How I run autonomous coding agents from my phone with OpenAI Symphony + Linear
TL;DR
From prompter to manager: The key shift is moving away from prompting agents step-by-step to managing them through a state machine like Linear, where you create tasks and let agents execute autonomously.
Symphony orchestrates the full lifecycle: OpenAI's open-source framework monitors Linear boards, spins up Codex agents when assigned work, and manages everything from task creation through PR review and rework.
Token usage reveals task complexity: One task consumed 221 million tokens because it required rewriting storage and deployment for Vercel, proving that token counts are a proxy for how many issues an agent encountered.
Mobile-first agent management: The entire workflow runs from a phone by creating Linear tasks, reviewing PRs, and moving items between states like human review and rework.
Real business applications beyond coding: Agents browse eBay to find underpriced Pokemon cards by comparing PSA certificate numbers and market prices, automating work that would take humans hours.
Purge your markdown files regularly: Models tend to add instructions rather than remove them, so agent instruction files accumulate confusing contradictions over time.
Summary
Alessio Finelli shows how he runs autonomous coding agents from his phone using OpenAI Symphony and Linear, managing entire software workflows without touching a keyboard, and applies the same approach to hunt for underpriced Pokemon cards worth thousands of dollars.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
The Retirement Email Isn't a Warning
Model retirements now arrive every few weeks; the config-eval-rehearsal loop turns each deprecation email from a fire drill into an afternoon swap.
Playbook
The Cheapest Model That Passes
OpenRouter lists 400 models behind one API. The fix for choosing isn't a better leaderboard, it's a four-step protocol that ends in a real eval.
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.