The team designs the software
Requirements, architecture, task decomposition, implementation strategy and acceptance remain engineering decisions. Developers can write or modify any code directly.
Task-Contract AI Development Framework
Requirements, architecture, task decomposition, implementation strategy and acceptance remain engineering decisions. Developers can write or modify any code directly.
The team chooses what to delegate. Each task defines context, authorized changes, preserved behaviour, checks and stop conditions; the agent cannot broaden the work on its own.
Small preservation-first changes follow the codebase’s existing methods and patterns, reducing hidden rewrites, avoidable rework and debugging debt.
Engineering first
TCAF does not hand an entire project to an agent, split the output into smaller prompts and reduce the developer to approving each result. The team still decides what to build, how to build it, how to divide the work and which parts are worth delegating.
AI can inspect, draft, implement or verify only where the team chooses to use it. Developers remain free to write code directly, adjust a task, replace an approach, perform a justified refactor or stop the agent at any point.
Operating principles
Product intent, architecture, code quality and final decisions remain with the developer or team.
AI is used only for the work the team chooses to delegate; direct development remains fully available.
New code integrates with working methods, guards, utilities and local patterns instead of rewriting them opportunistically.
The agent observes the actual project and nearby implementation before proposing or performing changes.
The Task Contract defines the authorized change surface, preserved areas and stop conditions.
Executed checks, limitations and pending verification remain visible before the team accepts the result.