Writing

The standard answer goes like this. The model has a limited context window; your data is bigger than that window, so you retrieve the relevant parts instead of sending everything. It is not wrong. It…

Everyone designs for the crash. Almost nobody designs for the lie. After running multi-agent pipelines in production for a while, I've come to sort failures into three kinds. Only one of them announce…

Somewhere along the way, while reading about and building multi-agent systems, I noticed the variety is mostly an illusion. Underneath, nearly everything is one of five shapes, sometimes combined. Lea…

When I first sketched a multi-agent redesign of one of my pipelines, I did the obvious math and almost abandoned the idea. The pipeline evaluates items in bulk. Say a thousand items. The single-agent…

A child agent knows only what the parent writes into its prompt. That's the sentence. Everything else in this article is just me convincing you to take it as seriously as it deserves. The bug that tau…

You might have spent a good part of this year building AI solutions, and then kept running into the same wall. Not a bug. Not a bad prompt. A wall that's built into the architecture itself. Let me wal…

MCP stands for Model Context Protocol. It is an open protocol that gives AI applications a standard way to connect with tools, data, and reusable prompts. The simplest way to think about it: MCP is a…

MCP becomes much easier once you separate the three ideas: The roles in the system The capabilities a server exposes The lifecycle of a request Together, these form the mental model you need before bu…

One of the fastest ways to misunderstand MCP is to describe it as "REST for AI agents." That sounds convenient, but it misses the point. REST and MCP solve different problems for different consumers.…

MCP servers sit between agents and real systems. That makes them powerful. It also makes them sensitive. If an MCP server can read private data, create tickets, update records, trigger workflows, or m…

Building a basic MCP server can be quick. Building one that agents can use safely and effectively takes design. The biggest mistake is treating MCP as an automatic export of your existing API. A good…

The Very Smart Intern in a Padded Room Imagine you hire an incredibly talented software engineer. They understand every programming language. They can design systems, write tests, debug issues, and ev…

Last time, we saw the 10,000-foot view of Codex: the orchestrator that lets you command an AI agent to build, debug, and explore code. We saw it as a black box with users on one end and executed comma…

Part 3 of "Anatomy of an AI Coding Agent: Dissecting the OpenAI Codex CLI" We've journeyed from the 10,000-foot view of Codex's architecture down to the nervous system that carries signals between com…

We've traced the journey from protocol to nervous system to the beating heart of the orchestrator. Commands arrive. The engine accepts them. But here's the sacred truth that separates a useful AI agen…

Welcome back to our deep dive into Codex, the OpenAI CLI agent that's reshaping how developers interact with AI. In Part 4, we explored how Codex manages execution—sandboxes, approvals, and rollback.…

Welcome back to our journey through Codex's architecture. In Part 5, we explored how Codex manifests in four different frontends—TUI, exec, app-server, MCP server. But we only touched the surface of M…

Hooks, Policies, and the Approval Chain Welcome to Part 7 of our journey into Codex's architecture. We've explored the brain, the nervous system, the vaults, the frontends, and the plugin ecosystem. N…

Standing on the Shoulders of Giants We've climbed seven mountains together. In Article 1, we saw the 10,000-foot view of the four buildings that make up Codex. In Article 2, we traced the nervous syst…

You onboard a new engineer once. You walk them through the architecture, the folder layout, the three things that look weird but exist for a reason, and the rule about never putting money in a float.…

Every team has a style guide nobody fully reads and a linter someone quietly disables when it gets annoying. You write the conventions down. You mention them in code review. And they still erode, one…

You can write a genuinely useful tool, wire it up perfectly, and then watch your AI assistant never reach for it. You ask it to do exactly the thing your tool was built for, and it improvises somethin…

Installing a plugin is one of the most trusting things you do all day, and you probably do it without reading a single line of what's inside. Think about what an install actually means. The moment a p…

Picture the task you have done so many times, your hands move on their own. The PR review. The pre-deploy checklist. The support-ticket triage. None of it is hard. All of it is friction, repeated, for…

Most people think the hard part of AI agents is the model. I don’t. The hard part is giving the agent a real place to work. That’s why I’m excited to share that SprintFoundry now supports Agent Sandbo…