Posts from LinkedIn
Short takes on architecture decisions, platform engineering and AI-augmented delivery.
Follow me on LinkedInAll posts
newest firstWhat is Agentic AI?
After completing my API Architecture Series, I’m starting a new series focused on Agentic AI. A traditional LLM generally follows: Prompt → Generate Response. An AI agent goes further: Goal → Plan → Use Tools → Act → Observe → Reflect →…
The building blocks of a production AI API
After seven parts, we’ve covered the major building blocks of a production AI API: Part 1 → API + AI. Part 2 → APIs + MCP. Part 3 → AI Orchestration. Part 4 → RAG & Context Engineering. Part 5 → Caching, Cost & Model Routing. Part 6 →…
How do we know what is actually happening inside an AI API?
In Part 6, we looked at security, guardrails and safe AI APIs. Traditional API monitoring is not enough for AI workloads. An AI request can involve: Request → RAG → Model Router → LLM → Tool → Response. Each step can introduce latency…
How Jev works: an interactive demo of fast, structured AI decisions
I built a small interactive demo to make the concept easier to understand. Jev by TypeSafe AI is a System One model designed for fast, structured decision-making. Instead of generating long-form text like a traditional LLM, Jev evaluates a…
How do we make AI APIs secure and controlled?
In Part 5, we looked at caching, model routing, performance and cost optimization. An AI API cannot simply trust every input, model response or tool call. A production architecture needs multiple layers of protection: • Input moderation. •…
How do we make it fast, reliable and cost-efficient?
In Part 4, we looked at RAG and context engineering. Calling the same large model for every request is rarely the right architecture. A production AI API can use: Semantic Cache → Model Router → Guardrails → LLM → Observability. The Model…
How does the API give the LLM the right information before generating a response?
In Part 3, we looked at what happens inside an AI-powered API and the role of the AI Orchestrator. That’s where RAG (Retrieval-Augmented Generation) comes in. A typical flow looks like: Query → Retrieve → Rerank → Build Context → Generate.…
What actually happens inside an AI-powered API?
In Part 1, we looked at API + AI. In Part 2, we looked at why MCP complements APIs. It’s not simply: API → LLM → Response. A production AI API needs an AI Orchestrator to manage: • Context / RAG. • Model routing. • Guardrails. • Semantic…
Why MCP when we already have APIs?
MCP doesn't replace APIs. APIs are built for applications; MCP gives AI agents a standard way to discover tools, resources, prompts and existing services. Application → API, AI agent → MCP → APIs / tools / data.
Traditional API vs API with AI: an architectural shift
An AI-enabled API adds an intelligence layer between the contract and execution, bringing new responsibilities: context management, model orchestration, tool execution, guardrails, semantic caching and cost observability.
Argus: an autonomous QA agent with a hundred eyes
An agent that explores your web app in a real browser, writes a living spec of how it behaves, re-walks flows nightly to catch regressions like silent API 500s, and files findings to Jira. Runs entirely on your machine.
Agentic AI Series
1 partFrom LLMs that answer to agents that plan, use tools, act and reflect toward a goal.
API Series
8 partsWhat changes when an API puts an LLM behind its contract: orchestration, MCP, RAG, cost and model routing, security, and observability.
Traditional API vs API with AI: an architectural shift
An AI-enabled API adds an intelligence layer between the contract and execution, bringing new responsibilities: context management, model orchestration, tool execution, guardrails, semantic caching and cost observability.
Why MCP when we already have APIs?
MCP doesn't replace APIs. APIs are built for applications; MCP gives AI agents a standard way to discover tools, resources, prompts and existing services. Application → API, AI agent → MCP → APIs / tools / data.
What actually happens inside an AI-powered API?
In Part 1, we looked at API + AI. In Part 2, we looked at why MCP complements APIs. It’s not simply: API → LLM → Response. A production AI API needs an AI Orchestrator to manage: • Context / RAG. • Model routing. • Guardrails. • Semantic…
How does the API give the LLM the right information before generating a response?
In Part 3, we looked at what happens inside an AI-powered API and the role of the AI Orchestrator. That’s where RAG (Retrieval-Augmented Generation) comes in. A typical flow looks like: Query → Retrieve → Rerank → Build Context → Generate.…
How do we make it fast, reliable and cost-efficient?
In Part 4, we looked at RAG and context engineering. Calling the same large model for every request is rarely the right architecture. A production AI API can use: Semantic Cache → Model Router → Guardrails → LLM → Observability. The Model…
How do we make AI APIs secure and controlled?
In Part 5, we looked at caching, model routing, performance and cost optimization. An AI API cannot simply trust every input, model response or tool call. A production architecture needs multiple layers of protection: • Input moderation. •…
How do we know what is actually happening inside an AI API?
In Part 6, we looked at security, guardrails and safe AI APIs. Traditional API monitoring is not enough for AI workloads. An AI request can involve: Request → RAG → Model Router → LLM → Tool → Response. Each step can introduce latency…
The building blocks of a production AI API
After seven parts, we’ve covered the major building blocks of a production AI API: Part 1 → API + AI. Part 2 → APIs + MCP. Part 3 → AI Orchestration. Part 4 → RAG & Context Engineering. Part 5 → Caching, Cost & Model Routing. Part 6 →…
More posts
2 postsBuilds, demos and one-off takes.
How Jev works: an interactive demo of fast, structured AI decisions
I built a small interactive demo to make the concept easier to understand. Jev by TypeSafe AI is a System One model designed for fast, structured decision-making. Instead of generating long-form text like a traditional LLM, Jev evaluates a…
Argus: an autonomous QA agent with a hundred eyes
An agent that explores your web app in a real browser, writes a living spec of how it behaves, re-walks flows nightly to catch regressions like silent API 500s, and files findings to Jira. Runs entirely on your machine.