A single daily edition of announcements from the AI coding tools you use, read from a curated source list, summarised, and linked back to the original. No infinite feed, no account, no advertising.
Agents pointed at licensed private production code finish under two tasks in five
Specific Labs published Real-SWE, a benchmark whose ten tasks come from private production codebases licensed from real companies rather than from public repositories. Fable 5.1 leads at 38.8%, ahead of GPT-6 Astra at 33.8% and Gemini 3.8 Flash at 31.2%, and no model solves every task. The most common failure is a missed requirement, behind 36.7% of Fable 5.1's failures, followed by unverified assumptions about the system. A cofounder says every codebase predates 2023, so the models are not being scored on AI-generated code.
A Jira ticket can carry its own context into the branch and the pull request
GitHub Copilot's app now pulls Jira issues into a shared canvas where work items are prioritised. Copilot then holds that context through investigation, implementation and pull request preparation, rather than losing it at the hand-off.
The same model ID on OpenRouter answers differently depending on who serves it
OpenRouter's automatic fallback spreads a request across several backend providers, and those providers run different serving software and configurations, so one model endpoint does not behave consistently. Some providers lack vision support for a vision model, and reasoning effort is handled differently between them. The fix is the provider.only option to restrict routing, plus the /endpoints method to find which providers actually serve a given model ID.
One flag runs a single Kiro session on the V2 harness and leaves your default alone
Kiro CLI 2.21.4 adds --v2, which runs one session on the V2 agent harness. It works on both kiro-cli and kiro-cli chat, and takes precedence over a saved chat.agentEngine value without overwriting it. The same release lets session search cover prompts only or prompts and agent responses, though tool output is indexed in neither.
Claude Code, Codex and Cursor share one workspace that survives a dropped connection
AgentsDock is an open-source IDE that puts Claude Code, Codex and Cursor in one workspace reachable from desktop and mobile. Sessions persist through tmux, so a dropped connection does not end a run, and the client attaches to remote servers for file editing and live training output. It is beta, at v0.2.13-beta.33, installed by cloning the AgentsServer repository and running its install script.
