Specify the outcome · Then implement
Spec-driven development uses an explicit specification as the reference for planning, implementation and review. With Anvil Coder, you prepare requirements in Spec Studio, curate the resulting tickets and delegate selected tasks to AI coding agents.
Record the audience, goals, non-goals, behaviour and open questions. Decisions become an editable document.
Describe the stack, interfaces, quality rules and check commands. Write acceptance criteria against which the result can be assessed.
Review the proposed epics and tasks. Creating tickets does not start a coding run; you decide which story is implemented.
Compare changes and executed checks with the specification. Feed new findings back into requirements and subsequent tasks.
Illustrative task · not a benchmark
“We need invitations” names a feature. A useful specification also records decisions that would otherwise be made during implementation:
Review now has an explicit reference. If the team later changes the expiry period, update the specification, implementation and related tests together.
Spec Studio guides you through vision and context, goals and non-goals, functional requirements, stack and constraints, quality requirements and guardrails, and epic slicing. You edit the actual text and decide which suggestions to accept.
A stack profile describes tools and verification commands. Their availability must be established in the execution environment. A specification and profile alone do not establish that tests passed.
Spec-driven AI development gives agents and reviewers a shared reference. It helps expose conflicting requirements and missing decisions early. An inaccurate requirement can still lead to an unsuitable solution.
Keep the specification current as you learn. Existing tests, new checks and human review examine different aspects of the result and complement each other.