Key Signals

• Cursor acquires Continue, consolidating the AI coding assistant market under SpaceX's ownership umbrella. Continue — the open-source coding assistant with 34,300 GitHub stars and an Apache 2.0 license — has been shut down following its acquisition by Cursor. With SpaceX having simultaneously acquired Cursor for $60 billion, Elon Musk's enterprise now owns one of the most prominent open-source coding tools. Two of Continue's founding engineers have joined OpenClaw rather than Cursor, suggesting talent is fragmenting across the rapidly consolidating landscape. [1]

• Claude Code creator Boris Cherny declares agentic loops "just as important" as the shift from handwritten code to agents. Speaking at Meta's @Scale conference, Cherny described running continuously-looping agents that submit architecture-improvement PRs without human prompting. This represents a fundamental shift from discrete task completion to always-on background coding agents — raising both capability ceilings and cost concerns as token consumption becomes theoretically unbounded. [2]

• GitHub Copilot in JetBrains IDEs now supports Claude as an agent provider in public preview. This marks a significant interoperability milestone: developers can select Claude Code as their agent backend directly within the Copilot Chat panel, alongside native Copilot agents. The same update brings organization-level custom agents, message queuing during long-running CLI sessions, and makes cloud agent generally available. [3]

• OpenAI launches Daybreak — Codex Security and GPT-5.5-Cyber for automated vulnerability patching — alongside "Patch the Planet" for open-source maintainers. These initiatives position Codex as a security tool rather than just a coding assistant, using AI to find, validate, and fix vulnerabilities at scale. The open-source-focused "Patch the Planet" program specifically targets maintainers who lack dedicated security resources, potentially shifting the economics of open-source security maintenance. [4][5]

• Google Labs publishes proactive agent evaluation research for Jules, arguing coding agents must shift from tasks to goals. The paper "Agentic Coding Needs Proactivity, Not Just Autonomy" introduces a benchmark where agents explore codebases to surface diagnostic insights before being asked. Preliminary results show Hit@5 accuracy jumping from 33% to 57% when the exploration budget increases from two to three rounds — quantifying how additional compute directly improves proactive code understanding. [7]

• Codex v0.142.0 introduces rollout token budgets and multi-agent delegation controls. Configurable token budgets track and cap usage across agent threads, aborting turns when exhausted — a direct infrastructure response to the runaway cost problem identified with always-on loops. Multi-agent delegation can now be set to disabled, explicit-request-only, or proactive per thread, giving enterprises granular control over agent autonomy. [8]

AI Coding News

• Qodo ships cross-repo code review to address the governance crisis in AI-flooded engineering teams. According to Google's DORA 2025 report, pull requests from high AI-adoption teams are 154% larger, take 91% longer to review, and ship 9% more bugs. Qodo's new Cross-Repo Code Review, Custom Rules Miner, and Skill Review Standards capabilities aim to close this gap by understanding dependencies across repositories and automatically discovering enforceable coding patterns from existing behavior. CEO Itamar Friedman argues that "speed without independent verification isn't velocity; it's technical debt in disguise." [6]

• OpenAI publishes a developer guide on "Codex-maxxing" for long-running work. The article demonstrates techniques for preserving context, managing complex multi-file projects, and maintaining continuity beyond single-prompt sessions — practical strategies for sustained engineering work rather than one-shot tasks. This reflects the broader industry trend of AI coding tools evolving from interactive assistants into persistent background workers. [9]

Feature Update

• GitHub Copilot CLI v1.0.64-2 adds inline image rendering, per-model token totals in /usage, and autopilot-to-interactive mode transitions. The pre-release introduces inline image rendering directly in the terminal, argument-hint frontmatter support for skills, and OpenTelemetry compaction markers. Autopilot mode now automatically returns to interactive after the agent calls task_complete. Fixes address PowerShell spurious directory access prompts, musl Linux SEA package resolution, mouse wheel scrolling after exit, and OpenTelemetry semantic conventions compliance for cache and reasoning token metrics. [10]

• GitHub Copilot CLI v1.0.64-3 adds HTTP proxy configuration via a user setting. This quick follow-up pre-release enables proxy configuration for enterprise environments, fixes session resume when names contain spaces, and hides unsupported slash commands in remote-hosted sessions. [11]

• Claude Code v2.1.186 adds CLI-based MCP authentication and automatic bash command responses. The claude mcp login/logout commands enable MCP server authentication directly from the CLI with --no-browser SSH support. The ! bash command prefix now triggers Claude to automatically respond to output. Background subagents now surface permission prompts in the main session instead of auto-denying. Workflow agent subagents no longer loop forever on validation failures, aborting after 5 attempts. The /review command now uses the same engine as /code-review medium. [12]

• OpenAI Codex v0.142.0 ships plugin marketplace organization, indexed web-search mode, and scheduled time reminders. The /plugins view now categorizes remote plugins into OpenAI Curated, Workspace, and Shared sections with contextual recommendations. An indexed web-search mode permits live searches while restricting direct page access to server-approved URLs. Performance improvements reduce startup latency through deferred DNS, model cache warming, and parallelized skill metadata reads. [8]

• GitHub Copilot for JetBrains IDEs: Claude as agent provider, org/enterprise agents, and message queuing. Beyond Claude integration, the update adds a /models slash command for the model picker, larger context window selection, recently used models section, and per-turn AI credits visibility across all session types. Cloud agent moves to general availability. Agent debug logs gain a new summary view with aggregate session stats. [3]