Automate Recruiting and Build Interactive Personas with Michal Peled of HoneyBook

By: Michal Peled Host: Claire Vo Source: How I AI (ChatPRD) Type: podcast

Summary

How I AI episode with Michal Peled, Technical Operations Engineer at HoneyBook. Three “little helper” workflows demonstrating practical AI for internal tools: (1) LinkedIn recruiting agent with ChatGPT agent mode — crafted a structured prompt by interviewing recruiters about their step-by-step workflow, gives the AI a recruiter role with specific constraints (candidates from Israel or Israeli companies, active within 3 months, matching title and seniority, 1+ year tenure or recently unemployed); agent mode opens a “magic computer” browser window to navigate LinkedIn, search profiles, apply filters, and vet candidates; “thoughts” panel shows real-time reasoning; returns formatted table with 5 candidates, profile links, and match scores in ~10 minutes; validation showed 4/5 were new high-quality prospects never found manually, 1/5 was already in the interview pipeline (confirming accuracy); includes collaborative handoff for login; (2) Interactive AI personas from static customer research — uploaded hundreds of pages of buyer persona research documents into NotebookLM (chose for source-grounded answers with citations); prompted NotebookLM as “expert prompt engineer” to read all research and generate detailed custom GPT prompts for 5 distinct personas; included strict instruction not to add or modify information beyond the source; refined prompts with ChatGPT and Claude to fit 8,000-character limit and add guardrails (no off-topic, no slang, no political/religious content); created 5 custom GPTs that give distinct in-character responses to product and marketing questions; (3) Game-day parking calendar automation — single ChatGPT prompt finds all Oracle Park home games in next 6 months, filters to morning-2PM starts, generates downloadable ICS file as all-day events with “free” availability and game details in description; imported to Google Calendar and shared with team; solved recurring $40+/hour surge parking problem in 36 seconds.

Key Ideas Extracted

Agent mode as collaborative co-pilot: Instruct the agent what to do autonomously but include handoff points (“let me take control and log in”) — agents don’t have to be fire-and-forget
Interview the experts to write the prompt: Map out the exact human workflow step-by-step by talking to the people who do the job, then translate directly into an AI prompt — produces much better results than generic instructions
“Thoughts” panel for agent transparency: ChatGPT agent mode shows real-time reasoning (“Now I will go to the feed page,” “I plan to click on…”) — builds trust and helps debug agent behavior
NotebookLM for source-grounded prompt generation: Use NotebookLM’s citation-backed, source-only mode to generate custom GPT prompts from research docs — prevents the AI from inventing persona details not in the research
Meta-prompting: AI as prompt engineer: Prompt one AI tool to generate prompts for another AI tool — NotebookLM reads research and creates the custom GPT system prompts, a more reliable pipeline than writing them manually
Guardrails for enterprise custom GPTs: Add explicit constraints (no off-topic responses, respectful tone, no political/religious content) — should be a default for all enterprise-facing AI personas
Right-sized AI for right-sized problems: Complex recruiting → agent mode; static research → NotebookLM + custom GPTs; parking schedule → single prompt with ICS file output — match the tool complexity to the problem complexity
Internal tools teams as AI innovation centers: Technical ops/internal tools teams can often move faster than customer-facing product teams because they face less red tape and can directly solve their colleagues’ daily friction

Notes

Published Dec 8, 2025 on How I AI (ChatPRD). ~9 min read. (Note: filename says 2025-10-20 but actual publication date is Dec 8, 2025)
Sponsors: Brex, Google Gemini
Michal Peled background: Technical Operations Engineer at HoneyBook; builds internal tools and automations
Tools: ChatGPT (agent mode for recruiting, ICS generation), NotebookLM (source-grounded persona generation), Claude (prompt refinement), Custom GPTs (5 interactive personas), Google Calendar
Key stats: 5 candidates found in ~10 minutes; 4/5 new high-quality; ICS file generated in 36 seconds
HoneyBook office: near Oracle Park (SF Giants), hence the parking problem
Three companion workflow guides published Jan 8, 2026
Cross-references: Agent mode workflows, interactive personas, NotebookLM, internal tools innovation, recruiting automation

Raw Content

Re-scraped from ChatPRD 2026-02-16. Full article content captured in Summary and Key Ideas above.