Tiden: Intent verification for coding agents

The product works from a map of the codebase. Tiden reads the repository and writes a spec for each behavior, then flags the ones no test proves. The demo project spans six components and 17 specs, drawing on web and API code, a billing service, Jira issues and error tracking across three environments. Before a task starts, the agent gets a working contract of promises, such as a rate limit returning a 429 above a set request count. The session cannot close until each promise has a passing test. Specs are meant to stay current. Tiden drafts an update on every merge, including merges made without it, and each branch shows what changed and which test proves it. Releases get a verdict broken down by component, and production errors point back to the release that caused them. Human reviewers still do the code review. Intent-driven verification is becoming a crowded area. Augment Code's Intent product checks work against a living spec before a pull request is created. Open-source tools such as OpenSpec and intent-gate also move requirements ahead of coding. For founders building on coding agents, the pitch targets a real weakness: an agent that writes both code and tests can pass its own checks. Teams selling or adopting agent-written code may need an independent record of intent. Because Tiden sits beside Claude Code, Cursor, Codex and Copilot, it does not depend on any one model vendor.