This beautifull, incredible, dexterous,smart, clever and competent woman named Michal Peled walks through three real examples from her work at HoneyBook in a podcast episodes across Lenny’s Podcast Network.

Michal is using ChatGPT’s agent mode to automate LinkedIn recruiting (and surface candidates humans missed), turning static customer research into interactive AI personas, and even building a parking-avoidance calendar app for Giants games. Along the way, Michal breaks down how to structure great agent workflows, why interviewing humans is the key to good automation, and how tools like NotebookLM can turn messy research into reliable, citation-backed prompts.

According to Michal, ChatGPT agent mode is your “little helper” that can perform real-world tasks. Unlike regular ChatGPT, agent mode can navigate websites, perform searches, and execute complex workflows—all while narrating its thought process. This transforms it from a text generator into a true digital assistant that can replicate human workflows.

Michal says that :

1) The best AI automation starts with interviewing humans about their process. Michal’s recruiting automation succeeded because she asked the hiring team exactly what criteria they use : “Candidates must be from Israel or working at an Israeli company, active on LinkedIn within 3 months, and employed at their current job for over a year.” By codifying these specific requirements, she created an agent that found candidates the team had missed.

2) The best prompt improvement technique : ask AI to fix your prompt. When AI isn’t giving you the results you want, Michal’s approach is to say : “This is the prompt I’m using. This is what’s wrong with the output. This is how I want it to be. Take away everything that doesn’t work well.”