From zero to hero
Meet the tool, learn to talk to it, organise your work...
A chatbot answers then you're on your own...
A better workflow: you bring context and judgement, then work on any problem together, with a system that understands and speaks your language: "intelligence".
Same "intelligence" whether you reach it on web, desktop or mobile, built on three principles: helpful, harmless, honest. The apps are just front doors.
Talk to it like a capable colleague.
Add context by uploading files, and let memory and styles carry your role and preferences forward so you stop repeating yourself.
The first draft is a starting point. Most disappointing answers trace back to one thing: The prompt was missing something.
Run a quick eval (5–10 of your own examples) to learn where Claude already fits your work.
The desktop app gives you one Claude in three shapes.
Pick by how big and how hands-off the work is.
Cowork and Code run on Claude Code underneath: local to your machine, able to spin up subagents and sustain long tasks. Cowork stays inside the folders you share; Code works directly in your project.
Stop re-uploading the same files and re-explaining the same rules. A project holds them once; every chat inside it starts already briefed.
Best for anything ongoing rather than a one-off.
Name your files descriptively: Claude uses the filename to find the right one.
Not a wall of code buried in the conversation. A finished thing in its own window: rendered, editable, ready to use / download.
Be specific and iterate one change at a time.
A skill is a folder of instructions and resources Claude loads when it's relevant. Two kinds, and one clean rule of thumb.
Only install custom skills from trusted sources: they can include code.
Instead of starting cold every time, Claude works with your actual data: searching, reading, and acting across the apps you already use.
Claude sees ONLY what you can see: connecting your email doesn't expose anyone else's. Permissions are scoped and revocable; only connect tools you trust.
A pre-built project over your organisation's knowledge. "What happened while I was out?" answered from across every connected source.
Team and Enterprise plans only: an admin sets it up, each person authenticates their own accounts. It returns only what you already have permission to see.
Not a single lookup. Claude works agentically: planning, running many searches that build on each other, then writing it up with citations.
Reach for Research when a thorough answer needs many sources. For a quick fact use web search; for pure reasoning use extended thinking; for internal-only questions use Enterprise Search.
The point of all this isn't the feature list, it's the work.
Pick one recurring task this week and try it. The skill sharpens with practice: you build intuition for where Claude earns its place.
claude.ai is one front door. The same Claude shows up inside the tools you spend your day in... each shape tuned to that surface.
Where you go next depends on the work: building software (Code), team context (Slack), models (Excel), decks (PowerPoint), or the live web (Chrome).
Set the stage, define the task, specify the rules > then refine.
Projects, artifacts and skills organise the work; connectors, search and research expand its reach; and it meets you in the tools you already use.
tutto.one/praxis · daniel@tutto.one