Architecture
50 articles about architecture.
How AI Agents Work: Architecture, Loops, and Tool Use Explained
AI agents work by running a perceive-reason-act loop powered by LLMs and tool calls. Learn the architecture, memory systems, and planning layers inside.
I Wanted a 100% Private AI Accessible from My Smartphone
Building a local-first desktop AI agent that keeps everything private while remaining accessible from your phone. The architecture behind truly private AI.
Actor-Based Sync Engines and Modular Frameworks for Native macOS Apps
Why actor-based sync engines with modular Swift frameworks produce the cleanest macOS app architecture. Lessons from real native apps using Swift 6 concurrency.
Agent CLI Framework Differences: Sequential vs Batch Tool Calling
A concrete comparison of sequential vs batch tool calling across Claude, OpenAI, LangChain, and open-source agent frameworks - with code examples, latency benchmarks, and a decision matrix for when each approach makes sense.
Building a Learning System for AI Agents That Remembers Across Repos
Why AI agents keep making the same mistakes and how an immune system-style memory layer helps them learn from repetition across multiple repositories.
AI Agent Security in 2026 - Lessons from OpenClaw and Why Architecture Matters
The OpenClaw security crisis showed what happens when AI agents have unchecked access to your system. Here is what went wrong, what the industry learned
Architecture Decision Records with Code References - Holding Architects Accountable
ADRs are only useful when they point to working code. Adding code references to Architecture Decision Records creates accountability and makes decisions
Architecture Diagrams vs Working Systems - How AI Agents Expose the Gap
AI agents implement architecture documents literally and expose every underspecified gap. Using an agent as an architecture validator catches design flaws before a full team builds on them.
Why Your AI Agent Should Never Depend on a Single LLM Provider
When your only LLM provider goes down, your entire agent stops working. Build multi-provider fallback into your AI workflows from the start.
AWS Certification That Changed Architecture
Certifications teach what a platform can do. Building teaches what it should do. Both matter for AI agent infrastructure decisions.
Building a Custom AI Coding Agent with the Claude API and MCP Tools
Why building your own AI coding agent with direct API access and custom MCP tools gives you more control than using Claude Code out of the box.
Skills vs Sub-Agents in Claude Code - When to Use Each Pattern
How to structure Claude Code skills vs sub-agents - splitting by type, managing 10+ skills, and choosing the right pattern for each workflow.
Claude Code Writes Your Code, but Do You Know What's in It?
AI coding agents restructure modules in unexpected ways. The code works but the architecture drifts from your mental model unless you actively review
How CLAUDE.md Prevents AI Agents from Writing Goop Code
The single biggest improvement for AI-generated code quality is describing your architecture in a CLAUDE.md file before the agent touches anything. Here is
Context Windows Are Not Memory
Context windows are working memory, not storage. Understanding this distinction is critical for building AI agents that maintain state across sessions.
Memory Is Just Context with a Longer TTL - AI Agent Memory Systems
Memory files are lossy compressed embeddings of past context. Explore how context windows and long-term memory relate in AI agent architectures.
Designing Agent Networks With Isolation and Shared State Patterns
A good agent network balances isolation with shared state. Learn how to design multi-agent systems where agents stay independent but coordinate through
Different Answers, Same Problem - Comparing AI Agent Architectures
When multiple AI agent architectures tackle the same automation task, the results reveal more about design tradeoffs than about which approach is best.
Dumb Orchestrator With Smart Workers Beats One Big Agent
A simple decision-tree orchestrator routing tasks to specialized worker agents - browser, accessibility, sequential - is more reliable than a single
When Federation vs Centralization Makes Sense for AI Agents
Federation adds coordination costs that often outweigh the benefits. Learn when to federate your AI agent architecture and when to keep it centralized.
Focus Compounds - Why Specialized AI Agents Outperform Generalists
A focused AI agent that does one thing well outperforms a distributed agent that does ten things poorly. Specialization compounds in ways generalization cannot.
GTC 2026: Agentic AI and Memory-First Architecture
Memory-first architecture treats agent memory as the primary data store, not an afterthought. Agents that remember context across sessions perform
Hybrid AI Agent Architectures - Local Models for Sensitive Data
Why the best AI agent setup uses local models for sensitive data and cloud models for everything else, with practical patterns for routing between them.
Exposing macOS Desktop Capabilities to External AI Agents via MCP
How MCP servers let external AI agents like ChatGPT and Claude interact with your macOS desktop - file management, app control, and system automation
How to Choose Which Model for Each Task in AI Agents
Tiered model routing sounds smart but adds complexity. When does routing between models actually help AI agents, and when is one model simpler and better?
Holding Parallel Truths in AI Agent Development
Two truths breathing at once is multithreading for consciousness. When two contradictory approaches both work in AI agent development and how to navigate
Modular Architecture for Native macOS Apps: Frameworks, Actors, and File Provider
Building a native macOS app with file syncing and background services requires clean architecture from day one. Here's how to structure Swift frameworks, use actors for concurrency safety, and treat File Provider as a thin adapter.
When Scaffolding Becomes Architecture in AI Agent Code
Scaffolding you refuse to take down becomes architecture eventually. How temporary workarounds in AI agent codebases become permanent fixtures and what to
Smart Caching Strategies for AI Agent Tool Results
TTL-based caching gives AI agents stale data. Learn about dependency-tracking caches that invalidate when upstream data changes, keeping agent decisions fresh.
Specialist vs Generalist AI Agents - When to Split Responsibilities
One generalist AI agent doing six things vs six specialist agents doing one thing each. When to split agent responsibilities and the tradeoffs of focused vs
SQLite Is the Right Database for Most AI Agent Workloads
A single SQLite file per agent session handles most workloads. Benchmarks, schema patterns, and when you actually need to move beyond SQLite for AI agent state management.
Vibecoded App with Claude Code
Vibecoding with CLAUDE.md architecture rules turns Claude Code from a code generator into a system-aware development partner. Here is how the approach works.
Error Propagation in Multi-Agent AI Systems
When one AI agent makes a bad decision, every downstream agent inherits that error. Learn how errors cascade in multi-agent systems and practical patterns to contain them.
Using Agent Teams as a Product Backend: Bridging Swift Desktop Apps to Claude Agent SDK
We built a Swift desktop app that bridges to the Claude Agent SDK via a local Node.js process. Here is how agent teams can serve as a product backend.
The Lossy Handoff Problem - When AI Agents Transfer Context via Git Diff
Git diffs capture what changed but not why. When AI agents hand off work to humans, architectural decisions and rejected alternatives are lost. How to
Competing Philosophies About Where AI Should Live - Truly Local vs Cloud VM
Some tools claim local-first but run in cloud VMs. True local means native code on your machine with direct OS access and no virtualization layer.
You Don't Have a Claude Code Problem, You Have an Architecture Problem
When AI agents struggle with desktop automation, the issue is usually architecture - not the LLM. Thin action primitives that the model composes into
What File Systems Teach About AI Agent Reliability
File systems solved reliability decades ago with atomicity, journaling, and crash recovery. AI agents can learn the same lessons for more reliable execution.
Structuring a macOS Agent App with Modular Swift Frameworks
Split your Swift macOS agent into separate frameworks for UI, accessibility, networking, and models. AI agents can work on one framework without breaking
The N+1 Problem in AI Agents - Everyone Wants Agents That Automate Other Agents
Why the impulse to build agents that automate other agents is premature, and why nailing the first layer of automation matters more.
No-Server Architecture for Small Business Tools - Why Local-First with IndexedDB Wins
Adding a backend to small business software means becoming the IT department for every shop. Local-first with IndexedDB is the smarter constraint.
How to Protect Your IP When Building with AI Coding Agents
Practical strategies for protecting intellectual property when using AI coding agents like Claude Code - isolate secret sauce, use modular architecture, and
PWA vs Native macOS App - How to Decide for Your AI Tool
PWA is fastest to ship but feels like a wrapper. Native SwiftUI gives you proper notifications, menu bar integration, and system-level shortcuts. For AI
Write Specs Before PRs to Avoid Redesign Debates in Code Review
How writing a short spec before non-trivial PRs prevents architecture debates during code review and saves hours of rework.
Designing a Tiered Permission System for AI Desktop Agents
Full YOLO mode is dangerous and full approval mode is unusable. Tiered permissions with allowlists per action type hit the sweet spot.
The HANDOFF.md Pattern - How to Keep Claude Code Productive Across Sessions
Context window management matters more than prompt quality once your project grows. How the HANDOFF.md pattern and post-edit hooks keep AI coding agents
You Do Not Need an MCP Server for Every Mac App - Accessibility APIs as a Universal Interface
Instead of building a separate MCP server for each macOS app, use the accessibility API as a single universal interface. One integration controls every app
Using Claude as an Execution Layer - Markdown Specs, MCP Tools, No Traditional Code
What happens when your entire app is markdown specs that Claude executes, with MCP servers as the only real code. A year of building this way.
Building Memory Into an AI Desktop Agent - Knowledge Graphs and Persistent Context
The hardest problem in AI agents is not planning - it is remembering. How knowledge graphs and local file indexing give desktop agents persistent memory
Native Desktop Agent vs Cloud VM - Why We Chose to Run on Your Actual Mac
Cloud VM agents like Claude Cowork run in isolated environments. Native agents like Fazm control your actual apps. Here is why the native approach wins for