From chatbot to digital coworker in three hours — every Claude feature worth knowing, in the order you'll actually use them.
If you've used a chatbot once, you're ready. We start with how Claude actually works and walk all the way to a digital coworker and a governance brief.
People who want Claude doing real work in their week.
Drowning in spreadsheets, inboxes and admin.
Deciding how the team adopts AI — and defends it.
No technical background needed.
The old "type a prompt, copy the answer" model leaves 90% of the value on the table. Today is about moving from advisor to executor.
Models, the heuristic, the product map.
RCTF prompting, Projects, Memory, Styles, Artifacts.
Cowork, Chrome, the Office add-ins, scheduled tasks.
Connectors & MCP, Skills, Claude Design.
Data, certifications, the honest limits, a brief.
A staged plan you'll actually follow.
Language work — drafting, summarising, reformatting, explaining, reasoning over the material you give it.
Recalling things it was never given. That's hallucination — confident gap-filling, not lying. The fix: give it the source and verify what matters.
In 2026 Claude also uses tools — web search, code, your files, your apps — which removes much of the old weakness. The behaviour is the same: good input, clear job, check what counts.
Fast & cheap. High-volume, low-latency work — classification, summarising, bulk jobs.
The workhorse. The default for most business work. Start here.
The flagship. Hard reasoning, long documents, multi-step agentic work.
Frontier. State-of-the-art — but currently suspended. Don't build on it alone.
Model choice is gears. You don't drive everywhere in fifth — Sonnet for the commute, Opus for the hills, Haiku when you just need to move fast and cheap.
Anthropic launches Fable 5, a Mythos-class model — state-of-the-art on most benchmarks.
Suspended for all customers after a US export-control directive. Opus 4.8 stayed up.
Frontier-model access can be revoked overnight. Never depend on a single model — build with fallbacks.
Think, draft, analyse, make Artifacts. The front door.
Hand it whole jobs; it does the work while you supervise.
The connective tissue that makes it one system.
Every time you copy-paste between Claude and another app, that's a job an executor product should be doing for you.
Who Claude should be — sets the standard and perspective.
Situation, audience, source material. 80% of quality lives here.
One specific, clear instruction.
Exactly what you want back — length, structure, tone.
"Make it good" isn't an instruction — it's a wish. The model is rarely the problem; the prompt is.
= your knowledge. Load documents and instructions once; every chat inside already knows them.
= your continuity. Claude remembers across conversations — view, edit, pause or reset it.
= your voice. Feed it your writing once; every draft starts sounding like you.
Build and iterate real things inside Claude — "add a column", "make it blue". Co-editing, not re-generating. Live Artifacts even refresh their own data.
Download real .xlsx, .pptx, .docx, .pdf files — available to all users, including Free. "Turn this CSV into a model, a memo and a deck."
Take the task you brought. Rewrite the prompt with all four of Role, Context, Task, Format. Run it, compare to your first attempt — then improve it with two follow-ups. Notice how much of the lift is just Context.
Can't touch your files. It advises; you do the multi-step work.
Give it a goal and a folder; it reads, edits and creates files and returns finished deliverables.
Human oversight by design: it shows a plan and waits for approval, and loops you in before anything significant. That gate is your safety belt.
Claude inside Excel, PowerPoint & Word (Outlook in beta). Edits are highlighted for review; nothing goes out until you click send.
A browser agent that acts on your tab. Prompt injection is real (better, not zero). Keep it away from banking, passwords and sensitive data.
Hourly, daily or weekly — a morning brief, a weekly report. Each run is a fresh session.
Cowork tasks run only while the computer is awake and the app is open.
Keep outputs as drafts; watch the first runs; earn autonomy run by run.
If you have the desktop app, point Cowork at a folder of low-stakes files and give it a genuine job. Approve its plan, watch it work, correct it once. Notice it asks permission before anything that matters.
Plug Claude into Gmail, Drive, Slack, Asana and dozens more. MCP is "a USB-C port for AI" — why the list keeps growing. Review every permission.
Graduate your best prompt into a packaged, shareable capability that runs your way every time — install one or build with the Skill Creator.
Describe a deck, one-pager or landing page and refine it live on a canvas. The fastest path from idea to something to show.
Skill = the moves · Project = the knowledge · Connectors = the reach · Cowork = the execution.
Capability is half the decision. The other half is "can I defend this to risk, legal and the board?"
No — not on business plans by default. Team, Enterprise and the API are excluded from training. For sensitive work, use that login.
SOC 2 Type I & II, ISO 27001:2022, ISO/IEC 42001:2023; GDPR via DPA + SCCs; HIPAA BAA on Enterprise.
Ad-free pledge, the Responsible Scaling Policy (active ASL-3), and Constitutional AI — transparent and auditable.
Hallucination (verify what matters) · prompt injection · human-in-the-loop on autonomous runs · vendor risk (keep fallbacks) · benchmark claims are vendor claims.
A one-page summary — plan, data stance, certifications, the guardrails you'll mandate — that turns "I tried a cool AI thing" into "here's a governed plan."
Teaching the limits isn't hedging — it's how you earn the right to recommend the tool.
No new features — just synthesis and commitment. You'll remember 30% of the demos, but keep 100% of the plan you write.
One task daily; Sonnet default; build your prompt library; turn on Memory.
A custom Style; your first Project; file creation on the doc you dread.
A real Cowork job; the Office add-ins or a scheduled task; watch the first runs.
Turn a task into a Skill; finish the Governance Brief; pick your next course.
Plan: start on Pro ($20) → Max if you hit limits weekly → Team for groups → Enterprise for SSO/audit/keys/BAA.
A true, end-to-end account of running a GST-registered Australian company's books — Claude Cowork orchestrating, Claude in Chrome driving QuickBooks Online. Every figure is real.
Drives the QuickBooks UI directly — no API needed.
Orchestrates the multi-step job; a sandbox does the heavy data work.
Pulls receipts and remittance advices to support the work.
The golden rule set up front: "Prior lodged years — do not touch." Watching Claude respect it is the whole lesson.
Three times — the Wise USD opening balance, the prior-year journals, and the BAS — Claude hit something that touched lodged returns or needed professional judgement, and handed back instead of guessing.
AI does the 90% that's tedious, surfaces the 10% that needs a human, and leaves a clean trail for both. Claude is not a registered tax agent — and that's the point.
The people who get value aren't the most impressed — they're the ones who delegate one real task next week. Pick it now, and open the playbook Monday.
18 interactive lessons to revisit, at your own pace, any time.
Mastering AI Tools for hands-on depth, or AI Agents for the frontier.
A private session on one of your real workflows — corporate training.
Rupert Chesman — AI educator, filmmaker, author of The AI-Native Playbook · ru@rupertchesman.com
Get notified when new courses drop and live workshops are announced.
You're in! Keep an eye on your inbox.
No spam, ever. Unsubscribe anytime.