recently discovered the most impressive project in the AI coding ecosystem.
Steve Yegge's Gas Town.
For those who don't know Steve Yegge:
- Senior engineer at Google for 10+ years
- Famous developer blogger ("Stevey's Blog Rants")
- Worked on Grok
- Recently co-authored "Wiring the Winning Organization" with Gene Kim
Gas Town, which he released on New Year's 2026, felt less like a tool and more like a manifesto for the future of AI coding.
The Core Insight of Gas Town
"AI agents are ephemeral. But work context should be permanent."
This matters because the biggest pain point we face with AI coding assistants is context loss.
You spend 3 hours designing architecture with Claude yesterday. Open a new session today? Blank slate. MIT Technology Review calls this one of the biggest limitations of LLMs.
Steve Yegge tackled this problem head-on.
Gas Town's Approach
1. Using Git as a trusted state store
Work state can be recovered even after agent crashes or restarts.
2. "Mayor" AI orchestrates multiple agents
20-30 agents working on different tasks simultaneously.
3. "Hooks" persist work state to Git worktree
Embracing agent ephemerality while ensuring continuity.
4. "Convoy" bundles multiple issues for tracking
Visibility into progress on complex projects.
Looking at this architecture, I realized this isn't just an "AI tool" — it's a new development paradigm for the AI era.
steveyegge
/
gastown
Gas Town - multi-agent workspace manager
Gas Town
Multi-agent orchestration system for Claude Code with persistent work tracking
Overview
Gas Town is a workspace manager that lets you coordinate multiple Claude Code agents working on different tasks. Instead of losing context when agents restart, Gas Town persists work state in git-backed hooks, enabling reliable multi-agent workflows.
What Problem Does This Solve?
| Challenge | Gas Town Solution |
|---|---|
| Agents lose context on restart | Work persists in git-backed hooks |
| Manual agent coordination | Built-in mailboxes, identities, and handoffs |
| 4-10 agents become chaotic | Scale comfortably to 20-30 agents |
| Work state lost in agent memory | Work state stored in Beads ledger |
Architecture
graph TB
Mayor[The Mayor<br/>AI Coordinator]
Town[Town Workspace<br/>~/gt/]
Town --> Mayor
Town --> Rig1[Rig: Project A]
Town --> Rig2[Rig: Project B]
Rig1 --> Crew1[Crew Member<br/>Your workspace]
Rig1 --> Hooks1[Hooks<br/>Persistent storage]
Rig1 --> Polecats1[Polecats<br/>Worker agents]
Rig2 --> Crew2[Crew Member]
Rig2 --> Hooks2[Hooks]
Rig2 --> Polecats2[Polecats]
Hooks1 -.git worktree.-> GitRepo1[Git Repository]
Hooks2 -.git worktree.->…I'm Also Building Something Similar
Gas Town is optimized for large teams and complex enterprise projects. Environments where you'd actually run 20-30 agents.
But what about solo developers like me, working on side projects?
I didn't need 20 agents. I wanted deep collaboration with one AI that remembers "where we left off yesterday" and "why we built it this way."
I'm building a tool called CodeSyncer.
Core idea:
- Gas Town stores state in Git
- CodeSyncer stores context in code comments themselves
/**
* Payment processor
*
* @codesyncer-decision [2026-01-15] Chose sync processing (UX priority)
* @codesyncer-inference Minimum amount $1 (PG policy)
* @codesyncer-todo Add refund logic
*/
async function processPayment(amount: number) {
// @codesyncer-why Idempotency key prevents duplicate charges
const idempotencyKey = generateKey(amount);
// ...
}
Next session, AI reads the code and instantly knows "why" it was built this way.
Start in 5 minutes, no complex setup.
npx codesyncer init
npx codesyncer watch
bitjaru
/
codesyncer
Claude forgets everything when the session ends. CodeSyncer makes it remember.
CodeSyncer
Claude forgets everything when the session ends. CodeSyncer makes it remember.
한국어 | English
The Problem → The Solution
| Problem | Without CodeSyncer | With CodeSyncer |
|---|---|---|
| Context loss | Every session = start from scratch |
@codesyncer-* tags = permanent memory |
| Forgot to tag | Changes go untracked |
codesyncer watch catches everything |
| Rules forgotten | AI forgets guidelines mid-session | Hooks auto-remind at optimal moments |
| Dangerous inference | AI guesses prices, security, APIs | Auto-pause on critical keywords |
Demo
Quick Start
# 1. Install
npm install -g codesyncer
# 2. Initialize
cd /path/to/your/project
codesyncer init
# 3. Let AI set up (say this to Claude)
"Read .claude/SETUP_GUIDE.md and follow the instructions"
# 4. Start coding (say this each session)
"Read CLAUDE.md"
Core Features
🏷️ Tag System
Put context IN your code. AI reads code, so it recovers context automatically.
// @codesyncer-decision: [2024-01-15] Using JWT (session management is simpler)
// @codesyncer-inference: Page size…Summary
| Gas Town | CodeSyncer | |
|---|---|---|
| Slogan | Scale your agents | Remember your context |
| Target | Teams/Enterprise | Solo developers |
| Agents | 20-30 simultaneous | 1, deeply |
| State storage | Git worktree | Code comments |
The philosophy is the same: "AI is ephemeral, but context should be permanent."
Just different targets and approaches.
Wrapping Up
In 2026, the real differentiator in AI coding won't be "smarter models" — it'll be "better context management".
Steve Yegge showed the way with Gas Town. I'm trying to make a small contribution for solo developers.
Both are open source. If you're interested in AI coding, check them out.
- Gas Town - For teams/enterprise
- CodeSyncer - For solo developers
If you've been struggling with this too, share your experience in the comments!
Top comments (1)
nice concept "CodeSyncer" i solo devs i like the idea, i use opencode maybe can use whit it