How I AI — Episode Summary
Episode Overview
Title: ChatGPT Agent Mode: The “Little Helper” That Transformed Recruiting, Crafted User Personas, and Solved Parking Nightmares
Host: Claire Vo
Guest: Michal Peled, Technical Operations Engineer at HoneyBook
Release Date: December 8, 2025
In this hands-on episode of How I AI, Claire Vo and guest Michal Peled explore how Michal uses ChatGPT’s agent mode to automate tedious business tasks. From revolutionizing HoneyBook’s recruiting process and turning user personas into chatbots, to solving the perennial San Francisco parking woes, Michal demonstrates practical, replicable AI workflows for everyday work and life. Expect transparent advice on prompt engineering, thoughtful discussion on user experience, and inspiration for making AI your own “little helper.”
Key Discussion Points & Insights
1. ChatGPT Agent Mode in Recruiting
[02:18–23:20]
-
Problem:
Recruiting teams at HoneyBook were losing countless hours manually searching LinkedIn for candidate matches against specific job descriptions. -
Solution:
Michal used ChatGPT’s agent mode as an interactive “little helper” to:- Log into LinkedIn
- Search for profiles aligned with a given job description
- Apply team-specific filtering criteria (e.g., candidates in Israel, employed at Israeli companies, actively using LinkedIn, tenure requirements)
- Present results in a scored table
-
Prompt Engineering Approach:
- Define the agent’s role: "You’re an IT recruiter."
- Describe the task in detail, including login steps and candidate criteria.
- List restrictions/instructions matching the exact recruiter workflow—gather these via interviews with the hiring team.
- Specify how many results are needed and request a “match score” for prioritization.
- Allow for human handoff: Build instructions for when the agent should pause or ask for user input.
-
Results & Team Reaction
- The agent generated five candidate matches—in just 10 minutes.
- Four of five candidates had not yet been found by the team; one was already in-process.
- Michal: “Now it’s going to be a real part of their hiring process, freeing their time to do other things that they love and appreciate a lot more.” (21:57)
- Improved not only speed, but quality—surfacing edge-case candidates.
Memorable Quotes:
- Claire: “If you can codify what a person’s step-by-step workflow is and put it in a simple prompt... you can replicate and automate that at scale.” (08:11)
- Michal: “The first thing they said was: out of these five, four of them were never found by us manually and they really fit the description.” (20:07)
Notable Tip:
“Think of agent mode as a little helper. That framing will help you come up with good prompts.” — Michal (10:26)
2. Creating Interactive Persona Chatbots from Research
[23:50–41:19]
-
Problem:
In-depth user research was trapped in documents, making it inaccessible for decision-making. -
Solution/Process:
- Turned five research-backed Personas into distinct ChatGPTs that embody each user type.
- Instead of just uploading files, Michal distilled behavioral, emotional, and practical traits into prompt instructions. The chatbot “became” the Persona, rather than just referring to summary notes.
- Used Google’s NotebookLM to extract and draft precise “identity” prompts, leveraging source verification and citation features.
-
Prompt Crafting:
- Defined attributes: core identity, mindset, decision-making style, business needs, technology stack, social media preferences, etc.
- Added strict guardrails: “Don’t make up text that is not in the research” to avoid hallucination.
- Capped prompts to fit ChatGPT custom GPT character limitations; further refined for clarity and safety (e.g., avoiding offensive language or inappropriate content).
-
Results:
- Now internal teams can “talk to” a Persona chatbot for ideas, ad feedback, UI testing, etc.—effectively simulating real target users on demand.
- Each Persona is used dozens of times for brainstorming and product planning.
Memorable Quotes:
- Michal: “I realized I’m not going to rely on uploaded files... The instructions here will be: you are that person, and this is your belief system.” (27:40)
- Claire: “So many people go to just plain ChatGPT, like ‘Give me five headlines for an ad campaign,’ as opposed to going and sitting with your fake Persona and saying ‘What ad campaigns would work on you?’” (40:09)
- Michal: “Personalizing those Personas has changed the way we work with them.” (40:09)
3. Solving the San Francisco Parking Headache with AI Calendar Automation
[41:19–48:15]
-
Problem:
HoneyBook’s office is near Oracle Park; on San Francisco Giants game days, parking prices spike unpredictably, catching employees off guard. -
Solution:
- Used ChatGPT to scrape the baseball schedule, filter for daytime home games (the only ones affecting morning parking), and generate a custom .ICS calendar file with “free” all-day events.
- Distributed the calendar team-wide, so everyone could plan commutes accordingly.
- Added a text-based list of all included games for transparency and verification.
-
Workflow Takeaway:
Demonstrated how even “dull” admin tasks can be streamlined with AI, saving time and money for teams.
Memorable Quotes:
- Michal: “ChatGPT doesn’t care [if I double-check its work]. So it thought for 36 seconds, provided me with a file and also with a list of all the games.” (45:50)
- Claire: “You probably could have found an SF Giants schedule calendar. That’s not exactly what you wanted... Now you have this filtered to what you want.” (47:12)
Notable Quotes & Moments
-
“Thinking about it as a little helper will really help you come up with good prompts for it.”
— Michal Peled (10:26) -
“As you watch this, what I hope our listeners and viewers are taking away is you don’t just have to rely on text prompts and chats when you’re using these tools... The next evolution of these LLMs…you actually have a lot more tools.”
— Claire Vo (12:22) -
“Four of them were never found by us manually... I want to approach and try to get them for an interview. And the fifth was one we caught manually who’s already coming in. To me, it was a great sign of quality.”
— Michal Peled (20:07) -
“This is your moment, internal tools teams. You can have really high impact and do great work... It’s moving faster than you can even get some of these experiences into product.”
— Claire Vo (51:17)
Pro Tips on AI Prompting (Lightning Round)
[52:35–55:52]
-
When facing unhelpful or hallucinated AI output:
- Copy your prompt into ChatGPT and explicitly tell it what’s wrong: “Output is inconsistent, contains too many hallucinations, invents things…”
- Outline exactly what you expect, then give ChatGPT explicit permission to add, remove, or rewrite as needed.
- Michal: “Giving permission to change, delete, remove whatever provides a better output... [Otherwise,] ChatGPT tends to be a pleaser.” (54:32)
-
All-caps for emphasis?
- Michal half-jokes about using all-caps in prompts for emphasis: “No one can persuade me otherwise.” (53:01)
- Ultimately, a structured, explanatory approach leads to much better results when revising prompts.
Timestamps for Key Segments
- Agent mode overview & recruiting use case: 02:18–23:20
- Persona GPT creation: 23:50–41:19
- Parking calendar automation: 41:19–48:15
- Discussion on Michal’s role & the value of internal tools teams: 49:16–52:35
- AI prompt improvement techniques: 52:35–55:52
Episode Takeaways
- AI Agents are most powerful when you take the time to precisely model the real process you want to supercharge.
- Interview teammates, codify steps and criteria, and treat LLMs as collaborators, not just passive responders.
- Custom AI personas unlock user empathy at scale in product and marketing.
- Even small personal pain points—like parking—can become opportunities for automation.
- Prompt engineering is recursive: use AI to improve your own prompts, and always provide permission and context for maximum effect.
Connect with Michal Peled:
- LinkedIn: linkedin.com/in/michal.peled
- Company: HoneyBook
For more episodes: