CHAPTERS
- 0:00 – 6:51
Why CEOs can’t just “send the AI memo”: creating safe space to experiment
Claire and Wade open by critiquing the common leadership failure mode: delegating “AI adoption” down the org without support. Wade argues leaders must create structured time—hackathons, show-and-tells, and play space—so teams can touch tools, reduce fear, and learn pragmatically.
- •The “delegation trap” leaves ICs to solve adoption for the whole company
- •Leaders should sponsor hackathons/show-and-tells to build comfort
- •Hands-on experimentation reduces anxiety and clarifies AI’s real strengths/limits
- •AI becomes less of a boogeyman once people see it fail and succeed in context
- 6:51 – 7:31
Making AI adoption measurable: AI fluency rubrics and incentives
Claire highlights Zapier’s approach to operationalizing AI capability with role-based fluency rubrics (e.g., for PMs). The point is to shift behavior by making expectations explicit and aligning rewards and measurement to AI-enabled work.
- •Rubrics clarify “what good looks like” at different role levels
- •Measurement and rewards drive adoption more than slogans
- •AI can help draft and improve competency frameworks
- •Focus is on practical skill-building, not vague transformation talk
- 7:31 – 10:16
Turning meetings into a culture handbook with Granola “Recipes”
Wade demonstrates a hack: using Granola meeting transcripts plus a prompt (“unspoken company culture handbook”) to infer how the company actually operates. He’s surprised by the specificity—capturing real norms beyond what’s written in official values docs.
- •Granola Recipes are reusable prompt templates over meeting data
- •Meeting transcripts reveal the lived culture (“how Wade works” and adjacent teams)
- •AI can extract what the org rewards vs. discourages
- •Longitudinal data makes AI outputs far more powerful and concrete
- 10:16 – 12:58
From inferred culture to hiring assets: rubrics, JDs, performance expectations
They discuss how the ‘unspoken culture’ output becomes actionable: feeding job descriptions, interview scoring prompts, and even performance management. Claire adds the idea of stress-testing stated values against actual communication patterns to spot misalignment.
- •Use inferred culture to generate interview scoring prompts
- •Convert culture insights into job descriptions and onboarding expectations
- •Stress-test stated values vs. observed behavior in communications
- •AI can help leaders do “the CEO job” of culture-carrying more effectively
- 12:58 – 14:13
Always-on coaching: using AI feedback bots to improve leadership and meetings
Wade explains why AI coaching bots are valuable, especially for CEOs who don’t always get candid feedback due to power dynamics. AI is framed as an infinitely patient coach that can offer continuous, scalable feedback across the organization.
- •Power dynamics limit honest human feedback; AI can fill the gap
- •Feedback bots provide both praise and critical improvement notes
- •Coaching extends beyond employees to executives and the CEO
- •Consistent feedback loops can reinforce cultural behaviors in meetings
- 14:13 – 16:44
Building a Zapier interview evaluation agent (Granola → Zapier Agents)
Wade walks through a practical agent that triggers when a Granola interview note lands in a folder, then evaluates the candidate against a job description and Zapier values. The agent returns a yes/no/maybe recommendation and emails Wade a concise rationale.
- •Trigger: new Granola note in an interview folder
- •Knowledge sources: job description + values rubric docs
- •Outputs: recommendation (yes/no/maybe) + 3–5 sentence reasoning
- •Used as bias check and thought partner across disciplines
- 16:44 – 19:27
Prompt iteration with Copilot: removing PII and improving the agent instructions
They show how Zapier’s Copilot helps rewrite agent instructions—first to strip personally identifiable information for safe demoing. Claire and Wade emphasize that prompt quality matters, and Copilot lowers the barrier to writing structured “SOP-like” instructions.
- •Copilot can rewrite prompts to enforce constraints (e.g., remove PII)
- •In-product ‘improve prompt’ features help non-experts write better instructions
- •Prompt quality materially affects outcomes, despite skepticism
- •Agents can be edited like living SOPs as needs change
- 19:27 – 22:32
Enhancing the interview agent: coach the interviewer + optimize decision speed
Claire proposes two upgrades: have the agent evaluate interviewer quality (missed questions, rubric coverage) and put the yes/no/maybe decision in the email subject line to speed hiring. Wade notes how fast it is to implement improvements once you have ideas.
- •Add an ‘interviewer feedback’ section to improve consistency over time
- •Surface decision outcome in the subject line to accelerate hiring motion
- •Treat agents as iterative systems that improve with examples and guidelines
- •The main bottleneck is idea generation, not implementation speed
- 22:32 – 25:45
Common agent mistake: copying today’s workflow instead of imagining “infinite interns”
Claire explains a creativity unlock: don’t only automate what you currently do; ask what you would do with more time, or with multiple interns, to reach an ideal process. Wade connects this to a broader theme: AI makes previously uneconomical tasks viable.
- •Start with current steps, then expand to the ideal end-state workflow
- •Ask: what would I do next, and after that, with more capacity?
- •AI enables low-cost, consistent execution of tedious or neglected tasks
- •Big value comes from new work that never happened before—not just speedups
- 25:45 – 30:28
Sourcing ‘diamonds in the rough’ with Grok on X (and beyond LinkedIn)
Wade shows a recruiting workflow using Grok to find under-the-radar social/media talent on X—tutorial creators, fans of Zapier/no-code/agents, with modest followings and outside the Bay Area. They iterate prompts to reduce bots and refine the candidate pool.
- •Grok enables natural-language sourcing queries that beat Boolean search
- •Targets creators who teach automation/no-code and have modest followings
- •Iterate constraints: geography, authenticity (not bots), profile photo cues
- •Use it to widen sourcing surface area beyond LinkedIn norms
- 30:28 – 33:40
Operational lessons from Grok results: segmentation, community strategy, and limitations
They note Grok’s outputs can be imperfect (bots, misfit accounts, missing links), requiring back-and-forth refinement. Claire points out that even imperfect lists can reveal market hotspots (e.g., regional pockets of no-code activity) that inform community and go-to-market ideas.
- •Key limitation: bot detection and noisy candidate lists
- •Prompt refinements can reduce low-quality results (e.g., username heuristics)
- •Insights can inform community/events strategy, not just hiring
- •Extend sourcing to YouTube/influencers; request channel links for usability
- 33:40 – 35:10
Recap of the end-to-end recruiting + culture AI stack
Claire summarizes the featured workflows: meeting-to-values culture extraction, always-on coaching, an interview evaluation agent, and Grok-based sourcing for “terminally online” creators. The episode frames this as an AI-native CEO approach to culture and recruiting.
- •Granola transcripts → inferred culture artifacts
- •AI coaching for meetings and leadership behavior
- •Zapier Agents for structured candidate evaluation + process improvements
- •Grok for discovering creator talent across social platforms
- 35:10 – 41:27
Lightning round: durable roles, organizational change, and Wade’s prompting style
Wade argues demand for top talent remains everywhere, especially engineering, but hyper-specialized “single-task” roles are at risk unless elevated with AI. He reflects on Zapier’s evolving culture (automation everywhere, more coaching) and shares his personal prompting habits when models misbehave.
- •Still hiring aggressively—especially engineering and engineering leadership
- •Top talent remains scarce; ‘top’ now includes AI leverage and adaptability
- •Risk: fleets of narrow analyst-style roles that AI can largely replace
- •Prompting style: polite by default, curt when iteration is needed
