18 posts
Context windows forget. Production agents don't. The difference is a layered architecture: working memory, session memory, and long-term memory split into facts, events, and skills. Here's how real agents remember — and why forgetting on purpose matters.
10 min readA language model without tools is an expensive autocomplete. This post dissects how a production AI harness defines, registers, validates, and executes 40+ tools — from file reads to shell commands to MCP integrations — with type safety, concurrency control, and deferred loading.
9 min readEveryone debates which AI model is best. The real engineering happens in the harness — the production system of tools, permissions, memory, and orchestration that makes any model actually useful. This is a map of that system, drawn from real source code.
9 min readAn AI with shell access and no guardrails will eventually destroy something you care about. This post dissects how a production harness implements layered permissions, hooks, dangerous pattern detection, and trust boundaries — balancing safety with usability.
10 min readThe orchestration loop is the heart of an AI harness — a state machine that coordinates API calls, streaming responses, concurrent tool execution, error recovery, and token budgets. This post traces the complete data flow from user input to final response.
9 min readAd-hoc prompting is fine for one-off questions. Repeatable workflows deserve structure. This post dissects how a production harness defines, discovers, loads, and executes skills — reusable AI workflows that turn tribal knowledge into executable automation.
8 min readThe invisible foundation beneath every AI harness layer: centralized state management, per-model cost tracking, rate limit handling, a custom React-to-terminal renderer, and multiple entry points. This post covers what makes 'works in a demo' become 'works in production.'
9 min readA production AI harness is not single-threaded. Background agents explore codebases, shell commands execute, remote agents run on cloud infrastructure — all while the main conversation continues. This post dissects the task system that manages this concurrency.
8 min readThe model is only as good as the context it receives. This post dissects how a production AI harness constructs system prompts, loads project instructions, manages persistent memory, and compresses context when the window fills — all to give the model the right information at the right time.
9 min readA chatbot answers. An agent finishes the job. The gap between them is not one feature — it's a stack of seven capabilities, built one rung at a time. This series walks that stack in plain English, for engineers and curious non-engineers alike.
9 min readMost Astro tutorials stop at pnpm create astro. By the time you need TypeScript strict mode, Tailwind 4, and a Cloudflare deployment pipeline enforcing Lighthouse 100, you are patching together Stack Overflow answers. This post gets you production-ready from the first command.
12 min readEvery client:load directive you add ships JavaScript to every reader. Most of them do not need it. The difference between Lighthouse 100 and 85 is understanding which components need interactivity and choosing the right hydration strategy for each.
14 min readA skill is not a long prompt. It is a structured workflow with a trigger, a prompt body, references, and an output contract. This post breaks down each component, shows how they interact, and explains the design decisions that separate skills that work from skills that frustrate.
11 min readYour BookStore API now has 200 customers who each want their own data. One missing WHERE clause leaks tenant data and kills the company. Here is how to make cross-tenant data leakage structurally impossible with context propagation, session events, and row-level isolation.
13 min readWhen order placement triggers inventory, email, webhooks, analytics, and distributor notifications — doing it synchronously makes the endpoint slow and couples everything. Domain events decouple 'what happened' from 'what should happen next.'
13 min readEight authentication and authorization methods, one decision tree. Here is the comparison matrix, production combination patterns, and pre-launch security checklist that turns auth knowledge into shipping decisions.
11 min readNot all hashes are created equal. Here is how SHA-256, bcrypt, scrypt, and Argon2id actually work under the hood — the math, the memory, the attacks they resist, and exactly when to use each one.
21 min readMost Python backend projects start clean and become unmaintainable. Here is the folder layout and naming convention that keeps a FastAPI + SQLAlchemy codebase navigable at any size.
7 min read