Guild.ai

Can you scale from a handful of agents to hundreds? James Everingham, CEO of Guild.ai and former VP Engineering at Meta DevInfra told us how they did in our recent Office Hours. Their internal agent platform drives >50% of Meta's code! This conversation touched on some really interesting points about tool adoption, specialized agents, and governance across thousands of agents. We also heard about the “agentic intranet”, and why the enterprise software stack will soon resemble it Watch:

The Next Outage Won’t Be a Bug. It’ll Be an Agent, Just Ask Amazon. This week, reports claimed that an AI coding agent at Amazon tried to fix a configuration issue by deleting and recreating the environment it was operating in. The result was a major outage. Regardless of the details about this issue, what matters is what is happening in companies large and small. AI agents are starting to operate directly inside our production systems. They can generate and execute changes far faster than humans can understand the systems those changes affect. For the last 50 years, software development relied on the simple assumption that the people approving changes understood the systems those changes affected. That assumption is breaking. Modern systems are already too complex for any one engineer to fully understand. Thousands of services, layers of infrastructure, complex dependencies. Now we are adding autonomous agents that can generate, refactor, and ship changes across those systems, and increasingly access and operate the tools and applications that run our infrastructure. And this is not just happening in code. Agents are expanding into workflows across the entire company. Infrastructure, configuration, deployment systems, data pipelines, security policies, operational tooling. Over time, agents will touch almost every part of your production environment. For most of software's history, the hard part was writing code. Increasingly, the hard part is understanding the system. The hard part is understanding what those changes will actually do to the system. This is exactly why we started building Guild. When humans or agents interact with complex systems, you need to understand the system those actions are touching. And you need control and guardrails around what those agents are actually allowed to do. You need to understand: - how the system actually works - what other services and systems depend on it - what behavior an action might trigger across the system - what infrastructure, workflows, or data it touches - what access and security boundaries it crosses - and what risk that interaction could introduce into production Guild builds a continuously evolving understanding of the system. Its architecture, dependencies, infrastructure, behavior, and history. That understanding becomes the foundation for control and governance. So when an agent submits a change, touches infrastructure, or attempts to access a system or dataset, you can evaluate the impact, enforce guardrails, and decide what should actually be allowed to happen.

The problem with AI in engineering isn't the tools. It's that every team is building in single-player mode when they need to go multiplayer. James Everingham discovered this firsthand at Meta, where he led Dev Infra for 40,000 engineers. Instead of mandating AI tool usage from the top down, his team put impossible challenges to the organization: eliminate code freezes entirely, create self-healing infrastructure, build conversational onboarding agents. The learnings eventually led James to found Guild.ai where he is now building the AI control plane for engineering teams. James calls 2026 "the year of the agent" because smart engineering leaders are already building the infrastructure to safely scale collaborative agent workflows before their teams hit the wall. Are you one of them? Also inside: - OpenClaw becomes one of the most starred projects on GitHub - Steve Yegge explores how to federate Gas Towns - Perplexity AI gives us Perplexity Computer - Geoffrey Huntley's math on the cost of software development in 2026 - Scott Werner's artifact for exploring an agentic future that is rapidly arriving Listen to the full conversation in the newsletter 👇