Skip to content

Getting the Most from Claude

Claude.ai is the primary way most Seer practitioners work — nothing to install, and your team's skills are already there. This page collects the settings, habits, and courses that make that experience faster and better.

Start Here

First Settings to Enable

Several useful capabilities are off by default. Turn these on under Settings → Capabilities:

  • Memory — Claude learns your preferences and context over time.
  • Search and reference past chats — pull context from prior conversations.
  • Artifacts — a canvas/doc editor for structured outputs like reports, apps, and pages.
  • Code Execution and File Creation — data analysis, scripts, and file generation.
  • Connectors — connect Slack, Google Drive, Notion, and Seer MCP.

Projects: Your Primary Workspace

Each Project is a workspace scoped to a client, workstream, or recurring task type.

Set up a Project when:

  • You're working the same client across multiple sessions
  • You have reference files you'll reuse
  • You want consistent behavior without re-explaining context every time

Put in a Project:

  • Key reference files (audits, briefs, templates, brand/style guides)
  • A clear system prompt (role, tone, output format)
  • One chat per task, rather than one giant thread

Avoid:

  • Leaving the system prompt blank
  • Re-uploading the same files every session
  • Cramming everything into one mega-thread

Choosing the Right Model

  • Opus — complex strategy, long documents, nuanced analysis, quality-critical work.
  • Sonnet — most day-to-day work (briefs, reports, deliverables, research). This is the default.
  • Haiku — quick rewrites, simple summaries, fast turnarounds.

You can switch models mid-conversation from the model selector. Default down to a smaller model when the task is straightforward — you'll get faster responses without losing quality.

If you moved from Sonnet 4.6 to Sonnet 5 and notice more meandering or over-thought answers, first try lowering the effort/thinking level, and state the goal directly ("Your goal is X — do it as concisely as possible") instead of a numbered list of steps. You can still manually select 4.6 while you adjust your prompts.

Efficiency Habits

The highest-leverage habits, worth building early:

  1. Front-load your prompt. Don't refine through 10 follow-ups — every reply re-reads the whole thread, so a complete first prompt is cheaper than a chain of corrections.
  2. Edit your original message instead of sending a correction. Claude restarts with clean context instead of carrying the mistake forward.
  3. Use Projects for persistent context, but keep Project files lean — a tight two-paragraph system prompt beats five uploaded docs.
  4. Ask "Audit what is using up my context" when a chat feels slow.
  5. Ask for outlines before full drafts on longer deliverables.
  6. Be surgical — paste only the relevant section, not the whole document.
  7. Specify output length/format ("under 200 words", "bullets only") to skip a revision round.
  8. Start a new chat for a new topic rather than continuing an unrelated thread.
  9. Track usage at a glance. A free browser extension (search your browser's extension store for "Claude Usage Tracker") shows remaining usage without leaving the tab.

Anthropic Academy Courses

Free courses at anthropic.skilljar.com:

  • Claude 101 — everyone, start here
  • AI Fluency: Framework & Foundations — responsible AI habits
  • Introduction to Claude Cowork — letting Claude use your computer
  • Claude Code in Action — developers and technical teammates
  • Introduction to Model Context Protocol — MCP servers and integrations
  • Building with the Claude API — developers

Going Deeper

Want the builder/power-user surfaces — Claude Code, Cowork, MCP server details? See Power Users: Claude Code & Cowork.

For community best practices (CLAUDE.md conventions, Plan Mode, subagents, agent-first design), see claudelog.com.

See Also