Emigo New Agentic Workflow Proposal

by ADMIN 36 views

Introduction

In this article, we will discuss a new agentic workflow proposal for Emigo, drawing inspiration from Aider's clear separation between Architect and Coding roles, and enhanced with principles seen in tools like Cursor, Devin, and Lovable. This workflow is designed to keep you, the user, firmly in control at every stage while maximizing clarity, safety, and precision.

The Problem with Current Workflows

Traditional workflows often involve a single agent or tool that attempts to handle both architectural and coding tasks. This can lead to a lack of clarity, safety, and precision, as the agent may not fully understand the project's context or requirements. Additionally, the user may feel overwhelmed by the complexity of the workflow, leading to frustration and decreased productivity.

The Proposed Workflow

Our proposed workflow addresses these issues by introducing two distinct agents: the Architect Agent and the Coder Agent.

Architect Agent

The Architect Agent is powered by a large-context LLM (such as Gemini 2.5 Pro) and uses MCP and Repomapper to deeply understand your project by integrating chat history, codebase, and file structure. Like Cursor and Devin, it reasons holistically across your codebase, generating a clear step-by-step plan, drafting pseudocode, and pinpointing exactly which files and which lines need to be touched.

Key Features of the Architect Agent:

  • Holistic reasoning: The Architect Agent uses a large-context LLM to reason about the entire codebase, ensuring a deep understanding of the project's context and requirements.
  • Clear step-by-step plan: The Architect Agent generates a clear and concise plan for implementing changes, reducing the risk of errors and increasing productivity.
  • Pseudocode drafting: The Architect Agent drafts pseudocode to help the user visualize the proposed changes and ensure they align with the project's requirements.
  • File and line pinpointing: The Architect Agent precisely identifies which files and lines need to be modified, reducing the risk of errors and increasing efficiency.

Coder Agent

Once the user is satisfied with the Architect Agent's proposed plan and approves it, the curated and approved context, limited to what the user has explicitly signed off on, is handed to a dedicated Coder Agent optimized for programming (such as Claude Sonnet). This agent focuses on implementing the changes, generating clean, localized code diffs based solely on the approved scope, and then sending it to Emacs to apply.

Key Features of the Coder Agent:

  • Focused on programming: The Coder Agent is optimized for programming tasks, ensuring that changes are implemented efficiently and accurately.
  • Clean code diffs: The Coder Agent generates clean, localized code diffs based solely on the approved scope, reducing the risk of errors and increasing productivity.
  • Emacs integration: The Coder Agent sends the code diffs to Emacs for application, ensuring a seamless and efficient implementation process.

Benefits of the Proposed Workflow

Our proposed workflow offers several benefits, including:

  • Increased clarity: The Architect Agent provides a clear and concise plan for implementing changes, reducing the risk of errors and increasing productivity.
  • Improved safety: The Coder Agent focuses on implementing changes, reducing the risk of errors and increasing efficiency.
  • Enhanced precision: The Architect Agent precisely which files and lines need to be modified, reducing the risk of errors and increasing efficiency.
  • User control: The user remains firmly in control at every stage, ensuring that changes align with the project's requirements and goals.

Conclusion

Our proposed workflow addresses the issues with traditional workflows by introducing two distinct agents: the Architect Agent and the Coder Agent. The Architect Agent provides a clear and concise plan for implementing changes, while the Coder Agent focuses on implementing those changes efficiently and accurately. By keeping the user firmly in control at every stage, our proposed workflow maximizes clarity, safety, and precision, ensuring a seamless and efficient implementation process.

Future Work

Future work will focus on refining the proposed workflow, including:

  • Integrating additional tools: Integrating additional tools, such as Lovable and Devin, to enhance the workflow's capabilities and efficiency.
  • Improving user experience: Improving the user experience by providing a more intuitive and user-friendly interface for interacting with the Architect Agent and Coder Agent.
  • Enhancing scalability: Enhancing the workflow's scalability to support larger and more complex projects.

Discussion

Introduction

In our previous article, we proposed a new agentic workflow for Emigo, drawing inspiration from Aider's clear separation between Architect and Coding roles, and enhanced with principles seen in tools like Cursor, Devin, and Lovable. In this article, we will address some of the most frequently asked questions about the proposed workflow.

Q&A

Q: What is the main difference between the Architect Agent and the Coder Agent?

A: The Architect Agent is responsible for generating a clear and concise plan for implementing changes, while the Coder Agent focuses on implementing those changes efficiently and accurately.

Q: How does the Architect Agent reason about the entire codebase?

A: The Architect Agent uses a large-context LLM (such as Gemini 2.5 Pro) to reason about the entire codebase, ensuring a deep understanding of the project's context and requirements.

Q: What is the role of MCP and Repomapper in the Architect Agent?

A: MCP and Repomapper are used by the Architect Agent to deeply understand the project by integrating chat history, codebase, and file structure.

Q: How does the Coder Agent generate clean, localized code diffs?

A: The Coder Agent generates clean, localized code diffs based solely on the approved scope, reducing the risk of errors and increasing productivity.

Q: What is the benefit of using a dedicated Coder Agent?

A: Using a dedicated Coder Agent ensures that changes are implemented efficiently and accurately, reducing the risk of errors and increasing productivity.

Q: How does the proposed workflow address the issue of user control?

A: The proposed workflow keeps the user firmly in control at every stage, ensuring that changes align with the project's requirements and goals.

Q: What is the benefit of integrating additional tools, such as Lovable and Devin?

A: Integrating additional tools can enhance the workflow's capabilities and efficiency, providing a more comprehensive and effective solution for implementing changes.

Q: How does the proposed workflow address the issue of scalability?

A: The proposed workflow is designed to be scalable, supporting larger and more complex projects by using a combination of Architect Agent and Coder Agent.

Q: What is the next step in refining the proposed workflow?

A: Future work will focus on refining the proposed workflow, including integrating additional tools, improving user experience, and enhancing scalability.

Conclusion

The proposed workflow addresses the issues with traditional workflows by introducing two distinct agents: the Architect Agent and the Coder Agent. By keeping the user firmly in control at every stage, our proposed workflow maximizes clarity, safety, and precision, ensuring a seamless and efficient implementation process.

Discussion

This Q&A article is a continuation of the Emigo GitHub discussion thread https://github.com/MatthewZMD/emigo/discussions/16. We welcome feedback and suggestions from the community to help refine and improve the proposed workflow.

Additional Resources