How Zapier’s EA built an army of AI interns to automate meeting prep, strengthen team culture, and scale internal alignment | Cortney Hickey

Cortney walks through her flagship automation: a Zapier agent that runs every Friday to prep for the upcoming week's meetings. The agent researches external participants via web search, checks HubSpot CRM for relationship history, searches Gmail and Slack for prior interactions, then delivers both Todoist tasks and a structured Slack digest with insights and recommendations.

• •Agent triggers Friday 8AM, identifies meetings requiring prep (excludes routine internal meetings)
• •Automatically researches external participants: current role, industry experience, company relationship in CRM
• •Searches Gmail and Slack for prior communications and mentions
• •Creates Todoist tasks 2 hours before each meeting with all context
• •Delivers weekly Slack digest with meeting intelligence, error handling, and AI-generated strategic insights
• •Serves as 'second brain' - consolidates info from CRM, email, Slack into one place instead of manual searching

Discussion of how to think about and improve AI agents over time. Cortney advocates for 'progress over perfection' - start basic, learn from execution logs, then iterate using natural language feedback to the copilot. She emphasizes teaching agents your system and way of thinking, just like training an intern.

• •Think of agents as interns: teach them your system, give them tools, they get smarter over time
• •Start with basic version, don't try to build everything at once
• •Review agent execution logs to see step-by-step reasoning and identify improvements
• •Use copilot to give natural language feedback (e.g., 'add LinkedIn hyperlinks to digest')
• •Can narrate your manual process to design agent structure: 'First I would Google, then check email, then Slack...'
• •Once working well, share templates across organization for others to customize

Courtney built an automated post-meeting coaching system that sends personalized feedback to all participants based on Zapier's values, meeting norms, and the Five Dysfunctions of a Team framework. Started as manual prompts in Fathom, now fully automated through Fellow integration, sending direct constructive feedback via Slack after every qualifying meeting.

• •Evolved from manually prompting Fathom after exec meetings to org-wide automation
• •Analyzes meeting transcripts against company values, meeting norms, decision-making frameworks, Five Dysfunctions of a Team
• •Sends each participant 1-2 specific growth opportunities and 1-2 actionable next steps via Slack
• •Iterated prompt with CEO to balance 'demanding and supportive' - initial version was too soft
• •Normalizes feedback culture: when everyone gets it automatically, reduces ego threat
• •Reinforces 'are we who we say we are?' - keeps cultural behaviors top of mind between quarterly offsites
• •2,978 people have used it to improve meeting effectiveness

To avoid becoming a bottleneck, Courtney created a ChatGPT custom GPT loaded with company strategy, values, meeting norms, and examples of good 'tee up' documents. Employees can stress-test their strategic docs before exec review, getting feedback on clarity, data inclusion, and alignment with how the CEO thinks - without waiting for synchronous feedback.

• •Built in ChatGPT (no conversation history) so employees feel safe sharing sensitive docs
• •Loaded with: team norms, revenue roadmap, strategy memos, good tee up examples, 'managing up to Wade' docs
• •Provides 'Wade-style' feedback: identifies if doc is too loose, suggests tightening, offers rewrites
• •Gives 'top fixes before the bullpen' and 'one bold coaching question'
• •Scales EA's thought partnership - people can get initial feedback 24/7 without being bottleneck
• •Sometimes makes tee ups so clear that meetings can be skipped entirely
• •Shows initiative when people come to exec meetings having already stress-tested with GPT

Using Google's NotebookLM, Courtney created an interactive strategy knowledge base containing dozens of sources: strategy docs, all-hands transcripts, org action plans. Employees can chat with it to understand how their work connects to company strategy, and it auto-generates AI podcasts summarizing strategic priorities.

• •Contains dozens of sources: top-level strategy, all-hands meetings, org strategic action plans
• •Employees can ask questions like 'As an EA, how can I contribute to 2026 strategy?'
• •Provides answers with source citations, connecting individual work to strategic priorities
• •Auto-generates AI podcast summarizing strategy for audio learners
• •Makes strategy interactive vs. static documents that get ignored
• •Continuously updated with latest information
• •Enables self-service strategic alignment without going to leaders for every question

Courtney addresses fears about AI replacing EAs, arguing it's a 'when not if' for adoption but that AI enables rather than replaces. She's received three promotions at Zapier by automating repetitive work to focus on creative, relationship-driven, high-impact projects. Discusses challenges of replicating constantly-learning executives versus consistent communication patterns.

• •AI adoption is 'when not if' for EAs - too much work, not enough time without it
• •Received three promotions by working herself out of manual tasks into higher-leverage roles
• •AI removes 'boring, repetitive, manual' work so EAs can do 'human stuff, fun stuff, creative stuff'
• •Can replicate consistent exec patterns (writing style, decision frameworks) quite well
• •Challenge: execs are constant learners ingesting new info, changing opinions - hard to keep models current
• •Old decisions become outdated as context evolves - 'that was an old decision'
• •Prompting technique: ramble via voice dictate, be very direct and demanding, give detailed feedback on reasoning