Skip to main content
WonkaChat is secure by design: your data is encrypted, isolated between organizations, and never used to train public models. These habits help you use it well day to day.

​
Scope your requests

  • Analyze this dataset and identify trends
  • Draft a reply to this customer inquiry
  • Review this document for clarity and errors
  • Summarize the key points from this meeting
Specific tasks get better results and are easier to check.

​
Think before executing

The main everyday risk isn’t a security flaw — it’s an agent doing something you didn’t mean to authorize.
  • Sending emails or messages to customers
  • Deleting files, records or accounts
  • Deploying or changing production configuration
  • Overwriting a shared document
  • Drafting content for you to review
  • Analyzing data and summarizing findings
  • Searching and retrieving information
  • Generating documentation or test data

​
Reduce risk for sensitive agents

In the agent builder, only enable the connector actions the task requires — for example, keep “draft email” but leave out “send email”. See tool restriction.
Try a new agent on a small, low-stakes case before relying on it for something important. Use anonymized or test data when you can.
For multi-step work, review the result of each step before moving to the next, rather than asking for everything at once.
Before connecting an agent to a tool, check what that connector can see and do. Remember it acts with your own permissions in that tool.
Combine these habits for higher-stakes work: restrict the agent’s tools and review its output before anything goes out.

​
Common-sense guidelines

Treat AI output like a colleague’s draft: check it before it reaches customers or production systems.
For code, commit AI-assisted changes separately so you can track and revert them if needed.
This is about professional ethics and your organization’s reputation, not a platform limitation.
For important decisions, check AI-generated information against authoritative sources. Use AI to speed up research and drafting, not to replace judgment.

​
Quick reference

For higher-risk tasks: restrict the agent’s tools, share narrowly first, and review output before anything ships.

​
The bottom line

  1. Think before executing — review before sending, deleting or deploying.
  2. Scope requests clearly — specific tasks are easier to check.
  3. Use good judgment — verify what matters, and treat AI as a fast collaborator, not a replacement for it.