coding agents

A new study of developer complaints about Claude Code, Cursor, GitHub Copilot, OpenAI Codex and similar AI-native development tools reaches a blunt conclusion: the biggest security and privacy failures are often created by what the agent is allowed to access and do, rather than by the underlying...

DeepSeek-V4 Flash and GPT-5.6 Luna do present a real cost-versus-capability trade-off for coding agents, but the widely circulated “4.8 times more tasks per dollar” comparison leaves out the assumption that makes its proposed cascade workflow work at all: someone or something must reliably know...

A browser game built around AI coding-agent permission prompts has exposed a problem that security teams should already recognize: asking a developer to click Approve hundreds of times is not meaningful oversight. In data from more than 40,000 game sessions and roughly 409,000 decisions, players...

Meta has launched Muse Code in beta, a terminal-based coding agent powered by the new Muse Spark 1.2 model, putting the company directly into the developer-agent market currently defined by Anthropic’s Claude Code and OpenAI’s Codex. The practical catch for Windows shops is immediate: Muse...

A new open-source Python project called Context Compiler makes a sensible case for treating a coding agent’s prompt as build output: keep the file being edited in full, reduce reachable dependencies to signatures and docstrings, and omit the rest. But the project’s headline claim — 69% to 74%...

GitHub on July 10, 2026, added improved filtering and sorting to Copilot session lists in the latest production builds of GitHub Mobile for iOS and Android, giving developers faster ways to isolate active, completed, attention-needing, repository-specific, type-specific, and agent-specific work...

Microsoft announced on July 6, 2026, that Dataverse is expanding its role as an agent data platform by bringing its coding-agent plugin to Claude, Cursor, and GitHub Copilot while broadening MCP support, certification, and governance across Microsoft’s AI stack. The news, detailed in a Microsoft...

OpenAI released GPT-5.5 on April 23, 2026, one week after Anthropic launched Claude Opus 4.7 on April 16, putting the two most visible frontier AI labs into a direct comparison across benchmarks, coding tools, subscriptions, APIs, and enterprise workflows. Mashable framed the matchup as a split...

Z.ai’s GLM-5.2, released in mid-June 2026 by the Beijing startup formerly known as Zhipu AI, has surged on developer platforms by offering near-frontier coding and agent performance at dramatically lower prices than leading U.S. proprietary models. The important part is not that China has...

Meta’s superintelligence chief Alexandr Wang told employees on July 2, 2026, that Meta’s in-training Watermelon model has caught up with OpenAI’s GPT-5.5 on closely watched AI benchmarks, according to Business Insider, while promising near-term gains in coding and agentic capabilities. That is...

The 2026 AI-agent platform market is no longer a clean leaderboard of chatbots with tool access; it is a fragmented contest among coding agents, Microsoft 365 workflow builders, open-source orchestration frameworks, customer-service specialists, and governance-heavy enterprise platforms. That...

Microsoft’s neat framing is simple and urgent: the era of “better answers” — where large language models serve mainly as glorified search-and-summarize tools — is yielding to something fundamentally different, and that difference matters more for business outcomes than model accuracy alone. The...