Tessl tackles the core reliability problem with AI coding agents: they rush to write code from vague prompts, hallucinate API calls for libraries they half-remember from training data, and break existing functionality when implementing new features without realizing it. The solution is spec-driven development — capturing what you want to build in structured specifications that include component descriptions, capability lists with linked tests, and public API definitions. Agents receive these specs as their primary input and generate code to match, rather than inventing their own interpretation of a loose prompt. When the agent drifts, the tests catch it before anything ships.
The platform consists of two products. The Tessl Framework integrates with your existing AI coding agent (Claude Code, Cursor, Copilot, or others) and guides it through a structured workflow: first create a written plan of action, then capture decisions in reusable spec documents, then generate code that satisfies the spec, then verify against linked tests. Specs live in the codebase as long-term memory, so when the agent comes back to evolve the application later, it has context about past decisions rather than reinventing the architecture from scratch. The Spec Registry is an open catalog of over 10,000 specifications explaining how to use external libraries — preventing the common problem where agents hallucinate nonexistent function signatures or confuse library versions.
The philosophy behind Tessl follows a maturity progression: spec-first (write a good spec, let the agent build the first version), spec-anchored (specs stay in the project and update alongside code), and eventually spec-as-source (developers maintain only specs, code is fully agent-generated). Tessl maintains a 1:1 mapping between spec files and code files, marking generated output with comments indicating it should not be manually edited. The Framework is currently in closed beta with a waitlist, while the Spec Registry is available in open beta for free. The project reflects a broader industry shift where multiple tools — GitHub's Spec Kit, AWS Kiro, and Tessl — are converging on the same insight: defining software through specifications rather than prompts produces dramatically better agent output.