
Stories often reach the sprint before they're ready: untestable criteria, a step that needs production access, an unfinished dependency. People can ask in stand-up. AI coding agents like IBM Bob, Claude Code, Copilot and Codex now pick up stories too, and they can't. Nothing stops an agent from: - building from an unclear story - marking work done without proof - changing work only a person should do SOLUTION Refinery is a readiness gate for backlogs shared by people and AI agents. - Assess: weighted checks for acceptance criteria, failure cases, rollout and the team's Definition of Ready, plus four "could an agent build this without asking?" checks. - Explain: one of four verdicts, from Ready to Not ready, with a question for the author on every failed check. - Enforce: unrefined stories stay in Backlog or Refinement. Done needs a test report, a screenshot if the UI changed, and a commit. Dependencies must finish first. Agents can't touch human-only work. Every change is logged with who and why. TARGET USERS Scrum teams handing work to AI coding agents: product owners, developers and Scrum Masters. HOW THEY USE IT - Story Board (web app): create, drag and assess stories, attach evidence, mark dependencies and human-only work. - Linear: add the ready-check label to assess an issue. - AI agents: connect to the MCP server with one command from Claude Code, Copilot CLI or Codex. WHAT MAKES IT UNIQUE - Built for agents: about half the quality score asks whether an agent could build the story unaided. - Rules live in the API, not the UI, so people and agents follow the same ones. - A refused agent is told why, so it reports back instead of working around the rule.
27 Sep 2026

Most website chatbots only reply. Talkie also acts on the page. It's a voice assistant SDK made of web components that work with any framework, or with none: add one script tag and a <talkie-assistant> element, and visitors can just talk. The conversation is hands-free and runs on AssemblyAI's Voice Agent API over a single WebSocket. The reply streams back as audio, and the words on screen appear in step with the voice. The key feature is tool calling. In our demo, a real-estate listing, a visitor asks "show me the kitchen" and the 3D tour moves there. They ask about nearby schools and the map filters to schools. They say they'd like a viewing and the booking form opens. Each assistant's persona, greeting, voice and tools live on your server, so you can change its behaviour without releasing a new version of the page. A playground lets you tune the look and copy the embed snippet, and a console lets you edit personas.
30 Sep 2026