Everything you need to understand, try, and build each one — with real tools, steps, and prompts.
Stop drowning in your inbox. An agent reads, prioritises, drafts replies, and books meetings — you just approve before anything sends.
Authenticate your email. Set trigger: "New email received."
Send subject + body. Ask: "Is this Urgent / Reply needed / FYI / Junk?"
Urgent → priority list. Reply needed → draft. Junk → archive.
Pass email + your tone context to Claude. Save as Gmail draft.
Use Calendly or Google Calendar API to check availability and suggest meeting slots.
"Read this email and classify it as [URGENT], [REPLY_NEEDED], [FYI], or [JUNK]. State why in one sentence. Email: {{email_body}}"
Give it a question. Get back a structured, sourced report — in minutes, not hours. Perfect for market research, competitor analysis, or briefing documents.
Sections: Executive Summary, Overview, Key Players, Trends, Recommendations.
Use Tavily API or Serper API to give the agent real-time search.
Ask Claude to decompose your topic into 5 sub-questions before searching.
One search per sub-question. Pass all results to Claude to synthesise.
Combine sections into a final document. Export as PDF or Word.
"I want to research: {{topic}}. Break this into 5 specific sub-questions that together give a complete picture. Return as a numbered list only."
Handle 80% of repetitive questions automatically — so you only deal with the 20% that genuinely needs you. Runs 24/7, never loses patience.
Write your 20 most common Q&As + policies + product info.
Use Botpress or Supabase+pgvector so the agent can search your docs.
Define brand voice, what the agent can/can't do, and escalation rules.
Embed on your site using Tidio, Crisp, or a custom widget.
Try angry customers, unusual questions, discount attempts. Refine.
Route unresolved chats to Slack. Log all conversations to improve over time.
"You are a friendly support agent for {{company}}. Acknowledge emotion first, then solve. If unresolved after 2 messages, say: 'Let me connect you with our team.'"
Brief it once. Get a full content calendar with posts, captions, and email copy — all tailored to your brand voice. Consistent content without the burnout.
Tone, banned phrases, 3 example posts, your audience persona. One page.
Fields: Topic, Goal, Key message, Channel, CTA. Fill once per campaign.
One prompt per channel with format rules (length, hashtags, tone).
New row in content calendar → generate content → save drafts to Notion.
Automate creation, not publishing — until you fully trust the output.
"Take this blog post and create: (1) a LinkedIn post, (2) 3 Instagram captions, (3) an email newsletter section. Match my brand voice. Post: {{content}}"
Describe what you want built. A coding agent writes it, runs it, finds the bugs, fixes them, and keeps going until it works. No code knowledge required to start.
Install Claude Code via npm or download Cursor IDE. Both are agent-ready.
Language, inputs, outputs, edge cases, libraries. More detail = better first draft.
Use a virtual environment or Docker so the agent can execute code safely.
Check git diff before merging. Agents can make unexpected changes elsewhere.
Give the agent a test suite. Instruct it to keep running until all tests pass.
"Build a {{language}} function that takes {{input}}, returns {{output}}, handles {{edge_cases}}, includes error handling and comments. Write tests alongside it."
Connect your tools. Define the triggers. Let agents move data and fire actions automatically — the highest-leverage use of AI for any professional.
Pick something you do 3× a week that always follows the same pattern.
"When X happens, do A, then B, if C do D, else do E, then notify me."
Start with a template. Customise rather than build from scratch.
Use Claude to classify, score, or enrich data mid-workflow.
Check every step before enabling automatic triggers.
Set up error alerts. Review logs weekly. Refine as you go.
"Classify this enquiry as [HIGH_PRIORITY], [STANDARD], or [SPAM]. High priority = budget over £10k or urgent deadline. Return label + one-line reason. Enquiry: {{text}}"