentic coding tool race with Anthropic.