Master Claude AI: Fable 5, Cowork & Agents for Business

Understand Claude's three doorways — Chat, Cowork, and Claude Code — and learn exactly when to reach for each, so you get the work done instead of just getting an answer.

See what Claude can do in the browser, the Desktop app, and on your phone — which features are tied to each place, and which one to make your home base.

Meet the four models behind every Claude answer — Fable 5, Opus 4.8, Sonnet 4.6, and Haiku 4.5 — and walk away with a simple rule for picking the right one for quality, speed, and cost.

Decode the five Claude plans from Free to Enterprise: what unlocks where, which features in this course are plan-gated, and which plan fits what you actually want to do.

Learn Cowork's two requirements — a paid plan and the Desktop app — and the good reasons behind both, so the hands-on sections never surprise you.

Preview the three ways a Cowork task becomes an automation that runs itself — scheduled tasks, Dispatch, and live artifacts — before you build them for real later in the course.

Understand what makes Claude Fable 5 a first: a Mythos-class model anyone can use, how it relates to Claude Mythos 5, and why the distinction matters for your business.

Get the numbers that define Fable 5 — how much it can read, how much it can write, how it benchmarks, and what it costs — translated into what each one means for real work.

See the four workload categories that a million-token context and long autonomy unlock — jobs that were impossible to delegate before, framed for the person who owns the outcome.

Learn how the Mythos-class safeguards work: when an answer is handled by Opus 4.8 instead, what you'd notice in practice, and why this layer is the price of public access.

Work through the cost decision: the price gap versus Opus 4.8, a worked example in real dollars, the one-line switching rule, and a four-question checklist for which jobs deserve the upgrade.

Know the dates that matter: the free Fable 5 window, what changes when it closes, and the one experiment worth running before it does.

Take a quick tour of the Claude chat interface — including the controls most people never notice — and get the map for every feature covered in this section.

Master the highest-leverage feature in Claude chat: Projects — persistent workspaces that already know your context — and the retrieval technology that loads a whole knowledge base without drowning the chat.

Watch a complete Project get built for a real recurring job, from setup to first deliverable — the exact four-step blueprint you'll reuse in the section assignment.

Stop expecting answers and start expecting things: documents, charts, calculators, and small interactive apps built live in a panel next to your conversation. You'll build three in five minutes.

Learn the two ways to hand off a finished artifact, which one your plan gives you, and the rule that catches most business teams off guard — a short lecture that prevents a real compliance mistake.

Reset the expectation about what Claude can draw — code-built diagrams, charts, and interactive visuals, not photos — and watch three genuinely useful ones get built live.

See how Claude's memory works across conversations: how to view and edit exactly what it has stored, how to take it with you, and when to switch it all off with an incognito chat.

Stop picking between web search, extended thinking, and research by vibes. Get a decision tree that makes the right choice automatic — and learn which two of the three combine.

Watch research run end to end: one real business question in, a cited multi-source report out — including the two verification checks to run every time before trusting the result.

Learn to drop contracts, spreadsheets, and whiteboard photos into Claude and get analysis you can act on — computed with real calculations, not guessed at.

Turn Claude from advisor into producer: real Excel workbooks with working formulas, Word documents, presentations, and PDFs — deliverables you attach and send, not text you reformat.

Delete the habit of retyping 'be concise' into every chat: set your standing preferences once, in one place, so every conversation starts the way you like it.

Close the section with two quick wins: talking to Claude out loud with voice mode, and what happened to the old Styles feature — so a changed menu never throws you off.

Learn to diagnose any disappointing Claude answer: the four fixable prompt problems behind nearly every weak output — and how to name exactly which one you're looking at.

Build the anatomy of a strong prompt — role, task, context, format, constraints — by watching one weak board-memo request get rebuilt part by part.

Steer Claude with the most reliable tool in prompting: a handful of well-chosen examples, positive and negative — for the qualities that are impossible to describe and easy to show.

Upgrade Claude from quick answers to careful analysis with two techniques: step-by-step reasoning, and labeled tags that keep complex inputs unambiguous.

Take precise control of the three output dials — length, tone, and format — using one vendor comparison as the test case. Right answer, right package.

First answers are first drafts. Learn the two moves that follow — course-correct or restart — and, more importantly, how to tell which one the situation calls for.

Install five habits that keep Claude's answers anchored to your actual documents — with exact quotes, citations, and visible gaps instead of confident-sounding guesses.

Get five fill-in-the-blank prompt patterns — summarize, extract, compare, draft, critique — that cover the bulk of what crosses a business desk.

Spot the sign you've outgrown ad-hoc prompting, and climb the four-rung ladder from chat prompt to full automation — with the questions that tell you when to move up.

Watch a genuinely bad Monday-morning prompt get rebuilt live into one that produces a forward-ready document — same task, same data, transformed result.

Learn what Cowork is and how it differs from regular chat and Claude Code — the shift from asking questions to handing over work, and why it changes what one person finishes in a day.

Watch one identical request go to Chat and to Cowork, side by side — and see exactly where the answer machine stops and the coworker keeps going.

Learn the strict order Cowork tries its routes to your tools and information — and why that order is what makes it fast and precise instead of clumsy.

See the full control loop that makes Cowork trustworthy, on a real task: the plan, your approval, the mid-task check-in, and the handback — and how much say you keep at every step.

Know exactly what your AI coworker can touch — granted folders, connectors, and the Office agents — and the finished deliverables it hands back from each.

Run your first complete Cowork task — turning a genuinely messy folder into an organized one — followed closely enough to repeat it on your own machine tonight.

Watch Cowork turn a folder of receipts, scans, and dashboard screenshots into a working spreadsheet — and learn the one verification habit this task demands.

Build the section's flagship workflow: scattered notes in, a polished report in your company template out — and see why this single task wins back hours every month.

Look under Cowork's hood — sub-agents, the isolated virtual machine, and plugins — and understand both why Cowork costs more than chat and where its real safety boundary sits.

Get the honest economics of Cowork: why it consumes your plan faster than chat, three budgeting rules that keep it economical, and the link to the model-cost decision.

Learn the four safeguards built into every Cowork install, what each one actually covers — and the honest reason none of them lets you stop paying attention.

Understand Cowork's most consequential setting — ask before acting versus act without asking — what the fast mode really costs, and a three-part rule for choosing safely.

Build the single most important safety habit in Cowork: the dedicated working folder — and the idea behind it, called blast radius.

Adopt the rule that removes the highest-risk category from a task before it starts: read-only by default — how to phrase it, how to grant it, and where it pays off most.

Understand computer use — the one Cowork capability that deserves more care than everything else combined — and the three properties that make it the highest-stakes tool in the box.

Build the fence around computer use: the permission that guards every app, the blocks Cowork ships with by default, and the four categories you should block yourself.

Become the person in the room who understands AI security: what prompt injection is, how a poisoned document steers an assistant — shown safely, step by step — and why your setup already blunts it.

Build the three-habit defense against prompt injection — watch the task, scope the access, treat content as data — and see why together they remove most of the risk without slowing you down.

Install the habit that defends against something more common than attacks: confident-sounding guesses. Unverified is not fact — so an unchecked claim never quietly becomes a business decision.

Learn why unattended runs deserve a higher bar — and the four rules that keep scheduled tasks safe when Cowork works while nobody is watching.

Close the section with the principle that frames all of it: you remain responsible for actions Claude takes on your behalf — made concrete, and scaled from one person to a team.

Start your first real Cowork workflow: a Monday-brief assistant built so it literally cannot send, change, or delete anything. Part one covers the goal, the rules, and the setup.

Run the Monday-brief assistant for the first time, learn to read its output like an editor, and put it on a weekly schedule so it's waiting for you every Monday morning.

Grade your first workflow against the four governance criteria — a two-minute check that turns 'looks right' into 'verified safe,' and that you'll reuse on every workflow after this one.

Raise the stakes with a marketing drafter that writes in your brand voice — and refuses to let an unchecked claim reach the public. Part one: the inputs and the fact-check gate.

Run the marketing drafter: read a real claims table, route drafts through a review queue, and wire the pattern that matters most — a human always presses publish.

Build the sales workflow: thirty minutes before a call, a one-page account briefing from your CRM export, your notes, and fresh research — read-only, cited line by line, facts walled off from guesses.

Automate the chore nobody loves: the recurring metrics spreadsheet, refreshed on a cadence — without ever touching your formulas, and with a change-log that makes every run reviewable in thirty seconds.

Collapse the four build-alongs into one repeatable five-step method that turns any routine into a governed Cowork workflow — plus the two questions that tell you where to start.

Meet the five building blocks that turn a general assistant into a specialist for your business — what each one is for, and when to reach for which.

See the whole extension system in a single diagram, built one piece at a time: how a skill, a connector, a sub-agent, and a hook fit together inside a plugin, around one real business request.

Get the availability map: which building blocks work in chat, which only run in Cowork — and what that means when you choose where a piece of work should happen.

Understand what a skill is, how it differs from a prompt or a Project — and why it's the highest-leverage thing you can build in Claude.

Learn the four kinds of skills — two you mostly consume, two you create or receive — and discover that one of them has quietly been working for you since early in the course.

Walk every surface where skills run — and see why the honest answer is: essentially everywhere you work with Claude, on every plan.

Fix the most common skills support question in two minutes: how to switch skills on properly, on every plan — and the one capability requirement behind it all.

Feel a skill work before you build one: a real, opens-in-PowerPoint presentation from a single prompt — and the exact triggering behavior your own custom skills will inherit.

Adopt the two rules that make installing skills safe — trusted sources only, and read before enabling — and know exactly what to look for before switching any skill on.

Learn the six-step method that turns any repeatable routine into a skill Claude runs your way, every time — no coding required.

Build a real skill from scratch, part one: the pick, the map, the working session, and the skill file Claude drafts for you — the first four steps of the method, live.

Finish the build: test the skill, watch one failure get found and fixed, then package it the way Claude expects, install it, and use it for real. By the end, it's yours for keeps.

Write the one field that decides whether your skill ever gets used — the description — with real good and bad examples, and a testing loop you can run in minutes.

Make your skill stop being fragile: the edge cases that break first drafts, the testing discipline from Anthropic's own guidance, and a simple versioning routine that keeps it improvable.

See the idea that turns skills into a management tool: a skill encodes rules, not just formatting — like 'never invent facts,' enforced automatically on every run.