Former GitHub CEO Thomas Dohmke officially launched Entire, a decentralized code hosting and review platform built from the ground up for AI agents rather than human pull requests.

• Trails replaces pull requests: every code push is scored by agents using customizable evals, shifting review from human opinion to automated, evidence-backed verdicts.
• Positioned as a GitHub competitor: developers can host and review code on Entire, but the workflow and data model are designed around agent-to-agent handoffs rather than async human review.

TestMu AI, formerly LambdaTest, made Agent Assurance generally available, giving engineering teams a purpose-built harness that tests what AI agents actually do rather than what the code says.

• Tests agent actions, not just code: the product checks tool calls, file writes, and record changes, then generates functional and adversarial test suites from an agent's own docs and endpoint.
• CI-native gate: teams get 4,000 credits and four concurrent runs on trial and can block deploys on evidence-backed verdicts with no demo required.

Generative Pulse opened early access to a standalone platform that shows brand, marketing, PR, and SEO teams how AI models describe and recommend their brands, which sources shape those answers, and what to do next.

• SEEN framework: four signals cover Share of Answer, Evaluation Assessment, and two additional dimensions, giving teams a structured read on AI portrayal rather than raw prompt outputs.
• Actionable next steps included: the platform moves beyond monitoring to recommend content and source strategies that can shift how AI models present a brand.

OpenAI's GPT-6.1 Terra is now generally available inside GitHub Copilot across VS Code, Visual Studio, and other supported editors, priced at provider list pricing under usage-based billing.

• Designed for agentic terminal work: Terra outperformed GPT-5.6 Terra on terminal and C# tasks in early testing and focuses on consolidated steps for well-scoped work at lower cost than GPT-6.1 Sol.
• Available to all paid tiers: Copilot Pro, Pro+, Max, Business, and Enterprise users can select Terra in the model picker immediately.

Clay shipped auto-mode for Claygent, a model router post-trained with Applied Compute that automatically picks the right model for each enrichment row and cuts cost 47% versus always running the most capable model.

• Per-row routing: rather than applying one model to an entire table, auto-mode evaluates each task independently, matching complexity to model capability.
• Post-trained router: Clay trained the routing layer specifically on GTM enrichment tasks, not generic benchmarks, so decisions reflect real prospecting and research workloads.

GitHub made assisted permissions generally available across Copilot CLI, the Copilot app, and VS Code, letting an LLM evaluate each agent request to run a command, access a file path, or open a URL and auto-approve only clearly safe actions.

• Risk-tiered approval: requests the model classifies as safe proceed automatically; anything else surfaces to the developer, reducing constant interruptions without eliminating oversight.
• Enterprise controls included: admins can configure assisted permissions policies at the organization level, giving security teams a governance layer over what agents can do autonomously.

Greptile extended its code review agents with GPT-6 Astra's computer use capability, enabling automated end-to-end testing of user workflows rather than static code analysis alone.

• Behavioral testing at scale: Astra's computer use lets Greptile simulate how a user actually experiences an application after a code change, catching bugs that line-by-line review misses.
• Codex for rework reduction: Greptile also uses Codex to reduce rework by catching issues earlier in the review cycle before they propagate downstream.