Launching today
Tiden
Intent verification for coding agents
247 followers
Intent verification for coding agents
247 followers
Tiden verifies your coding agents built what you actually asked for. Most checks run inside the session that wrote the code, so the tests agree with it by construction. Tiden keeps your intent as requirements that branch and merge like code, and checks every change against that record. The verdict comes from tests run in your environment; no model grades the diff. You decide what merges, and production errors trace back to the release. Works beside Claude Code, Cursor, Codex and Copilot.
Nikita and the team were shipping exciting features and products before, this one is a gem 💎
I do believe that no model grading is the right call. Our evals had deterministic routing tests before it could ship. That split is what kept people trusting it. How so you handle cases where deterministic tests aren't enough, like "this should feel fast"
@amartya_gaur1 Good question. Tiden checks against what was asked. When the original intent says "should feel fast", Tiden puts that intent in front of the agent and flags that nothing tests it. That's usually the moment the agent works out that to test it, it needs a number. It comes back with concrete proposals, like "export finishes under 2 s for 10k rows", and I pick the values. From there it's a requirement with a test like any other. Until someone picks them, it stays unverified, and Tiden says so instead of guessing.
Keeping requirements updated alongside code changes sounds useful. It could help teams catch things that get missed development.
@aarav_pittman Yes, living functional specification of product behaviour has been maintenance challenge and Tiden solves that by making sure product spec is maintained same as a codebase automatically. This ensures tribal knowledge is not lost when someone leaves and there is alignment for different stakeholders by creating cross functional visibility.
When Tiden flags a requirement as untested, what stops the agent from writing a shallow test just to turn it green?
@alexey_voronkov1 The system is designed to provide the agent with instructions that include the user’s original requirement. This means it evaluates not just the generated code itself, but the code against the requirement. Tiden also requires the agent to go beyond a superficial test and consider broader testing scenarios, such as negative cases and edge cases.
Does this work if the team never wrote requirements in the first place? Most of our "spec" lives in tickets and people's heads
@gregory_lesnikov yes, during the installation it goes through the code and figures out components and specs for every working feature