Tag: AI

A working vocabulary for building with large language models: models, agents, harnesses, retrieval, training, and the operations around them.

The practice of supplying AI systems with the right background information, in the right format, to produce useful outputs.

The design of the cycle an agent runs: its tools, its feedback, its stopping conditions, and where a human gets to interrupt.

Thinking deliberately about how you work, surfacing habitual patterns and tacit knowledge so you can identify opportunities for improvement or automation.

The practice of structuring instructions to AI systems for reliable, high-quality outputs through deliberate choices about structure, specificity, and iteration.

AI feels like noise, not leverage — scattered usage, no compounding.

Get more from the chat drawer — structured filters, the context meter, and conversation history.

Connect Claude or ChatGPT to the Nerd/Noir collection over Model Context Protocol for search, note retrieval, and resource browsing.

Margaret-Anne Storey's article proposing a triple debt model of technical, cognitive, and intent debt for reasoning about software health in the age of AI.

A 16-question self-assessment that estimates how much of your AI spend is structurally recoverable and ranks the levers that would recover it.

Where LLMs sharpen product discovery and where they quietly hollow it out — a stage-by-stage guide to the dos and don'ts.

Building reusable AI setups (projects, custom instructions, examples) that improve with use, turning one-time prompt wins into durable productivity systems.

Agent skills turn your team's engineering discipline into instructions a coding agent loads on demand. Four public libraries worth reading for reference, and why you should still write your own.

Systematic methods for measuring AI output quality so you can tell whether your prompts, context, and setups actually work.

How to wire a library of example files into Claude Projects, Cowork, and Claude Code so the right exemplars are selected per task and your quality bar compounds instead of resetting every session.

Using a model to build prompts as artifacts, either developing a reusable prompt on purpose or extracting one from a chat that already worked.

Running many short-lived insight-to-experiment tracks alongside a durable delivery track. An AI-era update to Jeff Patton's dual-track model.

Six named structures for prompts — CO-STAR, RISEN, RACE, CREATE, APE, and STOKE — with the components each one forces you to specify and a decision matrix for picking one.

Using AI to analyze successful outputs and reverse-engineer the prompts that would produce them, accelerating prompt development.

Arlo Belshee's two-character commit prefix that declares how risky a change is and what it intends to do, before anyone reads the diff.

A hands-on enablement program where a cohort builds a working AI system around a real problem, learning prompt, context, and loop engineering by applying them.