Aakash GuptaAutomate Your Entire Work Life With Claude Code — No Coding Needed
CHAPTERS
- 0:00 – 1:53
Claude Code as a personal operating system: why it beats a human assistant
Aakash introduces Dave Killeen and the premise: Claude Code is enabling “personal operating systems” that automate knowledge work end-to-end. Dave explains why this AI-driven setup is faster, more reliable, and compounds in value over time through “living files.”
- •Claude Code’s rapid adoption framed as a major shift in how work gets done
- •Dave’s claim: system never forgets, stays continuously up to speed, and improves over time
- •Concept of “living files” that get smarter as more context is captured
- •Dave’s background (BBC, MailOnline, now Field CPO at Pendo) sets credibility
- 1:53 – 2:50
Live demo: the morning “daily plan” command pulls everything together
Dave runs a single morning command that assembles a complete daily brief. The system automatically gathers calendar, goals, meeting notes, and multiple intel feeds into a structured plan without manual prep.
- •Daily plan command aggregates calendar, weekly goals, quarterly goals, and more
- •Meeting notes flow in automatically (e.g., via Granola)
- •External intel streams appear (YouTube, LinkedIn, newsletters)
- •Core promise: one command yields a prioritized, actionable daily view
- 2:50 – 3:45
Why Cursor works for non-engineers: everything is markdown, AI finds files for you
Aakash asks what tool they’re seeing; Dave explains Cursor as a developer environment that’s surprisingly effective for non-engineering workflows. The key is keeping everything as markdown/text files so the AI can read, append, and reorganize information easily.
- •Cursor used as a convenient UI for navigating many files during the demo
- •System is primarily markdown files: easy for AI to manipulate and extend
- •Users don’t need to know file locations—AI can retrieve and update content
- •Personal OS is essentially structured text plus automations
- 3:45 – 5:36
Connecting tools with MCP servers (and voice-to-text workflow)
Dave explains how he connects many services using MCP servers built from API docs, emphasizing guardrails and reliability. He also covers his voice-first workflow using Whisperflow/Superwhisper to avoid typing and speed up interaction.
- •MCP servers preferred over direct API calls because they provide better AI guardrails
- •Workflow: point Claude at API docs + API key → have it generate an MCP server
- •MCP helps AI propose new use cases after ingesting documentation
- •Voice stack: Superwhisper modes vs Whisperflow adoption and simplicity
- 5:36 – 9:03
Daily plan output: priorities, account intelligence, and multi-channel market signal
They review the daily plan output: top priorities, scheduling suggestions, and account-level guidance based on customer conversations and internal tools. The same page summarizes high-signal insights from YouTube/newsletters/Twitter and drafts messages to send.
- •Daily plan shows “three things that matter,” weekly priorities, and time-block suggestions
- •Redaction approach for sensitive company data via instruction to the AI
- •Account intelligence: flags which deals/accounts need attention and drafts Slack messages
- •Intel digests: clusters “novel/contrarian” insights across YouTube, newsletters, Twitter
- •LinkedIn outreach linked to CRM via PhantomBuster for prioritized responses
- 9:03 – 11:46
Dex skills + X-ray: making automations explainable and teachable
Dave introduces Dex (open source) and the idea of many reusable “skills/commands.” The X-ray skill generates mermaid diagrams and explanations that show how commands work, helping users learn and customize the system.
- •Dex contains many reusable skills (Dave mentions ~60)
- •X-ray skill “pulls back the curtain” with diagrams and hook usage explanations
- •Daily plan flow checks what intel files exist, generates missing digests, then composes plan
- •Goal: increase AI fluency by learning from your own working system
- 11:46 – 14:12
Terminal vs Cursor: Claude Code hooks and compounding sessions
They contrast using Claude via Cursor chat with using Claude Code in the terminal. Dave argues terminal is superior because hooks (especially session-start hooks) reliably inject context and prevent repeated mistakes, enabling compounding performance.
- •Hooks are more available/powerful in terminal than in Cursor
- •Session-start hooks preload goals, priorities, learnings, and guardrails into new chats
- •System becomes more effective as it accumulates context and corrections
- •Strategic implication: compounding workflows beat isolated chat sessions
- 14:12 – 15:20
Account health scoring and proactive leadership workflows
Dave demonstrates another key skill: generating account health scores to decide where to intervene. The aim is proactive support for sales/customer teams without needing people to escalate issues manually.
- •Health-score skill quickly surfaces at-risk accounts/deals needing attention
- •Enables proactive Field CPO interventions
- •Shows power of composable skills: each new command adds leverage
- •Narrative: focus attention where it matters most each day
- 15:20 – 19:00
From backlog idea to PRD: prompting for “taste,” not just output
They show how Dex ranks backlog ideas and then expands one into a full PRD. Dave and Aakash stress that AI can draft comprehensive documents, but humans still provide taste, commercial context, and refinement.
- •Backlog ideas are ranked by impact/alignment/token efficiency; AI can propose ideas too
- •Prompting style: push for 10x/delight/serendipity; emotional prompting can help
- •Taste and orchestration are positioned as the enduring PM/CPO value
- •PRD quality: strong first draft, but needs strategic/commercial framing and editing
- 19:00 – 21:37
Managing PRD overload: a local Kanban UI built in hours
Dave describes the emerging problem: AI can generate too many parallel artifacts (PRDs, agents, tasks). He demonstrates a locally hosted UI (React/localhost) that organizes PRDs into a Kanban board with scoring and next-step recommendations.
- •AI creates abundance—risk is cognitive overload and lost thread of work
- •Local Kanban organizes PRDs, shipped items, and what to do next
- •Built quickly via chat-driven development; future plans include Electron/mobile
- •“Malleable software” theme: build custom tools around your own workflow
- 21:37 – 28:32
Claude.md and “harsh truths”: using the AI to keep the system from sprawling
They inspect the Claude.md file that encodes product identity, behaviors, and guardrails for Dex. Dave explains “progressive disclosure” to keep prompts small, and highlights using the AI as a brutally honest sparring partner to reduce bloat.
- •Claude.md contains identity, principles, and sparring instructions (e.g., bloat radar)
- •Progressive disclosure: keep Claude.md short and link out to deeper files
- •AI can audit Claude.md and suggest moving key constraints into hooks
- •“Harsh truths” feedback loop: AI critiques the system to improve maintainability
- 28:32 – 33:35
Career planning with a Career MCP server: goals → quarters → weeks
Dave demonstrates a career-focused MCP server that collects feedback/evidence and maps long-term goals into quarterly and weekly plans. The system identifies gaps, performs skills analysis, and provides course-correction guidance.
- •Career MCP scans for evidence from meetings and work artifacts to build a feedback corpus
- •Skill gap analysis and promotion readiness scoring driven by accumulated evidence
- •Mermaid visualization: career goals ladder into quarterly goals and weekly actions
- •Weekly planning uses this to flag neglected priorities and recommend adjustments
- 33:35 – 36:26
Skills vs MCP vs hooks: what each is for and when to use them
Dave clarifies the architecture: skills/commands are procedural job descriptions, MCP provides deterministic guardrails for integrations, and hooks inject context or capture learnings automatically during sessions. This segment anchors how to design reliable personal OS behavior.
- •Skills/commands: step-by-step instructions that may be followed inconsistently
- •MCP: tighter, more deterministic interactions with external services and internal standards
- •Example: task MCP enforces consistent task structure and categorization
- •Hooks: event-based automation (e.g., session start) for persistent compounding context
- 36:26 – 38:57
Intelligence scanning and the compounding advantage over ChatGPT
Dave explains his daily “intel” pipeline: ingesting YouTube transcripts, newsletters, and Twitter bookmarks, then clustering themes and extracting contrarian/novel insights. They contrast this file-based, compounding system with isolated cloud chats.
- •Automated ingestion: transcripts/newsletters/bookmarks → markdown files in the system
- •Clustering across sources to surface themes and high-signal insights
- •“Living files” accumulate context for projects, stakeholders, and companies over time
- •Compounding differentiator: persistent, inspectable memory vs ephemeral chat threads
- 38:57 – 52:58
Dex improve + hooks deep dive + getting started (setup, tool choices, Pi/OpenClaw)
They cover Dex’s self-improvement workflow (scanning releases and community signals), then go deeper on hooks as the mechanism for durable learning (preferences, mistakes). The episode closes with onboarding steps, when to use Cursor vs terminal, and thoughts on LLM-neutral agent frameworks like Pi/OpenClaw.
- •Dex improve scans Anthropic changelog + Hacker News/Reddit and proposes upgrades
- •Hooks examples: session-start context injection, preferences capture, mistakes file
- •Onboarding: clone repo, open in Cursor, run /setup to scaffold role-based configuration
- •Tool choice guidance: start in Cursor; move to terminal for hooks; future UI planned
- •Pi/OpenClaw discussed as LLM-neutral agent harness direction; overhyped vs underhyped framing
