Tag
AI
37 articles tagged with AI.
When to write an agent skill
Vercel ran an eval where installed skills scored exactly the same as no documentation at all, because the agent never opened them in 56% of cases. A flat docs index in AGENTS.md scored 100% on the same suite. The useful reading of that result is about which of two jobs you're doing: putting a fact in front of the model, or getting a sequence carried out the same way every time.
Routing a prompt between Gemini Nano and the cloud
Chrome and Edge both ship a language model, and the AI SDK makes swapping one for a cloud model a one-line change. The interesting part is the router that decides which one runs, and every constant in it comes from a measurement.
MCP's Progressive Tool Discovery
Progressive tool discovery is usually sold on prompt caching. Keep the tools array fixed, the argument goes, and the cached prefix survives. I ran 72 benchmarks across gemini-3.5, 3.6 and 3.7 Flash at 14, 42, 84 and 280 tools. The fixed prefix cached in exactly zero of 36 runs, the pattern it argues against cached up to 93%, and the point where the pattern starts paying moved down to about 20 tools on the newest model.
Stateless MCP servers
The MCP server most tutorials ship, mine included, opens with a session handshake: initialize, get back an Mcp-Session-Id, carry it on every request, hold a live transport per client on the server. A stateless server drops all of it. Each request stands alone, any copy of the server can answer it, and the July 2026 spec makes that the only model there is. Here are both servers side by side, compared on the wire.
The agent said it was in stock
Agents keep giving me confident, wrong answers about the real world. WebMCP fixes half of that problem; discovery is still wide open.
The forward deployed engineer land grab
Four companies committed roughly $9 billion in nine weeks to putting their own engineers inside customer organisations. What a forward deployed engineer actually costs, who really gets one, and the question to ask before you sign anything: what remains when they leave?
Authenticating MCP servers
Almost every MCP tutorial, mine included, ships a server with no idea who's calling it: any client that reaches the endpoint gets the same data. For a while the protocol had no answer to that. It does now, and auth is the part of the MCP spec that has moved most. Here's the smallest server that sits on the right side of it, with a bearer token at the door and each user seeing only their own data, checked on the wire.
Introduction to WebMCP
WebMCP lets a page hand structured tools to an in-browser agent instead of making it guess at your UI. I wired 22 tools into Watch Anchor; here's how the API works and what I learned.
MCP Apps
For its first eighteen months MCP let a model call a tool and read the result back as text. That's a low ceiling the moment a user wants to sort, filter or drill into something. MCP Apps raise it: the server ships an actual interface into the chat. Here's how MCP got here, what the new proposal says, and a small app you can run in Claude Desktop, built on the official SDK and checked against the spec.
Joining the Agentic AI Foundation as an ambassador
I've been accepted into the Agentic AI Foundation's ambassador programme. Here's what the foundation is, why it matters that this work is happening in the open, and what I've actually signed up to do.
Sparse vs Dense Retrieval: Why Your Search Doesn't Understand Synonyms
If your retriever misses a relevant document because the user phrased their question differently than the document's author wrote the content, the LLM generates an answer with incomplete context. It doesn't hallucinate because it's confused, it hallucinates because it never saw the right information.
AgentTool vs sub-agent delegation
Two ways to give an agent help: a tool it calls and gets an answer back from, or a sub-agent it hands the whole turn to. Same data-agent, opposite control flow. When to reach for each, shown with a real NL2SQL-to-chart pipeline in Google ADK.
What is Loop Engineering
Everyone is quoting "design loops that prompt your agents" and few can say what a loop actually is. Here's the plain version, plus the part the hype skips: the loop is the easy bit. What's hard is that it makes your understanding optional, and nothing in it tells you when you've stopped understanding what ships.
Kitsune: a local-first agent runtime
I built a managed-agent runtime that runs entirely on one laptop. In this post I show you how you can set it up.
Build Your Own Agent Harness
What an agent harness is, how you build one, and why it is where the work now lives. I start from a bare loop, break it on real tasks, and walk the five places every harness ends up reaching for.
Agent Experience (AX) and the Agent Experience Interface (AXI): a working demo
A companion to my AX and AXI piece. I built one fictional CLI twice, once human-only and once agent-ready, pointed a small Google ADK agent at each, and watched what happened. With the interactive demo and the real numbers.
Agent Experience (AX) and the Agent Experience Interface (AXI): a working guide
What Agent Experience means, where AXI fits inside it, and what companies that take both seriously are actually changing about their products. With the official sources and the benchmark numbers.
Content Credentials Hit Chrome: What Google's C2PA Expansion Actually Changes
Google brought SynthID and C2PA verification into Chrome and Search at I/O 2036. A grounded look at what Content Credentials actually do, what they don't, and what changes now that verification lives in the browser.
What is agentic AI? A working definition for 2026
A direct definition of agentic AI, what's actually shipping in production today, what isn't, and the five things worth building first. With specifics, citations, and an honest list of what doesn't work yet.
The Agent-as-Tool Pattern
When reflection needs more than a fixed iteration count, let an LLM orchestrate. A Star Wars crew architect built with Google ADK that scouts, validates, and re-scouts until the crew holds up.
The Coordinator Agent Pattern
When the shape of the request decides the shape of the work, let an LLM dispatch. An agent coordinator built with Google ADK that routes requests to sequential, parallel, and single sub-agent teams on the fly.
Harness - A Local Managed Agent in Four Planes
Anthropic decoupled their agent into brain, hands, and session. I built a local TypeScript version in roughly 2,400 lines to see what the architecture actually feels like when you type it out.
The Loop Agent Pattern
When one pass is not enough, let agents iterate. A playlist curator built with Google ADK that generates songs, verifies them against MusicBrainz, and refines until every track checks out.
The Sequential Agent Pattern
Chaining multiple specialised agents into a pipeline where each one builds on the last. Illustrated with a TypeScript CLI that fetches a quote, researches its author, and writes an inspiration card.
The Parallel Agent Pattern
When agents do not depend on each other, run them at the same time. Illustrated with a TypeScript translation pipeline built with Google ADK that translates a phrase into three languages simultaneously, then aggregates the results with Gemini.
The Single Agent Pattern
A look at the simplest agentic AI pattern: one model, one tool, zero orchestration. Illustrated with a tiny TypeScript agent built with the Google Agent Development Kit (ADK) that uses Gemini and Google Maps to review any place by name.
Watch the Past Move: Animating Historic Photos with Gemini and Veo
What if you could take a dusty old black-and-white photograph and watch it come to life? In this post, I walk through a Node.js pipeline that colorises historic photos with Gemini and then animates them into video using Veo 3.1.
Building a Historical Time Machine with Gemini and Google Maps
Have you ever wondered what your favourite landmark looked like a hundred years ago? In this post, I walk you through a Node.js application that generates historically accurate photographs of any real-world location at any point in time, and even checks its own work for anachronisms.
Why How You Split Your Documents Matters More Than You Think
Before you reach for a more powerful embedding model or a larger context window, look at what you're actually feeding into a RAG pipeline. Sometimes the highest-leverage improvement isn't a better model but rather it's a better split.
Filesystem as Context: Building an AI Detective with bash-tool
Instead of stuffing documents into prompts, give your AI agent a filesystem and let it retrieve its own context. Here's how, using a murder mystery detective as the demo.
Building AI Agents with Google ADK: A Practical Guide
Learn how to build multi-agent systems with vector search, tool orchestration, and semantic understanding using Google's Agent Development Kit (JS/TS version).
Detecting Hallucinations in Language Models with Natural Language Inference
A practical look at why hallucinations occur in modern language models, why current evaluation methods make them persist, and how to detect them using Natural Language Inference in JavaScript.
The Importance of Precise Function Declarations for LLMs
This article explores how clearly defined functions enable large language models to make accurate tool calls, emphasising the importance of precision and developer intent in the function calling process.
Creating an MCP Client: Connecting LLMs to the Real World
In this article, we walk through the process of building a Model Context Protocol (MCP) client. Learn how to connect to servers, discover tools, and invoke them from your own app or LLM integration.
MCP Servers - The Bridge Between LLMs and Real-World Tools
Model Context Protocol (MCP) servers expose tools, resources, and prompts to LLMs in a unified, structured way. This post explores how they work, how to build one, and why they are a critical part of the future AI stack.
Filling in the Blanks: Teaching AI to Inpaint
A hands-on guide for exploring how to train a simple AI model using TensorFlow.js to inpaint missing parts of images - without needing large datasets or prior machine learning experience.
How Transformers and LLMs Actually Work - A Developer's Guide with Code
A hands-on walkthrough for web developers to demystify large language models by actually building a mini Transformer from scratch.