Via Annie Tsai

Mistake number one is […] wanting to automate the exciting things instead of the annoying things first. Start your agentic flows on automating the things that are repetitive, are data heavy, but are also low ambiguity

— The Three Most Common Mistakes People Make with Building AI Agents—Annie Tsai at 00:53

---

Most teams start their AI agent journey by reaching for the most ambitious use case — strategic insights, executive briefings, proactive customer engagement. Annie Tsai, experienced AI agent builder and agentic system design consultant, says this is mistake number one. “You’re wanting to automate the exciting things instead of the annoying things.”

When you’re early in the AI Adoption journey, use the RDL framework to find the ideal candidate use cases:

• Repetitive Tasks that you have to keep doing and which are time-consuming for your team
• Data-heavy Tasks that require a lot of data gathering, processing, or routing, but only if they have predictable inputs and outputs and use data from accessible sources
• Low variability and ambiguity — Tasks where the range of answers is narrow and predictable, the agent doesn’t need to make judgment calls, and it’s clear what “done” and “success” looks like

If a proposed agent project doesn’t pass all three criteria, it’s likely premature for an early AI project.

The sophisticated use cases — like waking up to a “magical, beautiful dossier of everything I need to know” — require significant data pre-processing and mature knowledge systems. Those come later, after you’ve built the foundational infrastructure through simpler agent projects.