CHAPTERS
- 0:00 – 2:32
Why “AI memos” fail: leadership needs hands-on play space
Claire and Wade open by critiquing the common CEO pattern of delegating AI adoption via a memo and pushing the work down the org. Wade argues leaders must create structured experimentation time so teams build comfort and reduce fear through direct tool use.
- •The “delegation trap”: AI initiatives die when leaders offload adoption to the lowest level
- •Create hackathons/show-and-tells/structured sandbox time to build confidence
- •Hands-on exposure demystifies AI and reveals strengths/limitations pragmatically
- •AI adoption is also an employee development investment (future interview expectations)
- 2:32 – 6:50
Why Zapier leaned in early: credibility, learning loops, and values
Wade explains Zapier’s motivation for aggressive internal AI usage: delivering more customer value and aligning internal behavior with external evangelism. He frames mistakes as valuable learning that can be shared, supported by Zapier’s value of “Don’t be a robot, build a robot.”
- •AI is a transformative lever for shipping more customer value
- •Internal usage must match external messaging to maintain credibility
- •Making mistakes is part of the learning loop; share what worked and what didn’t
- •Company values and automation culture make Zapier predisposed to adopt faster
- 6:50 – 8:37
Making AI fluency measurable: rubrics that change what gets rewarded
Claire highlights Zapier’s use of AI fluency rubrics (especially for PMs) to clarify expectations across levels. The conversation focuses on how rubrics turn vague guidance into measurable behaviors that people can invest in.
- •Rubrics make AI adoption tangible and reduce “where do I start?” anxiety
- •Measurement and rewards drive behavior change more than broad mandates
- •Role-level expectations (e.g., PM levels) help standardize growth paths
- •Using AI to draft and refine rubrics accelerates iteration
- 8:37 – 10:49
Turning meeting data into a culture handbook with Granola Recipes
Wade demos a workflow using Granola’s “Recipes” prompt to generate an “unspoken company culture handbook” from meeting transcripts. He emphasizes how aggregated meeting data captures real operating norms more specifically than traditional values docs.
- •Granola Recipes = reusable prompts applied to captured meeting transcripts
- •AI can infer “how we actually work” from repeated meeting behaviors and language
- •The output can be richer and more specific than formal culture statements
- •Continuous data collection over time increases the power/accuracy of insights
- 10:49 – 13:38
Operationalizing culture: from inferred norms to hiring and performance tools
They discuss applying the inferred culture output to practical artifacts like job descriptions, hiring/firing expectations, and scoring prompts for interviews. Claire notes the value of stress-testing stated values against observed behavior in daily communication.
- •Use inferred culture to generate interview scoring criteria and rubrics
- •Feed culture outputs into job descriptions and performance expectations
- •Compare stated values vs. lived behaviors to find misalignment or hidden strengths
- •Culture is a CEO responsibility that AI can support with new data sources
- 13:38 – 16:50
Always-on feedback: AI coaching bots for meetings (including the CEO)
Wade describes using AI as an “infinitely patient coach” that provides more feedback than humans can. This helps overcome power dynamics that often prevent candid coaching, especially for executives.
- •AI feedback bots provide consistent coaching across meetings
- •Helps counteract power dynamics that limit honest feedback to CEOs
- •Useful even when you don’t fully agree—acts as a prompt for reflection
- •Scales feedback volume beyond what managers/peers can realistically provide
- 16:50 – 18:49
Building an interview evaluation agent in Zapier Agents (Granola → Zapier)
Wade shows a Zapier Agent triggered when Granola adds an interview note to a folder. The agent evaluates the transcript against the job description and Zapier values, then emails a yes/no/maybe recommendation with rationale as a bias check and thought partner.
- •Trigger: new Granola interview note in a specific folder
- •Knowledge sources: job description + Zapier values rubric (Google Docs)
- •Outputs: yes/no/maybe recommendation + 3–5 sentence reasoning emailed to Wade
- •Benefits: bias check, cross-discipline support, consistent evaluation structure
- 18:49 – 22:31
Improving agent prompts with Copilot: adding guardrails and removing PII
They use Zapier Copilot to update the agent’s instructions—specifically to strip personally identifiable information from outputs. The segment emphasizes prompt copilots as practical tools that raise prompt quality without requiring expert prompt-writing.
- •Copilot rewrites instructions to remove candidate PII automatically
- •Prompt quality matters; tools that “improve my prompt” speed iteration
- •Copilot helps convert rough intent into more SOP-like structured instructions
- •You can still manually edit instructions after Copilot’s draft
- 22:31 – 25:11
Enhancements to the interview agent: coach the interviewer + faster triage
Claire proposes two upgrades: include feedback on the interviewer’s performance and put the hiring recommendation in the email subject line for speed. Wade implements the suggestions quickly, highlighting how idea generation is often the bottleneck, not execution.
- •Add “interviewer coaching” section (missed topics, better probing, rubric coverage)
- •Put yes/no/maybe in the subject line to accelerate decision-making
- •Agent-building = writing down steps like an SOP; Copilot helps formalize it
- •Constraint is creativity/idea generation more than tooling or build time
- 25:11 – 33:39
Common agent mistake: copying today’s workflow instead of imagining ‘infinite interns’
Claire and Wade discuss how people underuse agents by only automating what they currently do. The better approach is to imagine ideal execution with unlimited time/resources, then translate that expanded workflow into agent steps—unlocking tasks previously not worth the cost.
- •Start with current process, then ask: what would I do with more time?
- •Design for the “nth degree” version of the workflow, not the constrained version
- •AI enables economically-previously-impossible tasks to be done cheaply and consistently
- •Agents thrive on tedious, repetitive, low-value-to-humans activities
- 33:39 – 34:40
Sourcing ‘diamonds in the rough’ with Grok: recruiting beyond LinkedIn
Wade demos using Grok to find under-the-radar social media talent by querying X for creators aligned with Zapier/no-code/automation. They iterate on constraints (modest following, geography, avoiding bots) and extend the approach to YouTubers and influencer sourcing.
- •Use natural-language search to find niche creators on X outside LinkedIn funnels
- •Iterate constraints: modest following, location filters, real-face profile pics, bot avoidance
- •Use cases: recruiting, influencer marketing, product feedback/customer discovery
- •Insights can reveal market hotspots (e.g., geographic clusters of no-code activity)
- 34:40 – 41:27
Recap + lightning round: talent demand, job evolution, and prompt style
Claire summarizes the end-to-end recruiting/culture workflows, then asks about roles that remain competitive. Wade says top talent is in demand everywhere—especially engineering—while hyper-specialized “promptable” analyst tasks are at risk unless elevated; he ends with his pragmatic prompting style when models misbehave.
- •Recap: meeting-to-values culture extraction, coaching bots, interview agent, Grok sourcing
- •Enduring demand: engineering and leadership; top talent remains scarce across functions
- •Risk area: narrow specialist roles where the core task is now largely AI-executable
- •Prompting: polite by default, curt iteration when needed; ‘tricks’ likely don’t matter much
