Novice and intermediate AI learning paths with practical guides — free tools first, then prompting, paid plans and workflow automations.
Use a small synthetic example to learn a reliable request before putting real work material into an approved tool.
Notes: Maya will check the drawing list. The client wants a reply Friday. We still need an owner for the cost estimate.
Turn these notes into a table with columns Action, Owner, Deadline and Evidence. Keep missing owners or deadlines as Unassigned. Do not invent decisions. Add a final list called Open questions.Action | Owner | Deadline | Evidence
Check the drawing list | Maya | Unassigned | Maya will check the drawing list
Reply to the client | Unassigned | Friday | The client wants a reply Friday
Open questions
- Who owns the cost estimate?Compare every owner and deadline with the notes. If the answer adds a person, date or decision that is not in the notes, remove it before sharing.
Practice separating a claim from the evidence that supports it.
Source A: The course starts Monday. Source B: The first live session is Tuesday.
Make a two-row table with Claim, Supporting source and Uncertainty. Use only the supplied text and say when the sources do not agree.Claim | Supporting source | Uncertainty
Course starts Monday | Source A | Source B gives a different date for the first live session.Open the original source beside the table. Remove every claim that cannot be traced to a sentence in the supplied text.
Rewrite a short update while preserving decisions, names, dates and open questions.
Update: Maya approved the draft. We still need the cost estimate. Reply to the client by Friday.
Rewrite this update in plain language. Keep every decision, person, date and open question. List preserved facts underneath.Plain update: Maya approved the draft. We still need the cost estimate. Please reply to the client by Friday.
Preserved facts: Maya; approved; cost estimate missing; Friday deadline.Compare the preserved-facts list with the original. The rewrite must not add a cost, owner or approval that was not present.
Learn to replace sensitive details with a synthetic example before asking for help.
Private example: a client name, project address and contract value appear in a draft RFI.
Create a synthetic version of this workflow using placeholders CLIENT, PROJECT and VALUE. Do not repeat or infer any private detail.Synthetic workflow: Review the RFI for PROJECT, confirm the requested change with CLIENT and leave VALUE as a placeholder until an approved system is used.Scan the prompt and answer for names, addresses, contract values or identifiers. Replace them before using a consumer tool.
Start with the explanation, choose your client, then verify a small reading task.
MCP (Model Context Protocol) lets a compatible AI application use tools and context exposed by a server. Start with public documentation, confirm a real tool call, then connect a small test workspace.
Open guideChoose from nine publisher-maintained MCP servers by task, connection type and first test. Each entry links to the official setup documentation and carries a verification date.
Open guideAdd the public OpenAI documentation MCP to Codex, check the configured server and verify a source-backed tool call before connecting private accounts.
Open guideConnect GitHub's official remote MCP in VS Code, authorise your account and verify a read against one test repository before enabling changes.
Open guideConnect Notion's remote MCP to Claude Code and turn a synthetic page into an evidence-linked action table. Confirm access and source text before attempting a write.
Open guideDiagnose MCP problems in order: configuration, server startup, authentication, tool discovery and one small read. Capture the exact error before changing permissions.
Open guideCompare two synthetic specification excerpts, preserve uncertainty and produce an RFI draft with evidence. You can complete this exercise in an approved chat tool without MCP.
Open guideUnderstand what chat tools can do, start with free tiers and keep human review on anything contractual.
Use it for drafts, summaries and search — not for final legal, safety or clash sign-off.
Open guideCompare ChatGPT, Gemini, Claude, Copilot, NotebookLM and Perplexity free tiers before paying.
Open guidePick one recurring task — meeting notes, email drafts or spec search — and test for a week.
Open guideSummaries and practice questions are fine; submitting generated answers usually is not.
Open guideMove from ad-hoc chat to repeatable prompts, paid plans and agent-style workflows with clear guardrails.
Use role, context, constraints and output format so you edit less on the second pass.
Open guideMatch subscription vs API vs team Copilot to your weekly tasks and the live model value ranking.
Open guideDecide when prompts can stay on your machine versus when cloud tools and files are worth the trade-off.
Open guidePick Cursor, Copilot, Claude Code or Codex as a daily driver when edits span files and tests.
Open guideUse ChatGPT Work, Claude Cowork or Copilot agents when the deliverable needs files, apps or browser steps — with supervision.
Open guideMeeting summaries, spec search and RFI drafts — without replacing model checking or contractual review.
Open guide