Step01
Set Up Your Infrastructure
Set up a dedicated cloud VPS (Virtual Private Server) as an always-on environment for your agents. On this server, install and configure OpenAI's open source Symphony framework.
AdvancedEngineering
Automate Software Development with an AI Agent Manager using OpenAI Symphony and Linear
Set up a cloud-based 'software factory' where AI agents autonomously handle coding tasks assigned and managed through a Linear project board. This workflow turns you from a coder into an AI agent manager.
Source episode · 02:24
How Alessio Fanelli uses Open AI Symphony for Autonomous Coding and Pokémon Card Trading Workflowswith Claire Vo
From 02:24 to 18:05, Alessio Fanelli demonstrates an agent-manager workflow that pulls work from Linear, runs coding agents, and returns pull requests and previews. Clip range: 02:24 to 18:05.
Before you start
What you need
- Cloud VPS with persistent internet access
- Installed OpenAI Symphony framework
- Linear workspace with workflow columns configured
- GitHub repository with PR creation permission
- Vercel account for preview deployments
What you’ll make
A GitHub pull request linked to a Linear issue with a deployed Vercel preview and agent-generated implementation changes.
Tools used
- GitHub CopilotVisitGitHub Copilot
AI pair programmer
Step by step
The workflow
Follow the sequence once, then adapt the prompts, checks, and handoffs to your own setup.
7 steps
Step02
Create and Assign a Task
On your Linear project board, create a new issue. Describe the coding task in natural language, just as you would for a human developer.
Example prompt
clean up, pretty stable. let's remove the spread column too noisy. Let's also make the that name.
Step03
Trigger the AI Agent
Drag the new Linear issue from your backlog into the 'To Do' column. This action triggers Symphony, which monitors this column, to assign an AI agent to the task.
Step04
Agent Planning and Execution
The agent begins by creating a detailed plan, or 'workpad,' outlining acceptance criteria and validation steps. It then executes the plan, writing and modifying the necessary code.
Step05
Receive Work for Human Review
When the agent finishes its initial work, it opens a pull request (PR) in GitHub, deploys a live preview on Vercel, and moves the Linear ticket to the 'Human Review' column. This signals the task is ready for your inspection.
Step06
Initiate the Rework Loop (If Necessary)
If the code requires changes, leave specific comments on the GitHub PR. Then, move the Linear ticket to the 'Rework' column. The agent will read your feedback, turn it into a checklist, address each point, and submit an updated PR for another review.
Example prompt
In [file_name.js], the new function is missing error handling for null inputs. Please add a check and return an appropriate error. Also, the variable name `temp_val` is unclear; rename it to `user_id` for clarity.
Step07
Complete the Task
When you are satisfied with the changes, approve and merge the pull request in GitHub. The agent will detect the merge and automatically move the corresponding Linear ticket to the 'Done' column.
What good looks like
- Moving a Linear issue to To Do triggers agent execution
- Agent creates a workpad with acceptance criteria and validation steps
- GitHub pull request is opened with code changes tied to the issue
- Vercel preview URL is attached to the pull request
Build your next product with ChatPRD
Turn an idea into a PRD, user stories, and a plan.
After the steps
Runbook notes
How to recover when the loop fails and where human judgment helps.
Recover
If it goes sideways
- Symphony does not detect Linear status changes
- Verify Linear webhook configuration and API token scopes
- Preview deployment fails on Vercel
- Check environment variables and build logs before requesting review
- Agent rework ignores reviewer comments
- Require checklist extraction from each unresolved GitHub review comment
Keep building
Related workflows
intermediateContentJev
How to Analyze YouTube Comments for Audience Insights with AI
Quickly analyze thousands of YouTube comments for sentiment and new content ideas using the Jev AI model. This workflow helps creators turn audience feedback into actionable insights for their content strategy.
Open
advancedProductJevCodex
How to Build a Multi-Model AI Product Insights Engine
Create a powerful product insights engine by combining Jev with other LLMs. Use Jev for large-scale classification of customer feedback and development data, then apply reasoning models to analyze the structured output for deep insights.
Open
intermediateEngineeringJevCodex+1
How to Analyze GitHub PRs to Understand Engineering Effort with AI
Use the Jev AI model to cheaply and quickly cluster thousands of GitHub pull requests into thematic areas. This helps engineering and product leaders understand the breakdown of work between new features, tech debt, and maintenance.
Open
Start shipping
better products.
Join 100,000+ product managers who use ChatPRD to write better docs, align teams faster, and build products users love.
Free to start
No credit card
SOC 2 certified
Enterprise ready