How I AI: Automate Recruiting and Build Interactive Personas with Michal Peled of HoneyBook

In this episode, I was so excited to sit down with Michal Peled, a Technical Operations Engineer at HoneyBook. Michal is incredible at building internal tools and automations that get rid of friction and just make life easier for her team. She calls her creations "little helpers," and I absolutely love that framing—because don't we all want an AI little helper?

Michal's work is a great example of how you don't need to be on a customer-facing product team to have a huge impact with AI. She's part of a movement that's turning internal tools teams into real centers for innovation inside their companies. She’s solving real, everyday problems that save her colleagues time, money, and a lot of headaches.

Today, Michal walks us through three really smart and practical workflows. First, we’ll look at the advanced features of ChatGPT, using its agent mode to build an effective recruiting assistant that scours LinkedIn for great candidates. Then, she shows us how to bring static customer research to life by creating five distinct, interactive AI personas with Notebook LM. And finally, she shares a clever and personal automation that solves a uniquely San Francisco problem: avoiding surge pricing for parking during Giants games. Let's get to it!

Workflow 1: Automating LinkedIn Recruiting with ChatGPT's "Little Helper" Agent

Every hiring manager and recruiter knows the grind of sourcing candidates. It involves hours of manually searching LinkedIn, applying filters, and vetting profiles against a job description. It’s one of those tedious, repetitive tasks that’s necessary but doesn't feel like a great use of time. Michal saw this friction on her hiring team at HoneyBook and decided to build an AI agent to take on the load.

The goal was to take that load off of them. And ChatGPT Agent mode came just in time.

Step 1: Crafting the 'Little Helper' Prompt

It all starts with a well-structured prompt. Michal's approach is to give the AI a clear role, a detailed task, and a set of specific constraints—basically, she maps out the exact process a human recruiter would follow. She started by interviewing her colleagues to understand their step-by-step workflow and then translated that directly into the prompt.

Here’s the prompt she used:

You are an IT recruiter. 
Log into LinkedIn using my account. If not already logged in, let me take control and log in. Find up to five LinkedIn profiles where the current title and job description match the attached job description. 
Restrictions:
- Candidates must be from Israel or currently working at an Israeli company.
- They must be active in LinkedIn within the last three months.
- The current job role must be close enough to the open role in title and seniority.
- The candidates must either work in their current workspace more than a year, or they can be unemployed, but no more than a year, and have worked in their last workplace for over a year.

What I love about this prompt is how it blends autonomy with collaboration. It instructs the agent on what to do but also includes the crucial line, If not already logged in, let me take control and log in. This shows that agents don't have to be a fire-and-forget tool; they can be a co-pilot that you collaborate with.

Step 2: Unleashing the Agent

Once the prompt is in, things get really interesting. ChatGPT's agent mode opens what Michal perfectly describes as a "magic computer" right inside your chat window. You see a browser window where the agent navigates the web, clicks buttons, types into search bars, and scrolls through pages. The user experience is incredible.

One of the coolest features is the "thoughts" panel, which narrates the agent's reasoning in real-time: "Now I will go to the feed page," "I plan to click on...", "First, I need to make sure..." It's like looking inside the brain of your little helper as it works. While the agent ran, it efficiently executed the search, completely avoiding the usual distractions of the LinkedIn feed that would trap a human for minutes.

Step 3: The Surprising Results

After about 10 minutes, the agent returned a neatly formatted table with five candidates, complete with links to their profiles and a calculated "match score." Michal was rightfully skeptical, so she sent the list to her hiring manager for validation.

The feedback was impressive. Of the five candidates the agent found:

• Four were completely new, high-quality prospects that the team had never found manually but were a great fit.
• The fifth candidate was someone they had already identified and was in the interview process, validating the agent's accuracy.

This workflow is a great counterargument to the idea that AI trades quality for speed. In this case, it delivered both. The team not only saved hours of manual work but also ended up with a higher-quality candidate pool. For anyone involved in hiring, this is a huge deal.

Workflow 2: From Static Research to Interactive AI Personas

Most companies invest a lot in customer research, which results in detailed buyer personas. The problem is, this valuable insight often gets trapped in dense PDFs and slide decks that people rarely look at in their day-to-day work. Michal's team at HoneyBook faced this exact issue with five detailed personas they had developed. Her goal was to make them living, breathing entities that the product and marketing teams could actually talk to.

Step 1: Synthesizing Insights with NotebookLM

To start, Michal needed to distill hundreds of pages of research into a core identity for each persona. Instead of just dumping the files into a custom GPT, she turned to a more specialized tool: Notebook LM, which is powered by Google Gemini.

She chose Notebook LM for two key reasons:

1. Source-Grounded Answers: It can be instructed to only use the documents you provide as its knowledge base, preventing it from inventing information.
2. Citations: Its responses include citations that link back to the specific part of the source document, making it easy to verify accuracy.

Step 2: Prompting the Prompt Engineer

This is where the workflow gets wonderfully meta. Michal prompted Notebook LM to act as an expert prompt engineer. Its mission was to read all the research documents and, from them, generate a perfect, detailed prompt for a custom ChatGPT that would become each persona.

Here’s the core of her prompt to NotebookLM:

You are an expert prompt engineer specializing in creating custom GPTs by providing strong AI prompts. Your mission is to create AI prompts for custom GPTs representing entrepreneurs and small business owners... you'll craft highly detailed nuance and authentic ChatGPT prompts for five distinct buyer personas based on your sources.
Guidelines:
- Ensure that the prompt correctly and fully describe the core identity, mindset, decision making style, tone and communication style...
- ...business needs and the technology stack and the journey maps, social media preferences...
- Don't add or modify text that is not written or implied in the text. I know you're creative. I am turning you down. The text describe a specific persona must remain true to the original persona.

I have to laugh at that last instruction. The "don't make up stuff" directive is one everyone should have in their toolkit!

Step 3: Refining and Adding Guardrails

The prompts generated by NotebookLM were a fantastic starting point, but they needed refinement. Michal used ChatGPT and Claude to tighten the language, get them under the 8,000-character limit for custom GPT instructions, and add crucial guardrails to prevent misuse.

She added instructions like this to ensure the personas stayed on-brand and professional:

You do not act as a general purpose assistant. You do not ask follow up questions, you avoid slang, bad language or distasteful content and keep communication respectful and inspiring. You avoid political, religious, gender, or racial commentary.

This is such a smart move, and it should probably be a default for all enterprise GPTs. It anticipates how users might try to break the persona and proactively keeps the conversation focused and safe.

Step 4: Talking to Your Customers, 24/7

With the final prompts, Michal created five custom GPTs, one for each persona. Now, instead of digging through a PDF, her colleagues can directly ask a persona like "Balanced Blake" a question.

For example, when asked, "What kind of ad headline would catch your attention during a busy workday?", Blake responds:

A few would catch my eye. 'Save 10 hours a week with this tool. No tech skills needed' or 'From chaos to clarity. One dashboard to run it all.'

When the same question is posed to "Aiden," the answer is completely different, reflecting his unique mindset. This workflow successfully transformed static data into a dynamic, interactive tool that helps teams brainstorm, test messaging, and build empathy for their customers.

Workflow 3: Solving a Hyper-Local Problem with a Custom Calendar

This last workflow is a personal favorite because it solves such a specific, tangible, and frustrating problem. The HoneyBook office is located right next to Oracle Park, home of the San Francisco Giants. On game days, parking rates skyrocket from a flat ~$50/day to a whopping $40+ per hour.

As someone whose former office was also behind the ballpark, I have personally felt this pain. I once texted a friend in despair after having to pay nearly $100 to park for a meeting. Michal's team was constantly getting caught off guard, so she built a simple AI solution.

The One-Shot Prompt

Michal didn't need a complex agent or a multi-step process. She just needed a simple way to know about weekday, daytime games in advance. She went to ChatGPT with this beautifully simple prompt:

Find all home games that take place in Oracle Park in San Francisco during the next six months. Filter out only the games that start anywhere between morning to 2:00 PM. Using these dates create an ICS file for Google Calendar that will show these dates as an all day event. Availability: free. The event description should contain the game details and time. Also provide a textual list of all the dates, times, and events included.

The key instructions here are really clever: create an ICS file (the standard format for calendar events), make it an all day event for high visibility, and critically, set Availability: free so it doesn't block off everyone's work calendar.

The Instant Solution

In just 36 seconds, ChatGPT delivered two things: a downloadable .ics file and a text list of the games for verification. Michal imported the file into her Google Calendar and shared it with the team. Instantly, everyone had a heads-up on which days to avoid driving to the office.

This is a perfect example of a "little helper" automation. It's not a massive, company-wide system, but it solves a recurring, high-friction problem for a group of people, making their daily lives just a little bit better.

Final Thoughts

Michal's workflows are a perfect demonstration of practical AI. From a sophisticated recruiting agent to interactive marketing personas to a simple parking calendar, she shows how to identify points of friction and apply the right tool for the job. Her work is a testament to the growing importance of internal tools teams, who are now equipped to build high-impact solutions that can often move faster than customer-facing product teams.

My call to action for you is this: what's the recurring friction in your workday or your team's workflow? What's the equivalent of the $40/hour parking problem in your life? Take a page from Michal's book, start with a simple prompt, and go build your own "little helper." You might be surprised by what you can create.

Thanks to Our Sponsors!

Thank you to our amazing sponsors for supporting the show:

• Brex—The intelligent finance platform built for founders
• Google Gemini—Your everyday AI assistant

Episode Links

• Watch on YouTube
• Listen on Spotify
• Listen on Apple Podcasts

Find Michal Peled on LinkedIn and find me, Claire Vo, on LinkedIn, X, or at ChatPRD.