Move AI agents from working demos to operable systems
Connect agent internals, API boundaries, remote transport, and data access to identify the controls production systems need.
GUIDE OUTCOME
The goal is to design for failure, retries, permission boundaries, and audit requirements rather than only the successful demo path.
- Published articles
- 08
- Planned topics
- 03
LEARNING TRACK
Treat the agent as a system
Understand the internal loop and the human role before deciding where responsibility and verification belong.
- AI AgentInside AI Coding Agents: Models, Context, Tools, and the Agent LoopAn AI coding agent does not generate a complete solution in one shot. It combines a model, context management, tools, an execution environment, and safety controls in a repeated cycle of deciding, acting, and observing. This article follows a practical API change to explain that architecture, its failure modes, and how to use coding agents safely and effectively.
- AI AgentInside the Codex Agent Loop: Separating Model Reasoning from Tool ExecutionThis article decomposes the mechanism that lets Codex CLI inspect, edit, test, and revise software into four responsibilities: model, harness, tools, and environment. It follows the tool-call boundary through execution feedback, permissions, MCP, termination, and context management.
- AI AgentHow Software Engineering Changes in the Age of AI AgentsAI agents dramatically reduce implementation costs, but they do not remove responsibility for delivering correct software safely. This article explains why engineering value is shifting from code production toward constraint design, context engineering, assumption review, and evidence-based verification.
- AI AgentWhat Is Vibe Coding? A Beginner’s Guide to Building Apps with AI—Tools, Steps, and RisksLearn what vibe coding originally meant, how it differs from responsible AI-assisted development, and how to build a study timer by talking to an AI. This beginner-friendly guide covers tool selection, reusable prompts, hands-on testing, and the risks of handling personal data, credentials, and production systems.
LEARNING TRACK
Design the integration boundary
Separate API and MCP responsibilities, then expose existing services through a deliberate agent boundary.
- MCPWhat Is MCP? A Beginner-Friendly Guide to MCP, APIs, and Function CallingA beginner-friendly explanation of the Model Context Protocol, how hosts, clients, and servers work together, how MCP differs from REST APIs and function calling, and when building an MCP server makes sense.
- MCPBuild an MCP Server for an Existing REST API: Let AI Manage Tasks with PythonUse a runnable GitHub sample to turn an existing FastAPI task API into an AI-accessible MCP tool with the Python SDK and httpx. The article covers cloning and running the project, schemas, structured output, safe error mapping, stdio, Inspector, Codex, and three-layer pytest coverage.
- MCPHow to Remote an MCP Server with Streamable HTTP: Python, SSE, and DockerMove a stdio-based MCP server to a Streamable HTTP service that multiple clients can access by URL. This guide covers the Python SDK implementation, sessions, JSON versus SSE responses, Docker, Origin validation, and the distinct roles of CORS, TLS, authentication, and authorization.
LEARNING TRACK
Control data access
Replace arbitrary SQL with purpose-built Tools and include authorization, auditability, and transactions in the boundary.
ROADMAP / NEXT
Planned topics
- Idempotency and retries
- Evaluation and observability
- Incident response