Cognition added multi-level budget controls to Devin, its AI coding agent. Teams can set limits up to three levels deep, with alerts and efficiency-based auto-approvals.

Tsinghua researchers built HarnessPro, a four-agent loop that rewrites AI coding agent scaffolding to cut per-task costs 26% while improving accuracy. Factory Droid 1.0 ships the same idea commercially.

KloudStax CTO Vinay Thakker breaks down agent identity models, credential proxies, a 10-line terminal policy, and how red teams should test hijacked agents.

LinearB launched Agentic Studio to give engineering leaders and coding agents shared delivery context. Benchmark data from 8.1M pull requests explains why the tool exists.

Cognition's Devin now supports hierarchical AI spending controls across departments and projects, as 98% of FinOps teams report managing AI costs in 2026.

Anthropic outspent Duolingo's entire sales budget on ads in 2025, Meta's AI terms triggered a UK probe, and journalism's share of AI citations dropped from 27% to 22%.

Apiiro found vibe coding makes engineers 4x more productive but creates 10x the security issues. Five practices every non-engineer builder needs before shipping.

Free Claude Code promises 1.3 billion free tokens monthly and 61 AI providers. The GitHub repo is real. The marketing needs a closer read.

A bug bounty writeup shows how a single trailing dot in GitHub Copilot's security prompt becomes an attack vector. AppSec engineers and AI tool users should read this.

A fake Claude Code extension in the Cursor marketplace installed an SSH backdoor on developer machines. Here is what happened and what to check.

Dataiku launched Headless, a platform capability with an MCP server that lets AI coding agents like Claude Code, OpenAI Codex, and Cursor build data pipelines inside enterprise governance guardrails.

AI coding agents pushed GitHub's code commits to 7.38 billion in September, a 5x year-on-year jump. Here is what that means for dev infrastructure.