A completed December 2025 writing project: 31 entries about practical AI use, prompting systems, and workflow design.
Completed Series ā¢ Dec 2025
Introducing the article-spec-pack pipeline that powers this entire campaign.
Live Day 02Swap the sci-fi shorthand for a working definition of AI you can actually use.
Live Day 03Why AI works differently than every other tool you useāand why that matters.
Live Day 04Four copy-paste AI prompts for developers, project managers, creatives, and learners.
Live Day 05AI doesn't know what it doesn't know. Learn to spot hallucinations, bias, and verify facts.
Live Day 06AI has no long-term memoryālearn to manage context windows and keep the model on track.
Live Day 07Structure your prompts with Context, Role, Instruction, and Task Format for professional outputs.
Live Day 08Make the model think before it talks using CoT, RAG, and ReAct to handle complex work without hallucinations.
Live Day 09A practical framework to match models to the job across speed, depth, and cost tradeoffs.
Live Day 10Use AI as a critical thinking partner to falsify assumptions and debug faster.
Live Day 11Durable principles for working with AI so you can adapt as models and tools change.
Live Day 12A three-stage view of pre-training, fine-tuning, and RLHFāand what each stage explains about hallucinations and overconfidence.
Live Day 13Multimodal AI in real workflows: screenshots, diagrams, and evidence-first prompting.
Live Day 14How to turn a generic first draft into something specific using constraints and proof artifacts.
Live Day 15Design prompts like interfaces: inputs, constraints, outputs, and verification.
Live Day 16Turn messy threads into briefs, compare options, and run pre-mortems that catch failure early.
Live Day 17Practical guardrails: prompt injection fences, tool contracts, and evidence-based answers.
Live Day 18A clear gate for when to use AI, when to keep humans in charge, and when to say no.
Live Day 19How to design safe agents with scopes, receipts, and verification habits.
Live Day 20How to make your AI workflows durable: interfaces, evals, and a āplan, patch, proveā loop.
Live Day 21Sometimes you just need a quick fixāfast levers to pull for immediate output upgrades.
Live Day 22A useful frame: math over mind-readingāa mental model that matches how the machine actually behaves.
Live Day 23MoE is one of the most important scaling ideas in modern language models.
Live Day 24Long context doesn't remove the need for structureāit increases it.
Live Day 25Most AI problems that feel like "model quality" are really systems problems.
Live Day 26You tested with three prompts, it looked greatāthen you shipped without evals.
Live Day 27Retrieval doesn't make the model truthfulāit gives it more plausible text to be wrong with.
Live Day 28You give a model a toolāthen one day it does the thing you did not mean.
Live Day 29In incident response, the goal is a sequence of safe, reversible steps that reduce uncertainty.
Live Day 30The goal is docs that behave like a product: versioned, reviewable, and tested against reality.
Live Day 31A practical close-out on what worked, what surprised me, and the systems worth keeping.
Live