How I AIClaude Code for normal people: skills, voice mode, and how to collaborate with AI
CHAPTERS
- 0:00 – 3:49
Intent engineering over prompt engineering: letting Claude “pull” the prompt out of you
Grace reframes effective AI use as “intent engineering,” where you describe outcomes conversationally and let the model infer structure and next steps. She shares how short voice explanations can outperform meticulously crafted prompts, especially when you’re mobile and thinking out loud.
- •Prompt engineering is “dead”; focus on intent and outcomes
- •Voice-first workflows: talk for 2–3 minutes, let Claude infer the artifact
- •Claude should extract the prompt from you through conversation
- •Aim for interactive outputs (e.g., password-protected artifacts) rather than text blobs
- 3:49 – 4:49
Grace’s path: from marketing/teaching to building Claude-powered business workflows
Grace explains her background as a marketing consultant turned AI teacher and how documenting her learning led to teaching others. Her personal need for efficiency and better client experience pushed her toward building reusable Claude-based systems.
- •Relationship- and process-driven work created strong automation opportunities
- •Learning Claude/OpenClau turned into a teachable curriculum
- •She built real business tools first, then productized them for students/teams
- •Goal: reduce admin overhead so she can spend time teaching and serving clients
- 4:49 – 6:43
The “pipeline operator”: an hourly agent that runs key business ops
Grace introduces her most impactful system: a pipeline operator that runs every hour to manage email, client status, and deliverables. It’s designed to reduce tab overload, improve client communication warmth, and keep projects moving without constant manual triage.
- •Runs on a timer and checks multiple signals to move clients through steps
- •Ingests email + context to handle a deluge of communications
- •Improves client experience with more “emotion and love” via rich artifacts
- •Eliminates the chaos of 20 tabs and heavy weekly admin time
- 6:43 – 10:15
AI-first client deliverables: branded, interactive HTML proposals and onboarding
Claire and Grace discuss why highly customized, fast proposals and onboarding materials set a new bar for service. Grace shows HTML-based, branded, password-protected proposals that double as onboarding hubs and previews of what clients can build.
- •Proposal output is interactive HTML: branded, polished, and contextual
- •Password-protected artifacts create a premium, tailored experience
- •Auto-generates pre-work, pulls updated documentation, and tracks progress
- •Adds questionnaires to identify themes and readiness before sessions
- 10:15 – 12:15
Adoption is muscle memory: forcing functions to make Claude the default
Grace breaks down why seeing AI benefits isn’t enough—people must build the habit of using it. She shares practical forcing functions like reminders and screenshot-to-Claude workflows to create repeatable collaboration patterns.
- •Key barrier: defaulting to old tools (Gmail/Slack) instead of Claude
- •Forcing function: recurring reminders to screenshot your task into Claude
- •“Shadow me” prompting: ask Claude for help from an image, no text required
- •Teaching technical terms builds confidence and reduces “I’m not technical” fear
- 12:15 – 14:31
What’s inside a “skill file”: SOP verbalization, versioning, rules, and structure
Grace shows how she turns spoken SOPs into durable, invokable skills that Claude can run repeatedly. She emphasizes documenting standards (voice + proposal definition), versioning with changelogs, and writing clear workflow rules Claude can follow.
- •Three-step architecture: document standards → timer → publish (e.g., Netlify)
- •“Yap an SOP” exercise: narrate steps like training a new employee
- •Skill files include changelogs, naming conventions, and deal-shape checks
- •Defines teaching vs consulting philosophy and proposal layout rules
- 14:31 – 16:45
How she built the proposal maker: mobile voice notes, no “Claude interview me”
Grace explains her process for creating the proposal skill starting from a walk and a short voice note. She resists the interview-style prompting pattern, instead pushing Claude to propose the solution, then iterating with feedback and screenshots.
- •Built largely from a phone: quick voice prompt + iterative feedback
- •Rejects “let Claude interview you”; Claude should leverage existing context
- •Claude proposes solution (interactive password-protected artifact), user reacts
- •Iteration loop: generate HTML, pressure test, share screenshots of likes/dislikes
- 16:45 – 21:30
The voice guide: teaching Claude how she thinks (anti-slop, decision style, language)
Grace highlights her “pride and joy” skill: a voice-and-thinking guide that shapes outputs across tasks. It’s a living document updated via voice notes to avoid generic AI tone and to encode decision-making preferences, not just phrasing.
- •More than writing style: a “think like me” decision and philosophy guide
- •Captures words to use/avoid to prevent recognizable AI ‘slop’
- •Created by telling Claude the goal, then having Claude study and draft it
- •Continuously updated with examples of what she never wants to sound like
- 21:30 – 25:22
Rebuilding Gmail inside Claude: a personal workflow that compounds learning
Grace shares a Gmail replacement built from frustration that lives inside Claude (via Cowork artifacts). The motivation is to keep work and learning inside the AI environment so improvements compound, instead of being trapped in traditional email UI.
- •Goal: stop opening Gmail and handle email triage/replies through Claude
- •Artifact-based UI in Claude Cowork recreates a customized inbox experience
- •Keeping the workflow in Claude helps the system learn and improve over time
- •Positioned as a universally useful beginner project (fast, motivating, customizable)
- 25:22 – 28:05
Live demo: drafting and pushing replies to Gmail from the Cowork artifact
In a live run, Grace drafts a reply and uses a “Send in Gmail” action to create a Gmail draft and open a browser for final steps. Even with minor hiccups, the workflow keeps her mostly out of Gmail while still sending real messages.
- •Draft reply inside the Claude artifact, then push to Gmail as a draft
- •Workflow can open a browser to complete sending when needed
- •Even imperfect automation reduces time spent in Gmail’s interface
- •Key value: iterate with Claude in-context, then send without manual rewriting
- 28:05 – 31:23
Under the hood: connectors, custom plugins, and why she starts in Claude Code
Grace explains the technical setup: connecting data sources, adding custom connectors/plugins, and handling Google permissions. She prefers Claude Code for ambiguous builds because it’s faster and more proactive, then hands off to Cowork for better UX.
- •Connect “everything you can,” then build custom connectors for gaps
- •Google integrations required a Google Cloud project, service accounts, scopes
- •Claude Code is more proactive (alternative approaches when official connectors fail)
- •Session handoff: export a markdown file from Code, drag into Cowork to continue
- 31:23 – 35:07
Tiny daily Claude habits: workout tracker, plant ID, and living in voice mode
Beyond big systems, Grace uses Claude for small recurring personal workflows that keep the habit strong. She voice-notes workouts to update a spreadsheet automatically and shares plant photos to build a single, context-rich personal thread.
- •Voice-note workouts → Claude updates a tracking spreadsheet automatically
- •“Let go and let Claude file things” as a mindset for sustainable use
- •Plant photo logging/ID keeps gardening context centralized
- •Mobile-first usage reduces friction and increases consistency
- 35:07 – 38:50
Biggest misconception: people don’t need ‘mondo prompts’—they need collaboration practice
Grace reflects on teaching: heavy pre-written prompts can backfire by doing the thinking for students. The real unlock is learning to collaborate with AI and building the reflex to bring everyday work to Claude, especially for reluctant adopters.
- •Giving overly directive prompts can hinder learning and confidence
- •Core shift: from prompting to collaboration and iteration
- •Muscle memory and empathy matter, especially with “forced” trainees
- •Simple habit: screenshot a task and ask Claude for help to start collaborating
- 38:50 – 43:08
When Claude disappoints: direct feedback, iteration, and a real-time proposal check
Claire and Grace compare notes on what they do when outputs are mediocre: they get blunt, tighten feedback loops, and re-run. They then inspect a proposal Claude generated for Claire, verifying brand elements and content quality in real time.
- •Strategy is candid correction rather than “gentle parenting”
- •Frustration comes from knowing the model can do better—so feedback gets sharper
- •They review a Netlify-hosted proposal artifact Claude generated
- •Quality control: check visual brand (fonts/colors), interactivity, and homework/tasks
