An evolving shelf of self-contained learning collections — each one is a folder of interactive HTML guides, quizzes, and flashcards on a specific AI topic. New collections drop in over time; this page is the front door.
Content/ — click any card to open that pageHow agents actually work — from the bare loop to production infrastructure.
The agent loop from scratch — mini-swe-agent, smolagents, and Mini-Agent. ReAct, tools, and the read-think-act cycle laid bare.
Open guide →End-to-end: the harness, CLAUDE.md, Skills, MCP, Hooks, and Subagents — plus enterprise cases for putting it all to work.
Open guide →Memory, sandboxes, KV-cache, long-horizon execution, and the Kimi Swarm — the plumbing behind agents that run for hours.
Open guide →26 questions spanning all three guides, F → A difficulty. Test recall on the agent loop, the harness, and infrastructure.
Open quiz →45 flip-cards across every guide for fast spaced review of the key terms, patterns, and mechanisms.
Open flashcards →Twenty-one hands-on sites on putting AI and Copilot to work inside SharePoint.
How AI in SharePoint processes 12 files and 30 MB of data by writing code in a sandbox instead of guessing in chat.
Open site →Turn ordinary document libraries into Bento grids, blueprints, and whatever else you can describe — "that can't be SharePoint?!"
Open site →One-click buttons inside any list that quietly run AI in SharePoint — for users who never need to know it's there.
Open site →Document library, submissions folder, gap list, expert directory, and context file working as one organism — driven by four AI Skills.
Open site →Group-scope context lets you write learning rules that turn every correction into a permanent lesson. Audit failures, then teach.
Open site →A reusable hiring-pipeline Skill turns messy SharePoint lists into an interactive HTML dashboard — and a PowerPoint on demand.
Open site →A two-guide library on the new --record flag for producing polished demo videos of SharePoint AI in minutes.
Narrow, composable skills draft LinkedIn posts, YouTube descriptions, chapter timestamps, and titles — all from one transcript.
Open site →A three-guide library on the free, GitHub-hosted demo runner — the launcher, the demo file format, and a clean on-stage widget UI.
Open site →Reference files to keep context tight. Permissions to scope the AI's menu. Evals to let skills optimize themselves.
Open site →Two lists, four skills, one calendar, one homepage — built end-to-end with a Ralph Loop in about 20 minutes.
Open site →Build, edit, and list skills with plain-language utterances. Promote any ad-hoc action into a permanent, scoped skill.
Open site →One parent skill, two reference files, three prompts — choosing column formatting, view formatting, or both before it touches your list.
Open site →The Patterns & Practices catalog packaged as a skill, your brand layered on top — every list styled on-brand with no JSON by hand.
Open site →One conference list, three tiers. The Agentic Builder scaffolds it in a prompt; the Ralph Loop plans, builds, grades, and ships it.
Open site →Let SharePoint AI scan a messy library, propose folder schemes, and rename the worst offenders — with you in control of every move.
Open site →AI can summarize a contract in seconds but can't find every one missing a clause. Metadata — extracted at upload — is the missing layer.
Open site →Encode a brand guide as a reusable Skill, run it across a whole library in one prompt, and wire it to a Quick Step button.
Open site →List Creator, Assertion Extractor, and Citation Finder chain together to turn an uncited brief into a fully sourced table.
Open site →A pile of resumes becomes a structured candidate library through seven conversational prompts — extract, categorize, alert, track, color.
Open site →A Review Council skill channels seven reviewer personas — Adversary, Analyst, Storyteller, and more — in a single pass over your content.
Open site →A parent collection for YouTube-based AI learning tracks. Current track: FDE Deep Dive.
Defines forward deployed engineering, where it outperforms vanilla SaaS playbooks, and why the AI era makes it newly relevant.
Open guide →Covers team structure, deployment cadence, customer alignment, and field-to-product translation loops that keep momentum.
Open guide →Explains how to expand contract value and product leverage while avoiding the consulting trap as the motion scales.
Open guide →An 18-card deck across foundations, operating model, and scaling mechanics for rapid review and spaced repetition.
Open flashcards →Anthropic's playbook for building eval suites — from "what's an eval" to a zero-to-one roadmap for shipping with confidence.
Vocabulary first: task, trial, grader, transcript, harness, suite. Meet the three grader families and the split between capability and regression evals.
Open guide →How evaluation looks for coding, conversational, research, and computer-use agents — and why pass^k kills a 75% pass-rate agent for reliability work.
The eight-step playbook: collect tasks from failures, design a stable harness, read transcripts, treat suites as living artifacts. Closes with the Swiss-cheese coverage model.
Open guide →Forty flip cards covering every concept across the three guides. Filter by guide, shuffle the deck, or flip everything at once.
Open flashcards →A five-step protocol that turns "the user complained" into "team X ships fix Y" — three guides grounded in ten research sources.
Atomic dimensions, the five-step closed loop, G-Eval's structured judgment, and the LLM Judge biases — position, verbosity, self-enhancement — with the MT-Bench numbers behind them.
Open guide →Anthropic's grader stratification. Outcome, Answer, and Trajectory rubrics. Evidence gates, UNKNOWN as a first-class output, and pass@k vs pass^k for non-deterministic flows.
The contextual rubric chain, Agentic Rubrics for SWE agents, the four aging signals, and treating rubric changes like system changes — patch, calibrate, shadow, release.
Open guide →All 25 multiple-choice questions in one sitting — live score tracks as you go, with a per-guide breakdown at the end.
Open quiz →47 unique flip cards across all three guides. Multi-tagged cards (like RAGAS faithfulness) appear once with every relevant guide pill.
Stanford's reverse-engineered profile of the 5% of enterprise AI deployments that actually shipped — three guides on the operating habits, workforce choices, and lock-in budget behind every win.
The strict four-part inclusion bar. 51 deployments across 41 orgs, 9 industries, 7 countries. The central thesis — the model wasn't the moat — plus the 42% interchangeability finding and the "weeks vs years" gap.
Open guide →Six operating habits the high performers shared: 77% non-technical pain, workflow redesign (55% vs 20%), escalation-based operating models (71% productivity gain), gatekeeper triage, operational sponsorship, failure-as-precursor (61%).
Open guide →Strategic integration over experiment mode. Workforce strategy as a deliberate choice (45% reduced headcount, others redeployed). Cost-savings floor vs revenue ceiling. Where the lock-in budget goes when the model is interchangeable.
Open guide →All 20 multiple-choice questions in one sitting — live score tracks as you go, with a per-guide breakdown at the end.
Open quiz →32 flip cards across all three guides. Multi-tagged cards (like 42% interchangeability) appear once with every relevant guide pill.
Content/ with its own pages, then add a group section above pointing at them.
Content/<Name_YYYYMMDD>/ and keeps its own index.html as a collection home.Content/ is linked directly here, grouped by collection..site card in the matching .group and point href at it. For a brand-new collection, copy a whole section.group block and update the stats above.cat-agent (slate), cat-applied (amber), cat-advanced (vermillion). Reuse, or add a new .cat-* rule with a fresh --c.