Skip to content
Greg Lum

Learn

Resources

Courses, visual explanations, and practical help for understanding AI. Builder tools and sources follow.

AI learning links reviewed October 27 and 28, 2026. Builder inventory checked October 23, 2026.

Next Steps

Outside courses that go further than the guides. Each lists who it suits, what you need first, the time, and the cost. None is required for the guides.

  • Elements of AI: Introduction to AI

    A broad foundation that is not about any one product

    University of Helsinki and MinnaLearn

    Suits
    Anyone who wants the ideas behind AI, not only tips for a chatbot.
    You'll Learn
    What AI is and how it learns from examples, with exercises and questions about its effects on society.
    Before You Start
    Nothing. The course says it needs no complicated math or programming.
    Time
    Several parts of about 4 to 8 hours each. The provider suggests finishing within six weeks.
    Cost & Access
    Free, including the certificate. Sign up to submit exercises.
    Keep in Mind
    The certificate needs most exercises finished and reviews of other learners' answers. Waiting on reviews can delay it.
    Pairs With
    What do people mean by AI?, Why is everyone talking about AI now?

  • AI capabilities and limitations

    Understand why an assistant gets things wrong

    Anthropic · Claude Academy

    Suits
    People who use an assistant and want to predict where it will slip.
    You'll Learn
    How answers are built a word piece at a time, what a model knows, what its working memory holds, and which instructions tend not to stick.
    Before You Start
    Nothing technical. The AI Fluency course helps but is optional.
    Time
    13 lessons, about 3.5 hours, and a quiz.
    Cost & Access
    Free. Start without an account; sign in to save progress and earn the badge.
    Keep in Mind
    Made by Anthropic, the company behind Claude. The lessons describe language models in general.
    Pairs With
    How does an AI answer get made?

  • AI Fluency: Framework and foundations

    Learn a method for working with an assistant

    Anthropic · Claude Academy

    Suits
    People who want a structured way to decide what to hand to AI and how to judge the result.
    You'll Learn
    Choosing tasks for AI, describing them, judging the output, and taking responsibility for how you use it.
    Before You Start
    The course page lists none.
    Time
    14 lessons, about 4 hours, with a final assessment.
    Cost & Access
    Free. Start without an account; sign in to save progress and earn the badge.
    Keep in Mind
    Written by two university educators with Anthropic and hosted on Anthropic's site. It aims to hold up whatever tools come next.
    Pairs With
    Getting Useful Help from AI

  • Google AI Essentials

    A workplace course with a certificate

    Google · on Coursera

    Suits
    People who want guided practice using AI at work and a certificate they can show an employer.
    You'll Learn
    Five short courses: AI basics, everyday work tasks, writing prompts, using AI responsibly, and keeping up as tools change.
    Before You Start
    No technical experience.
    Time
    Google says under 5 hours. Coursera's listings for the five courses add up to about 8.
    Cost & Access
    Paid through Coursera: $49 a month in the US and Canada after a 7-day free trial; prices vary by country. Financial aid is available. Needs a Coursera account.
    Keep in Mind
    Practice centers on Google's Gemini. The free part is the trial, not the course.
    Pairs With
    Getting Useful Help from AI

  • AI concepts for developers and technology professionals

    Short modules on each kind of AI work

    Microsoft Learn

    Suits
    Readers who like short modules with a quiz, and people heading toward technical work.
    You'll Learn
    Seven modules: an overview of AI, then generative AI and agents, language, speech, vision, pulling data from documents, and looking up sources before answering.
    Before You Start
    Basic computing and math. Later modules build on the earlier ones.
    Time
    Modules are short. Microsoft lists the first at 40 minutes and the one on looking up sources at 34.
    Cost & Access
    Free to read without an account. Sign in to record a passed module assessment on a Microsoft profile.
    Keep in Mind
    Written for technical roles, with examples that point to Microsoft products such as Copilot and Foundry.
    Pairs With
    What do people mean by AI?, Finding Before Answering

  • LLM Course

    Build with open models in Python

    Hugging Face

    Suits
    Programmers who want to load, fine-tune, and share language models themselves.
    You'll Learn
    Working with models and datasets through Hugging Face's libraries, including the tools that split text into pieces, and its public library of models.
    Before You Start
    Good Python. Hugging Face suggests an introductory deep learning course first; no PyTorch or TensorFlow experience is expected.
    Time
    Each chapter is planned as a week of about 6 to 8 hours.
    Cost & Access
    Free, with no ads. Code runs in a hosted notebook such as Google Colab, which needs a Google account, or in your own Python setup. No certificate.
    Keep in Mind
    It teaches Hugging Face's own libraries. The ideas carry over; the tools are one company's.
    Pairs With
    Designing the System Around a Model, How to Read a Model Claim

  • Building effective human-agent teams

    Bring agents into a team's work

    Anthropic · Claude Academy

    Suits
    Team leads whose people already use AI on their own and are considering shared agents.
    You'll Learn
    How a shared agent differs from a personal chatbot, and four working rules: clear roles, a written goal, more work only as the agent proves reliable, and the right access to information.
    Before You Start
    No technical background.
    Time
    5 lessons, about 45 minutes, and a quiz.
    Cost & Access
    Free. Start without an account; sign in to save progress.
    Keep in Mind
    A beta course from Anthropic about team practice, not setup steps for a particular tool.
    Pairs With
    Catching What Agents Miss, One Rulebook for Two Agents

Understanding AI

A different explanation can make an idea click. Choose a course, a visual example, or a question worth following. Check the format and access notes before you start.

  • See how a language model builds an answer, one piece at a time.

    Illustrated article and video · Beginner; some technical ideas

    Public article and embedded video; no course registration

  • An Introduction to Artificial Intelligence

    The Open University / OpenLearn · Microsoft collaboration

    Explore how AI developed and the questions it raises for everyday life and society.

    Self-paced course · Beginner; longer course

    Free public reading; account needed for all activities and completion features

  • ChatGPT Fundamentals

    OpenAI Academy · Product guide

    Get a practical introduction to asking ChatGPT for help and improving an answer through follow-up questions.

    Vendor product guide · Beginner; workplace examples

    Public guide; trying ChatGPT has separate account and product conditions

  • Follow an example of learning patterns from house data, then see where its predictions go wrong.

    Visual interactive article · Optional deeper reading; graph interpretation

    Public page; scrolling graphics; no upload or registration required to read

  • Ethics of AI

    University of Helsinki

    Work through questions about responsibility, fairness and the choices people make with AI.

    Self-paced course · Beginner; reflective longer course

    Public learning material; check the provider for account and completion options.

  • Chatbots Decoded

    Computer History Museum

    Explore the long history of talking machines and the questions today's chatbots raise.

    Online museum exhibit · Beginner; long page

    Public online exhibit with text and optional videos

  • Learn why a familiar voice is not enough to verify an urgent request, and what to do next.

    Consumer advice article · Beginner; short practical read

    Free public article; US reporting links

  • How Content Credentials work

    Content Authenticity Initiative, led by Adobe

    Learn how Content Credentials can show where digital media came from and how it changed.

    Illustrated explainer and FAQ · Beginner; provenance vocabulary

    Public explanation; no upload needed to read

  • There's more to AI bias than biased data

    National Institute of Standards and Technology

    See why unfair outcomes can involve human decisions and institutions as well as training data.

    Public science article · Beginner article; deeper report

    Free public article; linked technical report is optional

  • Understand the physical resources behind AI, including electricity, cooling water and computing hardware.

    University science article · Beginner; longer science explainer

    Free public article

  • Generative AI and jobs: occupational exposure

    International Labour Organization

    Read research on which work tasks AI could affect, with an important distinction between exposure and job loss.

    Research summary and working paper · Optional deeper reading; research terminology

    Public summary and paper links; charts optional

Harnesses and Tools

  • Claude Code

    Anthropic

    My main implementation harness: hooks, plugins, skills, and worktrees.

    In use · Proprietary

  • My second coding agent; its catalog limits informed how much skill metadata I include.

    In use · Apache-2.0

  • Review, hook, plugin, and security plugins, and a read-only Git idea.

    In use · Apache-2.0

  • Evidence rules for frontend testing and for refuting security findings.

    Adopted · Per skill; MIT manifest

  • The SKILL.md format, evaluations with a no-skill control group, and descriptions that help agents select a skill.

    Adopted · Apache-2.0 code, CC BY 4.0 docs

  • Install skills as links; remove stale links only on request.

    In use · MIT

  • RTK

    rtk-ai

    Filtered command output: a hook in Claude Code, an explicit prefix in Codex.

    In use · Apache-2.0

  • Ponytail

    Dietrich Gebert

    Ask whether the behavior already exists before adding code.

    In use · MIT

  • CodeGraph

    Colby McHenry

    Symbol and call-path search in indexed repositories.

    In use · MIT

  • Context7

    Upstash

    Version-specific library docs for agents.

    In use · MIT

  • Parallel Search

    Parallel Web Systems

    Fresh web context with citations.

    In use · Commercial service

  • jevgrep

    dzhng

    Model-assisted code discovery when local search leaves gaps.

    In use · MIT

  • Ollama

    Ollama

    Local models for apps under development and bounded agent helpers.

    In use · MIT

  • MLX

    Apple

    MLX model builds that run under Ollama on Apple silicon.

    In use · MIT

  • Attach images to pull requests, and pass untrusted selectors after --.

    In use · MIT

  • Terraform

    HashiCorp

    Plan to a file, review it, then apply that file.

    In use · BUSL-1.1

Skill Authors

  • Specs from a conversation, writing for agents, and a diagnosis loop. to-spec is a licensed adaptation.

    In use · MIT

  • A Complete Guide to AGENTS.md

    Matt Pocock, AI Hero

    How I organize AGENTS.md and the docs it points to.

    Adopted · Article

  • pstack

    Lauren Tan

    Reviews for common warning signs, checks based on past corrections, and verification skills.

    Adopted · MIT

  • Explain the work before critiquing it, review from independent perspectives, and put each lesson in the relevant instructions.

    Adopted · MIT

  • compound-engineering-plugin

    Every (Kieran Klaassen)

    Remove unnecessary plan steps while preserving every required check.

    Adopted · MIT

  • A skill failure-mode list, blind evaluations, and an installer dry run.

    Adopted · MIT

  • HumanLayer skills

    Dex Horthy, HumanLayer

    Manage what enters the context window and split research into phases.

    Adopted · MIT, Apache-2.0

  • agent-stuff

    Armin Ronacher

    Read a dependency at the version the lockfile pins.

    Adopted · Apache-2.0

  • turbo

    tobihagemann

    Before adding a lesson, check what failed, why it failed, how to fix it, and how much the rule matters.

    Adopted · MIT

  • autoharness

    tigerless-labs

    Look for the rule a new rule replaces.

    Adopted · MIT

  • context-drift

    code-with-rashid

    CI is the authority for commands, and lockfiles for versions.

    Adopted · MIT

  • antfu skills

    Anthony Fu

    Attach before and after images to a pull request.

    Adopted · MIT

  • A narrow diagnostic pilot that became postgres-investigate.

    Adopted · MIT

  • MarkItDown

    Microsoft

    The document extraction workflow.

    Adopted · MIT

  • Data validation ideas behind reconcile-metric.

    Adopted · Apache-2.0

  • The installed frontend-design skill and its list of generic interface patterns to avoid.

    In use · Apache-2.0 per skill; document skills source-available

Testing and Review

  • AI evals skills

    Hamel Husain, Shreya Shankar

    Calibrate a model judge on separate labels.

    Adopted · Apache-2.0

  • Test-case writers see the job, not the skill text.

    Adopted · MIT

  • A parent folder's CLAUDE.md can leak into a control run.

    Adopted · MIT

  • open-steps

    kharmanskyi

    Deny tools that reach other sessions during an evaluation.

    Adopted · MIT

  • Route risky calls to recording mocks.

    Adopted · MIT

  • An output-safety check for skills.

    Adopted · Apache-2.0

  • Review the whole executable surface, and merge findings by identity.

    Adopted · Apache-2.0

  • Try to disprove each candidate finding, and keep a coverage ledger.

    Adopted · CC BY-SA 4.0

  • Findings are confirmed, need validation, or rejected.

    Adopted · MIT

  • Keep comments concise and trace the effects of a breaking change.

    Adopted · MIT

  • A green check counts only if it ran its target.

    Adopted · MIT

  • Plannotator

    backnotprop

    Organize a code review into chapters for the reader.

    Adopted · Apache-2.0

  • pr-review-canvas

    Vinta Software

    Write commit messages for someone reading git log.

    Adopted · Apache-2.0

  • Organize a review guide around the decision the reviewer needs to make.

    Adopted · MIT

  • Explain a change before listing its defects.

    Adopted · MIT

  • PR-Agent

    The-PR-Agent

    Keep a valid finding's wording when re-reviewing.

    Adopted · MIT

  • Two tests overlap only along a traced call path.

    Adopted · MIT

  • The accessibility check in this site's UX gate.

    In use · MPL-2.0

  • Puppeteer

    Google

    Drives the UX gate across widths, themes, and motion settings.

    In use · Apache-2.0

  • Browser inspection from an agent session.

    In use · Apache-2.0

Design and UI

  • Identify where motion helps before adding it, review the result, and check the longest and most difficult content.

    Adopted · MIT

  • Impeccable

    Paul Bakaus

    Preserve the existing design identity and check each flagged pattern in context.

    Adopted · Apache-2.0

  • Check only reachable states, with fixtures.

    Adopted · MIT

  • A worst-case pass on every fixture.

    Adopted · MIT

  • Test proposed rules before adding them to a skill.

    Adopted · MIT

  • hve-core

    Microsoft

    Separate evidence for a claim, a state, and a method.

    Adopted · MIT, CC BY 4.0

  • Reproduce a spec's motion exactly.

    Adopted · No license

  • superdesign-skill

    superdesigndev

    Compare design directions by structure.

    Adopted · MIT

  • Check whether readers understand the rendered page.

    Adopted · MIT

  • design-taste

    madebymustafa

    Compare designs on the same task.

    Adopted · MIT

  • canvas-ui

    DavidHDev

    A semantic fallback for visual effects.

    Adopted · MIT with Commons Clause

  • Paper Shaders

    paper-design

    Pause effects that are out of view, and clean them up.

    Adopted · Apache-2.0

  • uilayouts

    ui-layouts

    Inspect imports, rights, and lifecycle before adopting a component.

    Adopted · MIT

  • Mapping DESIGN.md tokens onto components.

    In use · MIT

  • The UI audit I run on interface source.

    In use · MIT

  • diagram-design

    Cathryn Lavery

    Ask whether a diagram is worth it, and keep a node budget.

    Adopted · MIT

  • Anatomy

    wheresryan22

    One model drives both the motion and the readouts.

    Adopted · MIT

  • Visual Explainer

    Nico Bailon

    Give each figure a specific question to answer.

    Adopted · MIT

  • Explain This

    analyticalmonk

    Take in sources, then reconcile claims against them.

    Adopted · MIT

Motion and Film

  • Remotion

    Remotion

    The film renderer, scenes driven by frame number, minimum text sizes, and safe areas.

    In use · Remotion license

  • The starting point for agent-built films.

    Adopted · Docs

  • Deterministic frames, reading holds, captions, and claims limited to the source.

    Adopted · Apache-2.0

  • ElevenLabs

    ElevenLabs

    Narration with word timestamps.

    In use · Commercial service

  • FFmpeg

    FFmpeg

    Transcodes, waveforms, and checks on the encoded file.

    In use · LGPL, GPL

  • Motion

    Motion

    The site animation library, and its performance guide.

    In use · MIT

  • Eases by direction, exits shorter than entrances, and stagger budgets.

    Adopted · MIT

  • Transitions between key poses, delayed follow-through, and a recipe for subtle movement during holds.

    Adopted · MIT

  • Click timing, and clip reveals for filled shapes.

    Adopted · MIT

  • Rest the cursor before a click.

    Adopted · MIT

  • Counters that round every frame and hold the sourced value.

    Adopted · MIT

  • Perceptual zoom, and caption reading speed.

    Adopted · MIT

  • Minimum hold durations and a check for audio peaks in every encoded file.

    Adopted · MIT (README)

  • How far to lower music under speech and how to measure loudness in the exported file.

    Adopted · MIT (README)

  • Follow brand sources when visual references disagree, and qualify claims to match the evidence.

    Adopted · MIT

  • A critic separate from the builder.

    Adopted · MIT

  • Named seeds, and strips of consecutive frames.

    Adopted · MIT

  • A pass, fail, or unverified record with evidence.

    Adopted · MIT

  • Check consecutive frames across cuts, and transcribe the final mix.

    Adopted · MIT

  • mg-styles-15

    Vincentwei1021

    Shared cues across contrasting styles.

    Adopted · MIT with exceptions

  • The difference between an original, a remake, and a hybrid.

    Adopted · MIT

  • lemo-opuscar

    lemomo-ai

    Style grammar and shot roles.

    Adopted · MIT code, CC BY 4.0 media

  • Shared lyric cues, and rendering in order.

    Adopted · MIT

  • Character acting, and boundaries against stale frames.

    Adopted · MIT

  • Production routes and evidence levels.

    Adopted · CC BY 4.0 annotations

  • How a video prompt breaks into parts.

    Adopted · No license; ideas only

  • Product discovery before a product film.

    Adopted · MIT

  • cinetic

    Leonxlnx

    A concept first, then named picture and sound cues.

    Adopted · MIT

  • GSAP skills

    GreenSock

    Animation lifecycle and scroll geometry. Its timing tables were not adopted.

    Adopted · MIT

  • Motion Prompts

    VanguardiaAI

    Questions about motion mechanics and conflicts.

    Adopted · PolyForm Noncommercial; linked, not copied

  • Collect references and choose the main point before animating.

    Adopted · Post; concept only

  • Brief fields, production gates, and shots with entry and exit states.

    Adopted · Post; concept only

  • One job per reference, and corrections from an independent critic.

    Adopted · Post; concept only

Writing

  • humanizer

    Siqi Chen

    The rewrite step and its lists of AI writing tells.

    In use · MIT

  • no-ai-slop

    Peter Yang

    Separate audit and fix modes, and an evaluation checklist.

    In use · MIT

  • stop-slop

    Hardik Pandya

    Quick checks folded into one search.

    Adopted · MIT

  • Signs of AI writing

    Wikipedia editors

    The basis of humanizer's patterns. Only keywords are kept.

    Adopted · CC BY-SA 4.0

  • Vale

    errata-ai

    Prose linting with write-good, proselint, and readability styles.

    In use · MIT

Articles and Courses

  • Rules for long runs, subagent prompts, and the order of a final report.

    In use · Article

  • 12-factor agents

    Dex Horthy, HumanLayer

    Choose what the agent keeps in its context window.

    Adopted · Code Apache-2.0; content CC BY-SA 4.0

  • Exercises that compare against a reference.

    Adopted · Article

  • Jev review loop

    Joao Queiros, Ray Amjad

    A fast, bounded review pass on each diff.

    Adopted · Article

Also Explored

Skills and tools I read or tried but did not adopt. Some confirmed a choice I had already made.