LLM as Co-pilot, Co-Integrator and Co-Facilitator is a three-stage division of AI-assisted software development proposed in 生成式 AI 与开源重塑软件研发, drawn from its authors’ experience of exploring AI assistance across the whole development process and of building generative-AI development tools and platforms. In the first stage, LLM as Co-pilot, the model helps a single role complete its tasks and affects individual work. In the second, LLM as Co-Integrator, it makes communication between different roles more efficient and affects how roles interact. In the third, LLM as Co-Facilitator, it assists with planning, forecasting and coordinating work and affects organizational decision-making. The progression runs from individual to team to organization.
Usage
The whitepaper places the industry of 2023 at the first stage: building Copilot-style tools was, it says, the consensus that could actually be put into practice, though model limitations and a lack of AI engineering capability kept results from being ideal. It treats the second stage as the next focus: once an organization has standardized the outputs of each role, such as requirement documents and test cases, using AI to assist communication across roles and teams becomes more important, and connecting the whole toolchain through AI and BizDevOps becomes the next exploratory goal. It cautions that this is not easy, because it involves people’s capabilities, reshaping role responsibilities, and changing and optimizing processes.
When It Applies
It is a framing for where an organization’s use of AI in development sits and what to aim at next, rather than a procedure. It assumes, for the later stages, that roles already produce standardized artifacts that AI can pass between them. It is one initiative’s proposal, set out in its own whitepaper alongside its own tools, not a measured or widely adopted model, and the whitepaper itself notes that how to measure the efficiency gains remains unsolved.