Orchestrating AI Agents: Combining Herdr and Firstmate

X postKun ChenTool

Kun Chen outlines the architectural combination of Herdr and Firstmate for multi-agent workflows. Herdr manages agent lifecycles and organization, while Firstmate acts as the primary orchestrator and single point of contact.

Key points

  1. Herdr handles lifecycle management and organizational structure for agent fleets.
  2. Firstmate serves as a single point of contact to coordinate multiple agents simultaneously.
  3. Task briefs are largely constructed deterministically rather than relying purely on LLM token generation.
  4. Multi-machine scaling is natively supported via remote second mates.
  5. Recommended workflow: launch Firstmate, share an initial concept, let it probe and draft a plan, then execute.
  6. Chen advises against manually managing context windows, arguing human attention should not be spent on mechanical operations.

Why it matters

Managing multiple autonomous coding agents across different sessions can quickly overwhelm developers. Pairing a dedicated lifecycle manager with a top-level orchestrator provides a structured approach to scaling multi-agent software development.

Analysis

Chen positions Firstmate as a management layer that only becomes necessary once manual handling of concurrent agent sessions hits practical limits. He highlights that orchestration does not replace rigorous planning: agents drifting off course typically stems from how tasks are specified rather than orchestration failures.

Original post

Kun Chen

@kunchenguid · 9 September 2026

herdr + firstmate is indeed a great combo. that's my daily driver too herdr provides the lifecycle management and organization of all the agents firstmate serves as a single point of contact to orchestrate all of them without going insane t.co/fBvRVeAJEF

687 likes28 replies1k bookmarks74.6k views

View on X
More posts in the thread (9)
  1. @ph000br you mean the "brief"? most of it is deterministically constructed, not all through output tokens but in general yes, it's like hiring a CTO to manage your team. its value comes when you have a ton of things to juggle
  2. @kmore lol that sounds like something's terribly wrong. what did you do with it?
  3. @tharshan_09 it’s already natively supported :) it’s called remote second mates
  4. @tharshan_09 very much depends on what you want to do :) in general i think you’d want something stable and always-on, and capability matches the actual need
  5. @kmore i wonder if you happen to have pulled a bad build - the latest main branch should be pretty good if you want to have another try! firstmate doesn’t automatically solve the “off rails” problem though - that depends more on how you plan and spec
  6. @itsmistermoon no i don’t recommend manually managing it in general, i think time is our most precious and finite resource - i would not spend human time and attention on mechanical operations even if it brings some small gain practically speaking i’d suggest just setting a reasonable
  7. @kmore from what you described, here’s my recommended sequence - 1. start agent harness as firstmate 2. tell it about your rough idea, and ask it to grill you + create plan. it may delegate it to a crewmate or not, either way is okay 3. once the plan is created, just tell firstmate
  8. @Helios_Rises don't use firstmate until you feel the pain of juggling between many sessions and it became super clear that you can't scale further without some help
  9. @nobitoshii the default flow is “just talk to firstmate about whatever you need” don’t worry about the “how” - firstmate will take care of that
Selected replies (5)
  • @ph000br @kunchenguid My experience with first mate was not good. Maybe was something with my workflow. But I requested first mate to verify an error msg and the autonomous agent wrote 1400 words report for a very simple problem. A lot of waste of tokens for a ai-to-ai communication...
  • @Helios_Rises @kunchenguid I love herdr. Shill me on first mate
  • @kmore @kunchenguid My friend, this is exactly the combo I was testing earlier this week. I've been wanting to really test out your Firstmate for awhile. I couldn't have failed more. I do not know why. Everything kept failing, with constant messages from the firstmate of "oh yah I missed that."
  • @tharshan_09 @kunchenguid Now that herdr can do multi machines. Have you thought about Firstmate spanning agent sessions across a fleet of machine - when a local machine resource limits hit
  • @kmore @kunchenguid So maybe it was the wrong kind of process for Firstmate? Dunno. But certainly willing to give it another go. Maybe'll I'll give it less detail, and see what happens with some Fable time.

Mentioned links

Summary generated by AI (Gemini 3.8 Flash, 27 September 2026). It may contain errors – the original source is authoritative.

Related entries

Similar in content, found by the AI search.

  • Video

    Video:David Ondrej

    Agentic Engineering Setup: Kun Chen's Terminal-Based Multi-Agent Workflow

    In conversation with David Ondrej, former Meta, Microsoft, and Atlassian engineer Kun Chen demonstrates his fully agentic development workflow. Rather than manually juggling dozens of agent sessions, he uses his custom tool FirstMate as a primary coordinator. FirstMate delegates sub-tasks to specialised crewmate agents inside Herder, an agent-aware terminal multiplexer running in WezTerm.

    AI & LLMs· Discussion

  • Video

    Video:CURT

    High-Throughput Agentic Engineering by Ex-Meta Principal Engineer

    Kun, former L8 Principal Engineer at Meta, Microsoft, and Atlassian, demonstrates his high-throughput agentic engineering setup. His open-source tool firstmate allows a single lead agent to orchestrate worker agents across dozens of active repositories.

    AI & LLMs· Demo

  • Repository:herdrdev/herdr

    herdr: Terminal Multiplexer and Runtime for Coding Agents

    herdr is a terminal multiplexer written in Rust, built specifically to manage and orchestrate coding agents like Claude Code, Codex, and Cursor. It maintains terminal sessions in the background, tracks agent states, and provides programmatic control interfaces.

    41.3kstarsRust

    AI & LLMs· Tool

  • Repository:Runfusion/Fusion

    Fusion: Multi-Agent Orchestrator as an Automated Software Factory

    Fusion is an open-source TypeScript framework designed to automate the software development lifecycle using AI agent teams. From plain-language task descriptions, agents plan, build, review, and deliver code inside isolated git worktrees managed via a kanban dashboard.

    1.2kstarsTypeScript

    AI & LLMs· Tool

Back to the knowledge base

Let's talk about your project

Get in touch

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.