How I run autonomous coding agents from my phone with OpenAI Symphony + Linear | Alessio Fanelli (Kernel Labs) · How I AI summary

Key takeaways

Autonomous management of AI agents using orchestration frameworks like OpenAI Symphony and Linear can vastly improve productivity and scalability for both software development and physical world business operations.
Moving from local setups to cloud VPS environments enables more reliable, persistent, and scalable AI workflows.
Monitoring token costs provides valuable insights into resource consumption and helps optimize AI task efficiency.
Automating mundane or repetitive tasks, such as data collection and price scouting for trading cards or cataloging inventory, can unlock new business models and increase small business competitiveness.
Keeping prompt and instruction files simple, clear, and periodically revisited prevents over-complication and maintains system effectiveness.
Effective AI workflows can democratize business creation, allowing small teams or individuals to compete in markets traditionally dominated by large corporations.

Summary

The Power of Autonomous AI Workflows

Alessio Fanelli discusses how he has transitioned from prompting individual AI agents to managing them as autonomous entities using OpenAI's Symphony framework and Linear. This setup acts as a comprehensive agent management system that automates entire development pipelines and tasks such as coding, review, and deployment without ongoing human intervention. He emphasizes the shift from local runtime setups like Mac Minis to scalable cloud VPS environments, enabling more robust and continuous AI operations.

Implementing Symphony + Linear for AI Management

Fanelli demonstrates how he built a workflow using the open-source Symphony framework, which orchestrates AI agents across projects via a structured markdown specification ('skills.md'). Symphony creates a managed environment where tasks are planned, executed, and reviewed in a loop, with automation from task inception to final approval. He explains the integration with Linear for task tracking and how tasks can be monitored for token costs, revealing insights into the resource consumption of large AI operations.

Use Cases: Coding and Trading Card Business Automation

The podcast explores practical applications of AI automation in the real world. Fanelli showcases how he uses AI for software development tasks such as code generation, rework, and validation, as well as managing a business that trades Pokémon cards. For the trading card use case, AI scans eBay and other sources to find undervalued cards, extracts PSA grades via browser automation, and flags potential deals—illustrating how AI can scale small business operations and create new business opportunities.

Advantages of AI in Small Business Creation and Physical World Intersections

Fanelli advocates for AI's role in democratizing business creation, emphasizing how AI-driven automation enables individuals and small teams to leverage inventory, physical assets, and process inefficiencies. He shares personal anecdotes about cataloging books in his house using AI and discusses how small businesses—like fish delivery—can benefit from automation tools previously impractical at small scales. The discussion highlights AI's potential for broad societal and economic impact, especially in small-scale and inventory-heavy industries.

Managing AI Prompts and Future Directions

Fanelli shares his insights on best practices for working with AI prompts, including strategies for troubleshooting when AI 'goes off the rails'—such as restarting conversations, breaking problems into smaller parts, and adjusting prompts. He notes that the simplicity of the Symphony specification file belies its power and encourages viewers to avoid over-engineering early workflows, focusing instead on clear, concise directives and incremental improvements.

Show notes

Alessio Fanelli, founder of Kernel Labs and co-host of Latent Space podcast, walks us through two very different AI workflows: (1) a fully autonomous coding setup using OpenAI Symphony + Linear, where Linear acts as a state machine and Symphony manages agents through the whole dev lifecycle with zero babysitting; (2) Codex with browser access searching eBay for underpriced Pokémon cards—autonomously browsing, extracting PSA certificate numbers, and flagging deals on $10K–$20K cards for his San Carlos card shop, Merlin Games.

What you’ll learn: