惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Google DeepMind News
Google DeepMind News
F
Fortinet All Blogs
阮一峰的网络日志
阮一峰的网络日志
Apple Machine Learning Research
Apple Machine Learning Research
爱范儿
爱范儿
WordPress大学
WordPress大学
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
J
Java Code Geeks
罗磊的独立博客
S
SegmentFault 最新的问题
V
V2EX
V
Visual Studio Blog
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
美团技术团队
博客园 - 三生石上(FineUI控件)
Stack Overflow Blog
Stack Overflow Blog
Y
Y Combinator Blog
MyScale Blog
MyScale Blog
D
Docker
Google DeepMind News
Google DeepMind News
Blog — PlanetScale
Blog — PlanetScale
M
Microsoft Research Blog - Microsoft Research
Martin Fowler
Martin Fowler
S
Secure Thoughts
B
Blog
cs.CL updates on arXiv.org
cs.CL updates on arXiv.org
www.infosecurity-magazine.com
www.infosecurity-magazine.com
Recent Announcements
Recent Announcements
MongoDB | Blog
MongoDB | Blog
C
Cisco Blogs
C
CERT Recently Published Vulnerability Notes
T
True Tiger Recordings
GbyAI
GbyAI
P
Proofpoint News Feed
P
Privacy International News Feed
Jina AI
Jina AI
The Cloudflare Blog
I
Intezer
AWS News Blog
AWS News Blog
Hacker News - Newest:
Hacker News - Newest: "LLM"
S
Security Archives - TechRepublic
NISL@THU
NISL@THU
The Register - Security
The Register - Security
Recent Commits to openclaw:main
Recent Commits to openclaw:main
P
Palo Alto Networks Blog
S
Schneier on Security
L
LINUX DO - 热门话题
C
CXSECURITY Database RSS Feed - CXSecurity.com
Security Latest
Security Latest
C
Cybersecurity and Infrastructure Security Agency CISA

The New Stack | DevOps, Open Source, and Cloud Native News

I buried 20 problems in a fake P&L to see if Claude for Small Business could find them Why enterprise AI keeps stalling — and how data streaming could unlock it JFrog report recaps a tumultuous year in supply chain security Kore counts down to Artemis, its moonshot for governable AI agents How to build your first end-to-end AI workflow in n8n CI wasn’t built for coding agents. Here’s what comes next. “Morally repugnant shortsightedness”: Why open source security leaders say companies must stop freeloading on maintainers After becoming cloud computing’s telemetry standard, OpenTelemetry graduates into the AI infrastructure era Building the agentic agreement enterprise: How developers are unlocking agentic experiences with Docusign’s MCP server and platform Cut your AI search costs without sacrificing quality NanoCo bets the future of enterprise AI is one sandboxed agent per employee Why six AI labs built the same product for knowledge workers in four months LLMs were trained on an inaccessible web — AudioEye data shows AI is still building one Cursor bets on cheaper coding with Composer 2.5 and Kimi K2.5 At Google I/O 2026, Antigravity gets a new job description Anthropic hires OpenAI co-founder Andrej Karpathy to lead Claude pre-training research Google launches $100 AI Ultra plan and cuts top tier to $200 Google’s Gemini 3.5 Flash beats the frontier models Google now lets developers use GPT and Claude in Android Studio Google wants to make the web agent-ready Google now lets you vibe code native Android apps in AI Studio Valkey just had a 17x year. Its lead maintainer still doesn’t want Redis to die. Anthropic debuts MCP tunnels and self-hosted sandboxes to lock down AI agent infrastructure Why production RAG systems give confident, wrong answers at scale Pulumi bets infrastructure’s next decade belongs to AI agents Why Google’s Remy leaks have enterprise architects rethinking the AI stack GitHub will start paying some bug bounty hunters in swag instead of cash AI security readiness is now the No. 1 obstacle to adoption, Linux Foundation finds The Mac mini just became infrastructure The cleanup cost of AI-generated code GitHub takes aim at Claude Code and Codex with its new Copilot app Forward deployed engineer is AI’s hottest job as OpenAI and Google race to hire. Here’s how to become one. Why Block handed Goose to the Linux Foundation AWS found bugs in 60% of software requirements. Its fix isn’t more AI — it’s a 50-year-old logic engine. The software fix that could shrink AI’s energy bill without new hardware Why AI is failing in the security operations center The hidden cost of build vs. buy for agentic AI in regulated industries OpenAI brings Codex to the ChatGPT mobile app Cloud code: Conductor joins rush toward remote coding agents GitLab is betting a 19th-century economic theory will shape its AI era Anthropic splits billing again: Agent SDK gets separate credit pools The Rust sidecar pattern that fixes Python AI’s biggest weakness Fivetran’s CPO: Closed data stacks won’t survive the agent era MinIO’s MemKV promises 95% better GPU utilization by ending AI recompute tax Red Hat’s skill packs give AI agents something a bigger model never could: 20 years of institutional memory Anthropic’s Claude Code agent view is a better dashboard. So why aren’t developers convinced? OpenAI’s Daybreak and Anthropic’s Glasswing have nearly identical benchmarks — and 3 of the same partners I tested OpenAI’s three claims about GPT-5.5 Instant, and only one fully held up Temporal hits 3,000 paying customers with its crash-proof workflow engine Cloud native application challenges: installing the walking skeleton Cimento emerges from stealth to secure the one thing no firewall can protect Why agent harnesses fail inside cloud-native systems How to build a skills library for your engineering team Why enterprise AI needs customization The new FinOps problem isn’t cloud bills Jensen Huang and Bill McDermott bet on OpenShell to secure enterprise AI agents The API portal is the clearest signal of whether your company can handle AI agents AI is creating a generation of developers who can’t debug their own code Red Hat is betting on AgentOps to close the gap between AI experiments and production AI teams are spending months on web scrapers that SerpApi replaces with one API call Living off the agent: The new tactic hijacking enterprise AI SAP launches AI Agent Hub at Sapphire 2026 to tame vendor agent sprawl SAP launches managed Joule Studio with Cursor and Claude Code support As agentic dev tools boom, workflow auditability becomes the constraint Anthropic’s Claude Platform comes to AWS Anthropic trains Claude to resist blackmail & self-preservation behavior via agentic misalignment How AI-native systems are built Why your AI agent doesn’t actually remember anything Why 157,000 developers are hedging against Anthropic with OpenCode Claude can now follow users across Outlook, Word, Excel, and PowerPoint Why Prometheus couldn’t see Cilium metrics at 2 a.m. Anthropic puts the “myth” in Mythos with its HackerOne bug bounty program The attack surface moved inside the agent. So did Arcjet. Tanzu Platform’s 15-year head start meets the AI moment Datadog and T-Mobile leaders reveal the reality of deploying AI agents in production How Anthropic and Elon Musk cornered Sam Altman this week OpenAI Codex arrives in the browser with new Chrome extension “Several known limitations”: Developers react to Cursor’s promising but still-moving SDK AI startups are scrambling to survive in big tech’s shadow “The terminal still matters”: Amp rebuilds its CLI for an agentic future beyond the command line Anthropic recruited SpaceX’s 220,000-GPU Colossus 1 to fix what Claude users kept complaining about How Microsoft is governing thousands of Kubernetes clusters without manual intervention Temporal reveals serverless option for its Durable Execution platform OpenAI brings GPT-5-level reasoning to its speech models Elastic architects reveal how to query observability data in plain English I tested the new OpenAI Codex features on a real Python codebase, and it’s the strongest Claude Code rival yet GitHub builds an immune system for AI coding agents running on MCP With the launch of Meko, Yugabyte targets the data layer that’s breaking multi-agent AI systems The introverts’ edge: How AI is leveling the developer floor How a Cursor AI agent wiped PocketOS’s production database in under 10 seconds Why long-running AI agents break on HTTP and how Ably is fixing it Anthropic will let its managed agents dream Developers will use whatever AI coding tool they want. ServiceNow is building for that reality. Why Atlassian is letting Claude Code into its own data graph Kubernetes finally lands user namespace support, but shared kernel problem remains The company that made RAG mainstream is now betting against it Why PHP performance keeps getting bumped from the roadmap How NetEase Games cut LLM cold starts from 42 minutes to 30 seconds Why the Linux Foundation adopted MCP, with Jim Zemlin and Mazin Gilbert “AI systems do not understand”: New report flags systemic failures in AI coding
Steve Yegge’s AI agent orchestration project Gas Town comes to the cloud — and brings the Wasteland with it
Paul Sawers · 2026-05-19 · via The New Stack | DevOps, Open Source, and Cloud Native News

Steve Yegge, a longtime software engineer and blogger who spent the best part of two decades at Amazon and Google, has emerged as a major advocate for multi-agent software development — launching Gas Town at the start of 2026, and expanding it into a growing open-source ecosystem that now includes Wasteland and, more recently, Gas City.

To support this effort, Yegge has teamed up with agentic coding company Kilo to move those ideas beyond self-hosted experimentation, toward something with a little more structure.

Founded in March 2025, Kilo is an open-source, model-agnostic coding agent platform designed to work across environments including VS Code, JetBrains IDEs, and the command line. The company’s founding team includes GitLab co-founder and former CEO Sid Sijbrandij, who has positioned the project as an attempt to build a more transparent and extensible alternative to closed AI coding assistants.

Orchestrator mode in Kilo
Orchestrator mode in Kilo

Playing host

At its core, Gas Town is designed as a multi-agent orchestration system for software development. The project splits work across coordinated groups of specialized agents operating simultaneously against the same codebase, with different agents handling coding, testing, orchestration, review, and operational maintenance tasks.

The system organizes those agents into different roles inside persistent “towns,” with long-running agent systems managing engineering work across repositories and projects.

In the five months since launch, Gas Town’s backend has migrated to Dolt, a Git-backed database system that Yegge says was better suited to the project’s coordination-heavy architecture. He also says newer AI models haven’t reduced the need for orchestration in the way he originally hoped, with Gas Town still relying heavily on specialized agent roles — including “Dogs,” which effectively function as reliability and maintenance agents responsible for keeping long-running towns operational.

“The rest of the architecture has been stable,” Yegge explains to The New Stack. “I expected it to get simpler as the models advance, but that didn’t happen.”

Kilo first announced Gas Town by Kilo in early March as a hosted version of the orchestration system, positioning it as a way to remove much of the operational overhead associated with running Gas Town locally, including model routing, persistent infrastructure management, and agent coordination.

The hosted service launched initially through a beta waitlist, with Kilo CEO Scott Breitenother telling The New Stack that thousands of towns were created during that period across both individual developers and enterprise teams.

“You can think of Gas Town as a normal agent harness, but multiplied across many coordinated agents.”

“You can think of Gas Town as a normal agent harness, but multiplied across many coordinated agents,” Breitenother says.

According to Breitenother, the kinds of workloads running through Gas Town broadly mirror those already seen across the company’s other coding agent products.

“There are folks working on greenfield projects, legacy projects, and everything in between,” Breitenother says.

The hosted version essentially combines managed infrastructure, integrated model routing through Kilo Gateway, and a dedicated interface for monitoring agent activity across persistent “towns.” Breitenother says the platform also reduces much of the setup and operational overhead associated with running Gas Town locally, while customers have reported more efficient token usage through features such as suspending idle towns and separating lightweight orchestration tasks from heavier coding operations.

“The value is really the combination of all of it working together. Hosting is part of it, but the bigger thing is making Gas Town actually practical and usable for more developers.”

“I think the value is really the combination of all of it working together,” he says. “Hosting is part of it, but the bigger thing is making Gas Town actually practical and usable for more developers.

For Yegge, meanwhile, the hosted Kilo version, which enters general availability today, has both broadened Gas Town’s reach beyond the original self-hosting community and helped validate the project as a more credible platform for developers and companies evaluating multi-agent systems.

“It lent Gas Town some additional credibility as a usable orchestration solution — with Kilo being independent — and it also opened Gas Town up to a potentially much broader audience of people who prefer to work in an IDE,” Yegge says.

Enter the Wasteland

Alongside the general availability launch of Gas Town by Kilo, the company is also introducing integrated hosted support for the Wasteland — a separate but closely related project intended to link “a thousand Gas Towns” together into a shared trust and coordination network.

As Yegge describes it, the Wasteland allows developers and agents to post tasks, claim work, and validate completed contributions through a shared stamping system built on top of Dolt and Git infrastructure.

“We believe this is the ideally transparent and scalable way for agents and humans to work,” Yegge says.

The Wasteland has already started seeing early organic adoption, and Yegge says they are developing a blockchain-backed work ledger to give the system enterprise-grade tracking and verification.

Next up is Gas City, a broader open-source orchestration framework that Yegge announced at the tail-end of April, and which he says is “the next big step.”

Unlike Gas Town’s more fixed structure of mayors, dogs, and specialized coding agents, Gas City is intended to let users build different kinds of organizational hierarchies and orchestration systems around long-running AI workers.

It was initially built by members of the wider Gas Town community, namely Julian Knutsen and Chris Sells, though Yegge says it has effectively become the ecosystem’s new direction. The project also represents a realization of ideas Yegge first outlined in January around large-scale agent coordination systems.

“I am only lightly affiliated with the code,” Yegge wrote in the Gas City announcement post. “But it’s exactly what I wished for, and it is being run by far more serious and disciplined engineers than me.”

TRENDING STORIES

Group Created with Sketch.