How I AI: Zapier EA Cortney Hickey's 4 AI Workflows for Meeting Prep, Culture Reinforcement, and Strategy

In this episode of How I AI, Zapier executive assistant Cortney Hickey demonstrates four systems she built for meeting preparation, meeting coaching, strategic document review, and company strategy access.

The workflows begin with tasks Cortney already understood deeply. She translates her manual process into steps, starts with a narrow version, watches the result, and adds context or actions only when they prove useful.

The practical effect is not just personal time savings. Several of the systems turn Cortney's operating knowledge into reusable support for the wider company, while leaving judgment and sensitive conversations with people.

The four examples move from personal preparation to organization-wide enablement: a weekly research agent, an internal meeting coach, a custom GPT for proposal feedback, and a NotebookLM strategy companion.

prepare for the coming week

Cortney used to reserve part of Friday for reviewing the week and preparing for the next one. A Zapier agent now completes much of the research and organization that went into that recurring block.

The agent follows the same path Cortney used manually across her calendar, the web, HubSpot, Gmail, and Slack, then puts the resulting context where she will use it.

How the meeting-prep agent works

Cortney began with a simple digest and expanded it over time. Her rule is progress over perfection: make one useful version, inspect it, and improve it as the need becomes clear.

1. Schedule: the agent runs each Friday at 8:00 a.m.
2. Calendar scan: it reviews the following week and excludes routine internal meetings that do not require preparation.
3. Research: for each external participant, it gathers several kinds of context.

It searches the web for the person's current role, experience, and other useful public information.

It checks HubSpot for relationship history, active deals, and recent sales notes.

It searches Cortney's Gmail and internal Slack for prior conversations or mentions of the participant's company.

1. Delivery: the agent creates two outputs.

Todoist receives a preparation task with the research brief, scheduled for two hours before the meeting.

Slack receives a weekly digest with the meetings, available context, missing information that may need manual follow-up, and suggested areas of focus.

I always tell people when they're starting with an agent like this is 'Progress over perfection.' I started this one with just a quick digest... And then over time I was like, oh, here's something else that might be helpful. Build something basic, see how it works, learn, and then make time to improve it.

Cortney can refine the logic in natural language through the built-in copilot. In the episode, she adds a request for LinkedIn profile links. She now rarely needs to touch the agent, but still has a clear place to investigate gaps and update its instructions.

coach meetings against company norms

Cortney first tested meeting coaching manually in Fathom, asking how an executive meeting reflected frameworks such as The Five Dysfunctions of a Team. Zapier later moved the idea into an automated workflow using Fellow transcripts.

The goal is consistent, expected feedback tied to Zapier's stated behaviors. The message is explicitly labeled as AI-generated, and each participant receives a small number of specific observations and actions.

How the meeting coach works

The workflow evaluates eligible internal meeting transcripts against a defined cultural and operating framework.

1. Input: after a meeting, it receives the transcript and metadata needed to identify participants and meeting length.
2. Eligibility: it excludes short meetings, external participants, and cases without enough context for specific feedback.
3. Context: the instructions include several layers of Zapier guidance.

Company values include expectations such as a growth mindset.

Meeting norms describe how Zapier expects discussions and decisions to work.

Team frameworks include The Five Dysfunctions of a Team and the balance between being demanding and supportive.

Impact behaviors describe what Zapier expects from employees.

The workflow then creates and delivers the coaching.

1. Feedback: the model identifies one or two growth opportunities and one or two actions for the participant's next meeting.
2. Delivery: it matches the participant's email address to Slack and sends the clearly labeled AI-generated feedback directly.

The first version was too soft. Cortney revised the instructions to be both more demanding and more supportive, producing feedback such as addressing misalignment directly or challenging the pace of a decision.

This is not a substitute for manager judgment or a complete assessment of someone's performance. It is a recurring prompt to compare observable meeting behavior with shared norms.

pressure-test a proposal before executive review

Colleagues often asked Cortney to review strategic tee-up documents because she understood how CEO Wade Foster evaluated them. She wanted to preserve that thought partnership without becoming the required stop for every draft.

Her Exec Prop custom GPT gives employees an independent review before a meeting. A clearer proposal can make executive time more productive, and in some cases can make a meeting unnecessary.

How the proposal reviewer works

The assistant's usefulness depends on the internal context behind it.

Knowledge base: Cortney supplied documents that explain Zapier's strategy and decision process.

Sources include the strategy memo and revenue roadmap.

Examples of strong tee-up documents show the expected structure and level of detail.

Team norms explain how decisions are framed and reviewed.

A "Managing Up to Wade" document records the CEO's communication and feedback preferences.

Input: an employee uploads a draft and asks for feedback on the tee-up.

Analysis: the GPT compares the draft with the supplied context and flags missing decisions, unclear reasoning, weak evidence, or unstated tradeoffs.

Output: the response follows a repeatable structure.

It begins with a short assessment of what is present and what is missing.

It recommends a small number of changes, such as making tradeoffs explicit.

When useful, it shows how a tighter version could be written.

It ends with a bold coaching question intended to test the author's thinking.

At the time of recording, Zapier's analytics showed 278 people had used the GPT to sharpen a strategy document. Cortney still provides direct coaching where it matters; the GPT gives more people a first review without waiting for her.

make company strategy easier to query

Strategy at a growing company lives across top-level documents, planning materials, all-hands meetings, and team action plans. Cortney and her colleagues collected those sources in NotebookLM so employees could explore them in one place.

The strategy companion launched about a month before the recording and can be updated as the underlying strategy changes.

How the strategy companion works

The notebook turns a few dozen existing sources into a searchable interface without replacing the original documents.

1. Sources: the workspace includes the company strategy, all-hands transcripts, strategy-meeting transcripts, and each organization's strategic action plan.
2. Questions: employees can ask how the company strategy connects to their own role. Cortney demonstrated this with: "As an executive assistant, how can I contribute to Zapier's 2026 strategy?"
3. Answers: NotebookLM synthesized themes such as clarity, focus, speed, and internal AI transformation, with citations back to the supplied sources. Citations make the basis inspectable, but employees still need to verify consequential interpretations.
4. Alternate formats: NotebookLM can also create an audio overview and quizzes from the same source set.

The maintenance requirement is as important as the interface. New strategy materials need to be added and outdated ones removed so the notebook does not present stale direction as current.

The shared pattern

Each system starts with a specific operating problem: repeated research, inconsistent coaching, a proposal-review bottleneck, or fragmented strategy. Cortney gives the model the context for that job and sends the output to a person who can use or challenge it.

Her progress-over-perfection approach keeps the first version small. Once the workflow proves useful, she adds sources, filters, output formats, and distribution so it can support more people without pretending the machine replaces judgment.