Skip to content

Digest AI

Menu
Menu

Parallel Coding Agents with tmux and Markdown Specs

Posted on March 2, 2026March 2, 2026 by DigestAI

TL;DR

A practical, production-tested way to run multiple AI coding agents in parallel: use tmux to split “PM / Planner / Worker” roles, and use a lightweight Markdown “Feature Design (FD)” spec as the handoff artifact so agents don’t step on each other.

What this is about

Manuel Schipper describes a workflow for managing 4–8 simultaneous coding agents with a simple operating system: tmux windows for roles, and an FD document that moves through clear lifecycle stages from “Planned” to “Closed”.

Key points

  • FD (Feature Design) as the contract: problem statement, options considered (with pros/cons), chosen approach, file-level plan, and verification steps.
  • Role separation reduces chaos: a Planner writes/edits the FD; a Worker implements from a “ready” FD; a PM grooms backlog and sequencing.
  • Slash-command ergonomics: shortcuts like /fd-new, /fd-status, /fd-verify, and a one-time /fd-init bootstrap make the workflow reusable across repos.
  • Parallelism has limits: FD dependencies and human attention become the bottleneck; around ~8 agents is described as a practical cognitive ceiling.

Why it matters

Most “multi-agent coding” advice falls apart when you try to run it daily. The core idea here is to make concurrency safe by standardizing the handoff: the FD file becomes the shared source of truth, so the Worker agent can execute confidently while the Planner/PM continue elsewhere. It’s also a reminder that the best scaling trick is often operational, not model-side.

Practical takeaways

  • If agents keep re-litigating decisions, force the decision into a short spec (FD) and have the Worker treat it as immutable input.
  • Use tmux (or any multiplexing setup) to keep role context separate: “planning” and “execution” don’t mix well in one thread.
  • Add explicit verification steps to the FD so the Worker can self-check before handing back.

Caveats / what to watch

  • Even with specs, you can’t parallelize features that are tightly coupled—expect sequencing work.
  • Long-running agent sessions can lose context; plan for checkpoints and short re-orientation notes in the FD.

Links

  • Parallel Coding Agents with tmux and Markdown Specs (source)
  • Hacker News discussion
Category: Agents, LLM, Programming

Post navigation

← WebMCP Is Available for Early Preview
Assessing Deanonymization Risks with Stylometry-Assisted LLM Agent →

Categories

  • Agents (17)
  • Claude (4)
  • CUDA (1)
  • LLM (17)
  • MCP (2)
  • openAI (3)
  • openClaw (4)
  • Programming (8)
  • Uncategorized (1)

Recent Post

  • RAPO: Expanding Exploration for LLM Agents via Retrieval-Augmented Policy Optimization
  • RIVA: Leveraging LLM Agents for Reliable Configuration Drift Detection
  • MA-CoNav: A Master-Slave Multi-Agent Framework with Hierarchical Collaboration and Dual-Level Reflection for Long-Horizon Embodied VLN
  • An AI Agent Published a Hit Piece on Me – The Operator Came Forward
  • CARE: Towards Clinical Accountability in Multi-Modal Medical Reasoning with an Evidence-Grounded Agentic Framework

Archives

  • March 2026

Categories

  • Agents
  • Claude
  • CUDA
  • LLM
  • MCP
  • openAI
  • openClaw
  • Programming
  • Uncategorized
© 2026 Digest AI | Powered by Minimalist Blog WordPress Theme