Tools

Part 7 of 7 — the capstone of the Logic and Reasoning series. The question that's been hovering since Part I: what is the difference between *doing* reasoning and *understanding...

Part 6 of 7. The previous five parts mostly treated reasoning as if it happened in a single moment. This part puts time back in. How a reasoning system learns from new evidence....

Part 5 of 7. A reasoning system — a single mind or a society of minds — can be unhealthy in characteristic ways. Bias, confirmation, polarization, echo chambers, conspiracy thin...

Part 4 of 7. Part III treated a multi-agent problem-solver as an instantaneous artifact. This part puts it in time. What changes when the team has a history, when it has institu...

Part 3 of 7. Single-agent problem-solving from Part II has a ceiling. Multi-agent problem-solving — where the work is decomposed across a team of specialised reasoners — is the ...

Part 2 of 7. Part I covered how a human mind reasons. Part II asks the harder question: how much reasoning is *computable* at all? Gödel, Turing, the halting problem, complexity...

Part 1 of a seven-part field guide on logic, reasoning, and how those capacities scale from a single mind to a society of minds — and what happens when that society is made of A...

Some AI doesn't chat. It learns to imitate a physics simulator so closely that it answers in milliseconds what used to take a supercomputer hours. NVIDIA has built a full ladder...

A real-time walkthrough of the agentic RCA framework running on the battery case from Case 1. Every step is shown: what the perceiver sees, what the grounder retrieves, what the...

The architecture that ties the five-case RCA series together. A human-gated agentic framework that runs the diagnostics-and-prognostics method as a closed loop — perceive, groun...

Part II of the human-in-the-loop monograph. Where Part I covered foundations and mechanisms, this one picks up the harder problem: learning the human's policy with RL, the advan...

The most-overloaded phrase in 2026 agentic AI is 'human-in-the-loop' — it gets used to mean everything from 'a human watches a dashboard' to 'a human approves each tool call' to...

An LLM agent that reasons about a propulsion fault is, by default, doing literary criticism on the symptom — pattern-matching against text it has seen before. A physics-grounded...

Most data-visualization advice in 2026 has the same flaw: it tells you what a good figure looks like and stops short of telling the agent how to produce one. A Copilot skill kit...

A demo reel for a Copilot kit of 26 visual-communication skills — paper-monograph shells, interactive parameter labs, algorithm step players, flowcharts, state-machine animation...

An AI agent that does literature surveys is only as good as the libraries it knows to search. arXiv for preprints. Semantic Scholar for citation graphs. Crossref for canonical m...

The premise is a lens, not a claim: agents do not suffer. But the human nosology of psychiatric disorders turns out to be a remarkably good index of how autonomous systems break...

Emotional intelligence — the ability to read, regulate, and respond to affect — is one of the more interesting capacities to track through the human-to-machine translation. In a...

The security and governance disciplines around AI are not the same discipline and treating them as one produces gaps in both directions. Security is keeping attackers out and li...

The same AI tools that have become indispensable for engineers have also become indispensable for attackers. Phishing emails are now grammatically perfect and personalized at sc...

Training a frontier LLM is a one-time cost, paid by a handful of companies. Inference is the recurring cost, paid by everyone who serves the model — and at scale, inference cost...

Every developer has shipped an application without really understanding what each component of the data stack does — SQL is the query language, SQLite is the embedded database, ...

The expanded edition of the multi-agent-systems introduction. Where the short version sketches the territory, this monograph builds it out: the four canonical roles fleshed into...

The 'one giant agent that does everything' design is a dead end. The pattern that actually works in production is the small team — three or four specialized agents with clear ro...

Every agentic coding system — Claude Code, GitHub Copilot agents, Cursor, Devin, your own harness — runs the same cycle: gather context, take action, verify the result, repeat. ...

When an agent system fails in production, it almost always fails in one of three ways: it loops forever, its plan was wrong from the start, or its tool calls did real damage. Th...

WebGL is the low-level graphics API your browser exposes for talking to the GPU. Three.js is the library that hides 800 lines of WebGL boilerplate behind a 30-line scene-camera-...

There is a particular shape of disappointment that follows a successful agentic AI demo. The demo runs once, looks magical, and then dies the moment it has to run a hundred time...

Volume I established the problem — cascading errors, super-agent bottleneck, missing verification. Volume II is the build manual: context engineering, tool surface design, orche...

The agent-framework landscape in 2026 is crowded enough that engineers are now confidently choosing one and being wrong about the trade-offs. Six frameworks have meaningful adop...

The reason Python won is not that it's the best language at any one thing. It's that it's the second-best language at almost everything, the ecosystem connects across all of tho...

Node.js is the JavaScript runtime that lets the language run outside a browser. npm is the package manager that lets you install other people's JavaScript code as a building blo...

Obsidian started life as a Markdown note-taking app and quietly turned into something stranger and more useful: a personal knowledge network. The killer feature is not the edito...

The single most important determinant of an LLM-powered system's quality is not the model. It's what you put in the context window. This monograph walks through the three archit...

Hermes Agent is one of the more interesting open-source agent projects to ship in 2026: a self-hosted, autonomous AI built by Nous Research around a closed learning loop — every...

Azure documentation is comprehensive in the same way that an unabridged dictionary is comprehensive — every term is defined, nothing is summarized, and you can spend a month rea...

The hardest part of learning the LangChain ecosystem is not any single tool — it is figuring out which problem each one solves and where one ends and the next begins. LangChain ...

VS Code Copilot has, over the past two years, quietly turned into the most-used AI development environment outside of dedicated tools like Cursor and Claude Code. It is also one...

API, CLI, MCP, ADK — four acronyms that get used as if they were rivals competing for the same job. They aren't. They live at different layers of the same stack, and the fastest...

Claude Code is Anthropic's agentic coding tool — a CLI that reads your codebase, edits files, runs commands, and plugs into your dev environment. It is also one of those tools w...

OpenClaw — formerly Clawdbot — is one of the more interesting open-source AI agent projects to appear in the past year: a long-running process that runs on your own hardware, sp...

You pick the same model in VS Code that you use in the Claude app, ask a similar question, and the editor's answer feels thinner. The usual conclusion — 'the IDE version must be...

Architecture decisions are the ones you make in the first week of a project and then live with for years. This primer is the working engineer's introduction — what architecture ...

An LLM reads text. Source code is text. So why isn't this trivial? Because raw .c files come with a preprocessor that has to be expanded before the code is meaningful, with #inc...

A 'skill' is the most underrated primitive in agent design: a folder of files — a SKILL.md, some examples, maybe a script — that tells the agent how to do one thing well. The fo...

There are two different questions you can ask of any LLM-powered system: where did this answer come from, and how long will the model remember it. The first is about knowledge —...

The companion to 'How LLMs See an Image.' Audio is more like text than you'd expect — a 1-D signal turned into a 2-D spectrogram, sliced into patches, and tokenized. Video is au...

There is no version of GPT-4o, Claude, or Gemini that 'looks at' an image the way you do. The model is a transformer; it reads tokens. So every image you upload is quietly turne...

CAN was designed by Bosch for early-1990s passenger cars and has, against all expectations, outlasted nearly every networking technology of its era. It runs the entire embedded ...

AI-assisted coding has changed the daily practice of engineering in ways the older how-to books haven't caught up with. This compendium is the long version of what I've learned ...

A field report from setting up an NVIDIA Jetson Orin Nano Super for AI development. NVMe migration, swap on a 1TB drive, GPU-aware Docker, MAXN power mode, and the small handful...

A 45-minute walkthrough for engineers who keep needing the terminal but never quite sat down to learn it. Five parts: what Linux even is, the file system, the essential commands...

A practical guide for engineers shifting from Claude Desktop to GitHub Copilot in VS Code. The same model, a different room — the three modes, project knowledge, file handling, ...

An interactive primer on CUDA for engineers comfortable with C but new to GPU programming. Twelve demonstrations covering thread hierarchy, memory model, warp execution, memory ...

Ten ideas about the gap between 'works on my laptop' and 'survives in production' — correctness across environments, performance under real load, resilience and operability, and...

Ten ideas about software engineering that have nothing to do with code — what you're actually building, the human surface around it, trust and law and data, and the long tail of...

A 36-chapter reference monograph covering the working vocabulary of the modern software stack — JSON, APIs, regex, parsers, JSON Schema, ReAct, tokens, embeddings, RAG, BM25, Ta...