Plan
Requirements are structured, open points are made visible as assumptions and turned into a directed work graph.
The principle
An AI software factory coordinates software implementation across multiple work steps. Anvil Coder starts with requirements and project rules. AI agents create commits; configurable checks and pull-request review make the result traceable.
A large project does not fit into a single prompt. The factory solves this problem not with a bigger wall of text but by cutting to size: every task becomes a small, comprehensible slice with exactly the context it needs.
The coordinator holds the overall plan. Workers implement the individual parts and leave their commits behind. Phase barriers pin down the order where it matters for the domain; within a phase there is room for parallel work.
Requirements are structured, open points are made visible as assumptions and turned into a directed work graph.
Every slice gets its own worktree. The model works with a bounded context instead of failing against one unmanageable overall prompt.
Configured checks report their results. Build and tests are optional and enabled in the hosted service; a matching toolchain is required. Results that cannot be verified remain visible.
Every station leaves traces: commits, checks and an audit trail make visible how the repository emerged from the requirement. People keep the authority to decide; the repeatable machine work covers the distance in between.
Understand task scope, autonomy and the evidence your reviewers need.
Follow an assignment through planning, changes and the review handover.
Turn requirements and acceptance criteria into actionable tickets.