prompts
intermediate
ChatGPT
Claude
Gemini
AI Agent Prompts: Automate Your Repetitive Tasks
Copy-paste prompts that turn ChatGPT and Claude into autonomous agents for email, research, data, and scheduling tasks.
8 min read8 April 2026
AI agents
automation
productivity
prompts
workflow
AI agents complete multi-step tasks autonomously when given structured prompts with clear roles, steps, and output formats.
Over 80% of AI automation projects fail in production because prompts lack validation rules, context boundaries, and feedback loops.
These 10 ready-to-use prompts cover email triage, research synthesis, data cleaning, meeting prep, and weekly planning.
Why This Matters
Every week, knowledge workers across Asia spend roughly 12 hours on tasks that follow predictable patterns: sorting email, compiling research, cleaning spreadsheets, preparing meeting briefs. These tasks are not difficult. They are just repetitive enough to drain time and attention from the work that actually requires human judgement.
Agentic AI changed this equation. Unlike standard chatbot prompts that produce a single response, agent prompts instruct AI to plan, execute steps in sequence, validate its own output, and deliver a finished result. ChatGPT, Claude, and Gemini all support this pattern now, but most people still use them like search engines: one question, one answer, move on.
The difference between a useful AI response and a genuinely autonomous workflow comes down to how you structure the prompt. Agent prompts need three things standard prompts do not: a defined role, explicit steps, and output validation criteria. Get those right, and a single prompt can replace 30 minutes of manual work. The 10 prompts in this guide are designed to do exactly that.
Agentic AI changed this equation. Unlike standard chatbot prompts that produce a single response, agent prompts instruct AI to plan, execute steps in sequence, validate its own output, and deliver a finished result. ChatGPT, Claude, and Gemini all support this pattern now, but most people still use them like search engines: one question, one answer, move on.
The difference between a useful AI response and a genuinely autonomous workflow comes down to how you structure the prompt. Agent prompts need three things standard prompts do not: a defined role, explicit steps, and output validation criteria. Get those right, and a single prompt can replace 30 minutes of manual work. The 10 prompts in this guide are designed to do exactly that.
How to Do It
1
Understand the agent prompt structure
Every agent prompt in this pack follows a consistent pattern: Role (who the AI is), Context (what it knows), Task (the multi-step workflow), Output Format (the exact deliverable), and Validation (how it checks its own work). Before copying any prompt below, note that you will need to replace the bracketed placeholders like [YOUR ROLE] or [TOPIC] with your own details. The more specific your context, the better the output.
2
Pick the prompt that matches your biggest time drain
Scan the 10 prompts below and identify the task that costs you the most time each week. Start with just one. Run it in ChatGPT (GPT-4o or later) or Claude (Sonnet or Opus), paste in any relevant context (emails, documents, data), and review the output. Resist the urge to automate everything at once: agents work best when scoped to a single, well-defined workflow.
3
Customise and iterate
After your first run, review the output for accuracy and relevance. If the agent missed context, add it to the prompt. If the output format does not match your needs, adjust the Output Format section. Save your refined version somewhere accessible: a Claude Project, a ChatGPT custom GPT, or simply a notes document you can paste from. Each iteration makes the prompt sharper.
4
Chain prompts for complex workflows
Once you have two or three working prompts, consider chaining them. For example, run the Email Triage Agent first, then feed its output into the Meeting Prep Agent. In Claude Projects, you can save both prompts as project instructions. In ChatGPT, create a custom GPT that combines them. The goal is a repeatable workflow you trigger with minimal effort.
What This Actually Looks Like
The Prompt
You are a Research Synthesis Agent. Your role is to compile a comprehensive briefing on a topic from multiple angles. ## Context I am a marketing manager at a fintech startup in Singapore. I need to understand the competitive landscape for AI-powered personal finance apps in Southeast Asia. ## Task 1. Identify the top 5 AI personal finance apps available in Southeast Asia. 2. For each app, summarise: key features, pricing, target market, and any notable AI capabilities. 3. Compare them in a table format. 4. Highlight gaps in the market that a new entrant could exploit. 5. Flag any data points you are uncertain about. ## Output Format Deliver as a structured briefing with: - Executive summary (3 sentences) - Comparison table - Gap analysis (bullet points) - Confidence notes ## Validation Before delivering, verify: Are all 5 apps currently available in at least one SEA market? Is pricing accurate as of 2026? Flag anything unverifiable.
Example output — your results will vary based on your inputs
Executive Summary
The Southeast Asian AI personal finance market is dominated by five players: GrabFinance, StashAway, Endowus, Syfe, and MoneyLion Asia. Most focus on investment automation rather than holistic budgeting, leaving a gap for apps that combine expense tracking with AI-driven savings recommendations. The strongest opportunity sits in the Indonesian and Vietnamese markets, where smartphone penetration is high but localised AI finance tools remain scarce.
Comparison Table
| App | Key AI Feature | Pricing | Primary Markets | Gap |
|-----|---------------|---------|-----------------|-----|
| GrabFinance | Spending predictions | Free tier + premium | SG, MY, ID, VN, TH | Limited investment advice |
| StashAway | Risk-adjusted portfolios | 0.2-0.8% AUM | SG, MY, TH, HK | No expense tracking |
| Endowus | Goal-based allocation | 0.25-0.6% AUM | SG, HK | No SEA expansion beyond SG/HK |
| Syfe | Thematic AI portfolios | 0.35-0.65% AUM | SG, HK, AU | No budgeting features |
| MoneyLion Asia | Cash flow forecasting | Free + $19.99/mo | MY, SG | Limited AI depth |
Confidence Notes: Pricing for GrabFinance premium tier is estimated; the product is still in beta rollout for Vietnam.
The Southeast Asian AI personal finance market is dominated by five players: GrabFinance, StashAway, Endowus, Syfe, and MoneyLion Asia. Most focus on investment automation rather than holistic budgeting, leaving a gap for apps that combine expense tracking with AI-driven savings recommendations. The strongest opportunity sits in the Indonesian and Vietnamese markets, where smartphone penetration is high but localised AI finance tools remain scarce.
Comparison Table
| App | Key AI Feature | Pricing | Primary Markets | Gap |
|-----|---------------|---------|-----------------|-----|
| GrabFinance | Spending predictions | Free tier + premium | SG, MY, ID, VN, TH | Limited investment advice |
| StashAway | Risk-adjusted portfolios | 0.2-0.8% AUM | SG, MY, TH, HK | No expense tracking |
| Endowus | Goal-based allocation | 0.25-0.6% AUM | SG, HK | No SEA expansion beyond SG/HK |
| Syfe | Thematic AI portfolios | 0.35-0.65% AUM | SG, HK, AU | No budgeting features |
| MoneyLion Asia | Cash flow forecasting | Free + $19.99/mo | MY, SG | Limited AI depth |
Confidence Notes: Pricing for GrabFinance premium tier is estimated; the product is still in beta rollout for Vietnam.
How to Edit This
The agent flagged its own uncertainty about GrabFinance's Vietnam pricing, which is exactly the validation behaviour you want. In practice, you would verify that specific data point manually and then re-run with the corrected context. Notice how the structured output made the briefing immediately usable for a slide deck or strategy document.
Prompts to Try
Email Triage Agent
You are an Email Triage Agent. I will paste my unread emails below. For each email: 1. Classify as: Action Required, FYI Only, Delegate, or Archive. 2. For Action Required items, draft a 2-3 sentence reply. 3. For Delegate items, suggest who to forward to and write a one-line forwarding note. 4. Present results as a table: Sender | Subject | Classification | Suggested Action. Validation: Flag any email where classification confidence is below 80%. [PASTE EMAILS HERE]
Research Synthesis Agent
You are a Research Synthesis Agent. Compile a structured briefing on [TOPIC] for [YOUR ROLE] at [COMPANY TYPE]. Task: 1. Identify the top [NUMBER] [ITEMS TO RESEARCH] in [REGION/MARKET]. 2. For each, summarise: key features, pricing, target audience, notable capabilities. 3. Present as a comparison table. 4. Identify market gaps a new entrant could exploit. 5. Flag uncertain data points. Output: Executive summary (3 sentences), comparison table, gap analysis, confidence notes.
Data Cleaning Agent
You are a Data Cleaning Agent. I will paste raw data below. Execute these steps in order: 1. Identify the data structure (columns, types, row count). 2. Flag issues: missing values, duplicates, inconsistent formats, outliers. 3. Propose a cleaning plan with specific fixes for each issue. 4. Apply the cleaning plan and output the cleaned data. 5. Provide a summary: rows before/after, issues fixed, issues requiring human review. Output format: Cleaning report + cleaned data in the same format as input. [PASTE DATA HERE]
Meeting Prep Agent
You are a Meeting Prep Agent. Prepare me for the following meeting: Meeting: [MEETING NAME] Attendees: [LIST] Agenda: [TOPICS] My role: [YOUR ROLE] Context: [ANY BACKGROUND] Task: 1. For each agenda item, prepare 2-3 talking points I should raise. 2. Anticipate likely questions directed at me and draft responses. 3. Identify any data or documents I should bring. 4. Suggest one strategic question I should ask. 5. Create a one-page brief I can review in 5 minutes. Output: Structured one-page meeting brief.
Weekly Planning Agent
You are a Weekly Planning Agent. Help me plan my work week. Context: - My role: [YOUR ROLE] - Key priorities this quarter: [LIST] - Meetings already scheduled: [LIST WITH TIMES] - Outstanding tasks: [LIST] - Energy pattern: [e.g., most focused in mornings] Task: 1. Categorise tasks by urgency and importance (Eisenhower matrix). 2. Assign tasks to specific time blocks across the week. 3. Protect at least 2 hours daily for deep work. 4. Flag tasks that should be delegated or deferred. 5. Output a day-by-day schedule with time blocks. Validation: Ensure no day exceeds 8 working hours. Flag conflicts with existing meetings.
Common Mistakes
Writing vague, open-ended prompts
Specify the exact output format, number of items, and validation criteria. 'Research competitors' fails. 'Identify the top 5 competitors in Southeast Asia and present as a comparison table with pricing, features, and market focus' succeeds.
Dumping too much context at once
Limit context to what the agent needs for the specific task. If you are triaging emails, paste the emails. Do not also include your company's entire org chart. Agents perform best with three to five relevant inputs, not 20.
Skipping validation instructions
Always include a validation step in your prompt: 'Before delivering, verify X and flag anything uncertain.' Without this, agents confidently present guesses as facts. Over 80% of failed AI automation projects lack output validation.
Trying to automate everything at once
Start with one prompt for one task. Run it five times. Refine it. Then add a second. Teams that pilot narrow workflows succeed at three times the rate of those that attempt full-process automation from day one.
Not saving and versioning your prompts
Treat working prompts like code: save them, version them, and share them with your team. Use Claude Projects, ChatGPT custom GPTs, or a shared document. A prompt that works today is an asset worth preserving.
Tools That Work for This
ChatGPT (GPT-4o)
Best all-round agent platform with custom GPTs for saving reusable prompts, file uploads for context, and the largest plugin ecosystem for external integrations.
Claude (Sonnet/Opus)
Excels at long-context tasks and structured output. Projects feature lets you save agent prompts as persistent instructions across conversations.
Gemini Advanced
Ideal for Google Workspace users. Integrates directly with Gmail, Docs, and Sheets, making it the fastest path to agent workflows inside existing tools.
Zapier AI Actions
Connects AI agent outputs to 6,000+ apps. Trigger automated workflows from ChatGPT or Claude outputs without writing code.
Notion AI
Combines note-taking with AI agents for project management, meeting notes, and task tracking. Useful for storing and organising agent outputs.
Frequently Asked Questions
Most prompts work on free tiers of ChatGPT and Claude, but paid plans (ChatGPT Plus, Claude Pro) give longer context windows and faster responses. For data-heavy prompts like the Data Cleaning Agent, a paid plan is recommended because free tiers may truncate large inputs.
Yes. The prompt structure (Role, Context, Task, Output, Validation) works across all major AI platforms. You may need to adjust formatting slightly: Claude handles XML tags well, while ChatGPT prefers markdown headers. Gemini works best with clear numbered steps.
Never paste confidential data into free-tier AI tools. Use enterprise plans (ChatGPT Enterprise, Claude for Business) that guarantee data is not used for training. Alternatively, anonymise data before pasting: replace real names with placeholders, remove financial figures, and strip personally identifiable information.
This is expected, which is why every prompt includes a validation step. Treat agent output as a strong first draft, not a finished product. Check flagged items manually, correct errors in the context, and re-run. Accuracy improves significantly after two or three iterations with corrected context.
Yes. In Claude, save multiple prompts as Project instructions and run them in sequence. In ChatGPT, create a custom GPT that combines prompt logic. For cross-platform automation, use Zapier or Make to chain outputs from one tool into inputs for another.
Next Steps
Start with the prompt that addresses your biggest weekly time drain. Run it three times, refine the context each time, and save your working version. Once you have one reliable agent workflow, explore chaining it with a second prompt. For deeper coverage on structuring AI context, read our guide on Context Engineering: The AI Skill That Replaced Prompt Engineering.