Skip to main content
Twin’s chat editor is built to guide you as you go — it will ask follow-up questions whenever something is unclear, and help you fill in any missing details so the agent can run correctly. That said, the tips below can help you get to a great result faster (and reduce back-and-forth), especially for more complex or high-stakes workflows.

​
Start With a Clear Objective

Twin works best when it understands exactly what problem you’re trying to solve.

What usually doesn't work

Vague requests make it hard for an agent to infer the right behavior:
  • “I need help with social media”
  • “Build something for marketing”
  • “I want to automate my business”
These describe a domain, not a task.

What works much better

Clear, concrete goals give the agent something actionable:
  • “Visit website X, extract the main article content, summarize it, and email me the result.”
  • “Collect Google Reviews for a list of restaurants and save them to a CSV file.”
  • “Transcribe a YouTube video, turn it into structured notes, and store them in a Google Doc.”
The key difference: these requests describe a specific outcome, not a vague intention.

​
Focus on Outcomes, Not Implementation

You don’t need to tell Twin how to do something. Tell it what you want done.

Avoid implementation-driven requests

These describe tools, not goals:
  • “I need a web scraper”
  • “Build an AI workflow”
  • “Use the OpenAI API”

Describe the result you want instead

  • “Track competitor prices weekly and report changes.”
  • “Analyze customer reviews and identify recurring themes.”
  • “Automatically reply to simple customer emails.”
When you focus on outcomes, Twin can choose the most reliable approach internally.

​
Provide Context, Not Just a Task

Context helps the agent make better decisions.

Too vague

“Build me a lead generation workflow.”

Much better

“Build a lead generation agent that:
  • Finds marketing agency founders on LinkedIn
  • Based in the US
  • Companies with 10–50 employees
  • Collects contact details and company info
  • Saves results to a Google Sheet
  • Processes ~50 leads per week”
The added detail removes guesswork and improves accuracy.

​
Explain Your Current Manual Process

One of the most effective ways to design a good agent is to explain what you do today.

Example

“Every Monday, I manually:
  1. Check Twitter analytics for mentions
  2. Review LinkedIn post performance
  3. Compile everything into a PowerPoint
This takes about 2 hours.I want an agent that does this automatically and emails me the report.”
This gives Twin a clear blueprint to replicate and automate.

​
Be Explicit About Your Data

If your workflow involves structured data, describe it clearly.

Example

“I have a Google Sheet with:
  • Column A: Customer name
  • Column B: LinkedIn profile URL
  • Column C: Company name
  • Around 200 rows, updated weekly
I want to enrich this with job title and company size in columns D and E.”
Concrete examples reduce ambiguity and speed up setup.

​
Specify Where Results Should Go

Tell Twin exactly what should happen once the agent finishes.

Example

“When new leads are found:
  • Add them to HubSpot under ‘Marketing Qualified’
  • Post a message in #sales on Slack with the count
  • Email me a summary if more than 10 high-priority leads are detected”
Clear destinations and conditions lead to predictable behavior.

​
Use This Request Template

If you’re unsure how to phrase your request, use this structure:
ElementDescription
GoalWhat you want to achieve
Current processWhat you do manually today
InputsData you start from (with examples)
OutputsWhat the agent should produce
DestinationsWhere results should go
Success criteriaHow you’ll know it’s working
Goal: Monitor competitor pricing and get alertsCurrent process: Every Friday, manually check 5 competitor sites and log pricesInputs: Google Sheet with product names, URLs, current pricesOutputs: Updated prices, alerts for 10%+ changes, weekly summaryDestinations: Google Sheet, email, SlackSuccess criteria: Changes detected within 24 hours, high accuracy, no false alerts

​
Start Small, Then Expand

Begin with a narrow, well-defined task. Once it works reliably, extend the agent with additional steps or logic.

Providing business context helps

“I run a digital marketing agency managing social media for 25 restaurants. My main challenge is producing enough content while understanding what performs best.”
This allows the agent to better align with your real-world constraints.

Start Building

Apply these tips in the Twin App