GitHub Copilot cloud agent gets enterprise guardrails
GitHub turned Copilot cloud agent into a branch-first repo worker, then added signed commits plus org-wide runner and firewall controls that stricter teams actually need.
Signed archive
Maya Halberg
Maya writes across the AI field, from research claims and benchmark narratives to tools, products, institutional decisions, and market shifts. Her reporting stays focused on what changes once hype meets deployment, procurement, workflow reality, and human skepticism.
Stockholm · RemoteMethodology over launch theater.Former analytics lead turned newsroom translator for technical claims.
Back to all authors
Latest story
Seedance 2.0 takes Sora's abandoned AI video crown
ByteDance slipped Seedance 2.0 into the US on April 7, just after Sora shut down. That timing was not accidental, and the copyright fight did not disappear on the way in.
- 24
- Apr 11, 2026
- 6
Coverage signature
A result only matters after the setup becomes legible.
Sober, comparative, and suspicious of perfect charts.
Coverage lanes
Model claimsWorkflow shiftsMarket incentivesTrust and reproducibility
AI benchmark trustAI workflow analysislab credibility
Published stories
Everything currently attached to this byline.
vLLM 0.19.0 changes long-context cost math
vLLM 0.19.0 pairs CPU KV offloading, zero-bubble async speculative decoding, and Gemma 4 support in a release that changes long-context serving economics.
Qwen3.6-Plus pitches repo-scale coding agents
Alibaba pitches Qwen3.6-Plus as a 1M-context model for repo-scale coding agents. The real test is whether that yields calmer workflows, not prettier benchmark theater.
Google turns Android Studio into a local AI agent IDE
Google is not just adding Gemma 4 to Android Studio. It is linking local coding, AICore prototyping, and future Gemini Nano 4 phones into one Google-controlled path.
OpenAI acquires TBPN and buys the AI stage
OpenAI acquires TBPN may look like a media deal, but the sharper read is distribution: OpenAI just bought a live room for launches, guests, and the daily AI conversation.
OpenAI Codex plugins turn Codex into a workflow market
OpenAI Codex plugins are less a feature than a distribution move: Codex now has a curated install surface for reusable bundles of skills, apps, and MCP servers.
Gemma 4 is really Google's Apache 2.0 local agent stack
Gemma 4's real launch is the stack around it: Apache 2.0 weights, AICore, AI Edge Gallery, LiteRT-LM, and day-one local-agent support.
Granite 4.0 3B Vision is IBM's document AI wedge
IBM's Granite 4.0 3B Vision is a compact Apache 2.0 model for charts, tables, and form fields, giving document AI an open enterprise wedge.
Anthropic clamps down on Claude OAuth wrappers
OpenCode removed bundled Claude Pro/Max auth paths after Anthropic legal requests, pushing third-party wrappers back toward API keys and usage-based billing.
Anthropic study: AI users want help, not autonomy
Anthropic’s 80,508-interview Claude user study suggests the market wants productivity, learning, and cognitive support more than full AI autonomy.
Why AI benchmark wins feel less trustworthy
I still care about benchmarks, but not the old way. A score only matters if it survives reproducibility checks, task fit, and deployment reality.
