The biggest productivity gain from Claude isn't better prompting — it's the tools you already have access to. Projects, Skills, Memory, Connectors, Deep Research, Extended Thinking, Artifacts, Cowork, and Claude Code, organized into four levels so you know what to turn on first.
After a year of daily use, the single biggest upgrade in my Claude workflow wasn't a prompting trick. It was finally configuring the tools that ship with every paid account and that almost no one touches.
This guide walks them end to end, grouped by how much leverage they give you. Start at Level 1 and don't skip ahead.
Level 1: the foundation everyone should set up
These are not optional. Skip them and you will re-explain context every session forever.
Projects
Projects is the single most under-configured feature in Claude. Without it, every conversation starts from zero — Claude doesn't know your voice, your codebase, your standards.
Spend 20 to 30 minutes collecting the context files you repeatedly paste and upload them into a Project. The quality delta shows up on the next message.
Claude Skills
If a Project is the environment, a Skill is a saved workflow that runs inside it. A Skill is a markdown file with pre-loaded instructions. Invoke it and Claude follows them instantly.
Example: a "Brand Voice" Skill that encodes tone, audience, banned words, and formatting. Instead of re-explaining, you write "use my Brand Voice Skill to draft X."
To build one: Customize → Skills → enable Skill-Creator. Then tell Claude "help me build a Skill for process X."
Memory
Claude is already accumulating memories about you across conversations. Most people never look at them.
Open Settings → Memory. Delete what's stale. Add the long-lived context you want it to keep.
Connectors
Connectors wire Claude into the tools you already use — Gmail, Google Drive, Slack, Notion, Calendar, and 50+ others. Settings → Connectors. Plug in whatever you touch daily.
Level 2: research and thinking
Foundation in place, now you use it.
Deep Research
Most people know it exists and ignore it. Activate it and Claude decomposes your query, reads dozens of sources, cross-references them, and returns a cited report. Duration: five to 45 minutes depending on scope. Open the chat, click "+", turn it on.
Extended Thinking
Claude answers fast by default. For strategy, analysis, or multi-step reasoning, you want it slower and more deliberate.
Two ways to enable:
- Phrase it: "think deeply before responding."
- Toggle Extended Thinking in the chat UI under the model picker.
Pair heavy reasoning with the strongest model available.
Artifacts
Artifacts are standalone outputs — HTML pages, React components, docs, diagrams, tables — that live as real files you can view, edit, download, and iterate. Ask Claude to "make this an Artifact" and you unlock a much wider surface area: landing pages, dashboards, small apps.
Artifacts compound with Skills. A Skill encodes how you want something built; Artifacts are where it lands.
Level 3: agentic tools
This is where Claude stops being a chat and starts being a teammate.
Claude Cowork
Cowork is desktop-only. Once installed, you hand Claude multi-step jobs that touch your real workflows. Three features earn their keep daily:
- Scheduled Tasks. Claude runs on a cadence — morning briefings from Gmail and Calendar, weekly research sweeps, whatever is recurring.
- File Access. The real differentiator. Grant Claude access to specific folders on your machine and it can read, edit, and work locally inside those workspaces.
- Plug-Ins. Skills on steroids: where a Skill automates one task, a Plug-In bundles several into a role.
Cowork Dispatch
Dispatch runs Cowork from your phone. Leave the laptop open, walk away, keep sending tasks. Open "Dispatch" inside Cowork and follow the prompts.
Claude in Chrome
This connector lets desktop Claude launch a task and execute it in the browser without you switching windows. Useful for anything research- or form-driven.
Level 4: building and coding
Claude Code
Claude Code is the most capable coding agent available right now. What it actually does well:
- Hunts down complex bugs across a codebase.
- Ships whole sites and apps.
- Writes and runs tests end to end.
- Runs security audits.
- Plans multi-session work before touching code.
Slash commands
A built-in command system. The ones worth memorizing:
/help— list all commands./clear— reset the current session's context./compact— compress history to reclaim context./init— generate a CLAUDE.md from the project./memory— edit the memory file./model— switch between Opus, Sonnet, Haiku./cost— current token spend for the session./review— review a pull request./mcp— manage MCP servers./resume— continue a previous session./doctor— diagnose environment problems.
You can also define your own. Drop a markdown file in .claude/commands/ (project-scoped) or ~/.claude/commands/ (global). The file's contents are the prompt that runs when you invoke it. Perfect for /deploy, /test-all, /fix-lint.
CLAUDE.md
A markdown file at the root of any project. Claude Code reads it automatically every session. Inside: coding standards, architectural decisions, file layout, forbidden zones.
This is the single biggest quality upgrade most Claude Code users haven't made.
A starter example:
This is a Next.js project with TypeScript and Tailwind.
Always use functional components. Never class components.
Run npm run lint before committing.
All API calls go through /lib/api. Don't call APIs directly
from components.
Don't modify /config without approval.Multi-agent mode and subagents
Claude Code can spawn subagents in parallel — one writing tests while another builds the feature, one refactoring while another writes docs. It decides when to spawn them based on task complexity.
How you trigger them: give outcome-oriented prompts, not step-by-step instructions. "Refactor X and add tests" invites parallelism. "First do A, then B, then C" does not.
Memory in Claude Code
Claude Code has its own memory, separate from the web app. Tell it to remember a preference or convention and it persists across sessions. Paired with a good CLAUDE.md, your project gets progressively easier to work in instead of starting from scratch.
The bottom line
The tools are already included. The bottleneck is not access — it's that nobody sets them up. Spend one afternoon this week wiring up Projects, a Skill, Memory, and a CLAUDE.md in your main repo. The compounding effect starts the next morning.
For the next step up, see the managed agents playbook, or learn how self-learning agents extend this stack in the Hermes guide.