How I AIHow custom GPTs can make you a better manager | Hilary Gridley (Head of Core Product at Whoop)
CHAPTERS
- 0:00 – 3:13
Managers, meet your “AI clone”: leverage without replacement
Hilary frames custom GPTs as a way to scale managerial judgment and coaching—handling the first 60–70% of feedback so managers can spend time on strategy and people development. She emphasizes that GPTs won’t replace good managers, but can remove busywork and accelerate iteration.
- •GPTs as managerial leverage: offload the “0 to 60%” work
- •AI as a coaching multiplier, not a manager replacement
- •Why this matters: reclaim time for strategic work and hands-on coaching
- 3:13 – 3:57
From intuition to rubric: defining what “good” looks like
Hilary explains that great managers articulate “good” and “excellent,” which is often hard even for experienced leaders. Her starting point for any GPT is capturing her taste as explicit criteria that can be communicated and reused.
- •Articulating taste is a core management skill
- •Start every GPT by asking: “What does good mean to me?”
- •Rubrics reduce ambiguity and frustration for teams
- 3:57 – 6:01
Reverse-engineering your preferences with good/bad examples (low-tech, high signal)
She shares an “easy but longer” approach: collect before/after examples of work (like slides) and let the model infer patterns. Hilary uses a simple PDF workflow—left column bad, right column good—to help the AI extract evaluation criteria.
- •Build datasets from real edits: before vs. after
- •Use simple labeling (bad vs. good) to teach taste
- •PDF upload workflow to keep examples portable
- •Option to constrain: “Only use these examples”
- 6:01 – 8:31
Prompting style: start vague to discover, then go hyper-specific
Hilary intentionally avoids overly specific prompts at first to prevent biasing the model’s interpretation. After the AI suggests initial criteria, she tightens language aggressively—using directives like “Be 100 times more specific” and even percentage-based creativity tuning.
- •Begin broad: let the model surface unexpected patterns
- •Then converge with precision to remove ambiguity
- •Favorite directive: “Be 100 times more specific”
- •Use “20% more creative” vs. “1000%” to calibrate outputs
- 8:31 – 12:35
Turning criteria into a manager-grade evaluation system
They connect management evaluation to AI eval thinking: criteria, scoring, and consistent feedback loops. Hilary iterates by challenging, re-ranking, and asking what’s missing, using a diverge/converge workflow until the rubric matches her judgment.
- •Criteria creation mirrors AI eval design
- •Iterate: re-rank importance, add missing dimensions
- •Use “What am I missing?” to broaden the rubric
- •Ask the model to justify which criteria matter most
- 12:35 – 15:24
Why articulating taste improves employee experience (even before GPTs)
Claire reflects on how unclear feedback (“make it look like a thing”) stalls growth, and how a rubric would have helped. Hilary underscores that managers have hard-won judgment but lack time/brain space to explain it repeatedly—GPTs can package that guidance for juniors.
- •Rubrics apply beyond slides: interviewing, talent, writing
- •Clear criteria reduces guesswork and churn
- •GPTs can deliver patient, always-available coaching
- •Accelerates junior growth without demanding constant manager bandwidth
- 15:24 – 19:08
Demo build: creating the “Deck Doctor” slide-deck evaluator GPT
Hilary shows how to convert her rubric into a custom GPT by asking the model to write the GPT instructions. She uses “my job / your job” framing, adds a “ruthlessly helpful” tone, and requests structured output like per-criterion ratings before detailed feedback.
- •Ask AI to write the GPT prompt/instructions
- •Use role clarity: define “my job” and “your job”
- •Design for honesty: avoid false praise with ‘ruthlessly helpful’ framing
- •Add structure: 1–5 ratings per criterion before narrative feedback
- 19:08 – 21:23
Testing with real artifacts: upload a PDF and evaluate results
They test the GPT by uploading a PDF deck, highlighting how GPTs eliminate re-prompting and lower the skill barrier for teammates. The evaluator produces scores by criterion and specific rewrite suggestions, and Hilary notes she’d iterate further to improve quality.
- •Testing workflow: create → upload deck PDF → run
- •Benefit: no prompting expertise required from the user
- •Outputs: per-criterion scores + targeted recommendations
- •Iterate based on results; ship early and refine later
- 21:23 – 21:37
Adoption and scaling across a team: beta test, then let useful GPTs ‘go viral’
Hilary shares a pragmatic rollout approach: give a GPT to one person, see if it sticks, then expand. She argues you shouldn’t over-polish upfront—use real usage as the filter for which GPTs deserve improvement.
- •Start with a single beta user to validate value
- •Silence = not useful; engagement = scale it
- •Don’t over-invest upfront; usage determines winners
- •Lowering onboarding friction increases AI adoption
- 21:37 – 23:43
Personalized coaching GPTs: anticipating exec questions and meeting dynamics
Beyond generic evaluators, Hilary tailors GPTs to specific feedback for specific individuals—like generating likely questions from different stakeholders and rehearsing responses. Claire adds that AI meeting notes can supply real question datasets to train these tools.
- •Create niche GPTs based on individual growth feedback
- •Prompt: generate questions by job title/stakeholder
- •Use AI as practice partner for executive Q&A
- •Leverage meeting notes as training/evidence of real questions
- 23:43 – 30:22
Demo: using AI to strengthen writing and “get pulled into” the cool meetings
Hilary explains that strong written POVs can earn influence and meeting invites by spreading internally. She demonstrates a writing workflow: paste a rough draft, ask the AI to restate the thesis and support (instead of “is it good?”), then pressure-test with blind spots and clarity-focused rewrites—while keeping the final voice hers.
- •Career tactic: write compelling POVs that spread
- •Prompt: ‘Restate my thesis and supporting points’ to test clarity
- •Use AI to find blind spots and challenge assumptions
- •Rewrite for clarity repeatedly; final draft remains human-authored
- 30:22 – 34:42
Lightning round: who’s adopting AI, fun personal use cases, and tool ‘triangulation’
Hilary discusses research suggesting women (even high-achieving women) are adopting AI tools less, which concerns her given the leverage AI provides. She shares personal uses—reading companion in voice mode and crafting shopping lists—and her technique for stubborn tools: consult another model/tool for a workaround.
- •Concern: women being left behind in AI adoption
- •Reframe AI as accessible and fun—not a ‘tech bro’ domain
- •Personal use: voice-mode reading companion (no spoilers)
- •Personal use: craft project shopping lists and planning
- •When stuck: ask other models/tools for strategies
- 34:42 – 36:08
Where to find Hilary and closing notes
Hilary shares where people can follow her work (Substack), learn in-depth (Maven course), and join a women-in-AI community (Grrls in the Loop). Claire closes with standard show sign-offs and where to watch/listen.
- •Hilary’s Substack: hils.substack.com
- •Maven course: using AI to be a ‘super manager’
- •Community: Grrls in the Loop (grrlsintheloop.ai)
- •Podcast wrap-up: subscribe, comment, listen on audio platforms
