Automate Your Entire Work Life With Claude Code — Akash Gupta with Dave Killeen
🎥 Watch the full episode here
📆 Published: March 2026
🕒 Estimated Reading Time: 10 mins. Time saved: 45+ mins! 🔥

Fair warning - this is a big one. This Pod Shot goes deeper than usual because, frankly, Dex deserves it. I've added external research, contrarian perspectives from HBR and MIT, and cross-references to my own Claude Code series to give you the full picture - not just what Dave says, but whether the evidence backs it up. Here's how to navigate it:

Pick your depth. No judgement. 🙂

Every morning, Dave Killeen wakes up and issues a single voice command: "Can we do a daily plan for the day today, please? Thank you. Love you." (Yes, he says "love you" — "I always say love you because I do think it changes how we feel about our AI, particularly when we're using voice to text.") Within seconds, his AI Chief of Staff aggregates his calendar, quarterly goals, sales forecasting data from Clary, email newsletters, YouTube summaries, and LinkedIn intelligence — then delivers three daily priorities aligned with his strategic objectives. No scrambling through tabs. No rebuilding context from yesterday.

After 8 months of building with Claude Code, Pendo's Field CPO for EMEA open-sourced Dex — a personal operating system with roughly 60 skills that turns your meetings, relationships, tasks, and career evidence into a living, self-learning system. It's "essentially just a bunch of text files" that Claude orchestrates — but those text files become what Dave calls "malleable software": living documentation that compounds with every interaction and can be reshaped to match exactly how you work. This isn't another productivity app you'll abandon in two weeks. It's a markdown-based PKM that runs on your machine, adapts to your specific role, and gets smarter every single day.

Full disclosure: I use Dex myself. I know Dave well. I am a big fanboi. It's become central to how I manage my professional life, and this episode is the best explanation of why.

Key insights:

Let's be honest: most knowledge workers have tried and abandoned at least one PKM system. You set up Notion, or Obsidian, or Roam, with the best intentions. Two weeks later, you've stopped capturing. A month later, you can't find anything. The system became one more thing to maintain rather than something that works for you.

The fundamental problem? Traditional PKM systems require massive manual discipline. You have to decide where things go, maintain your taxonomy, remember to capture, and somehow find time to review and connect ideas. The cognitive overhead of maintaining the system often exceeds the value it provides. This isn't anecdotal - a Forte Labs survey found that 68% of PKM adopters abandoned their system within six months. One practitioner on the Bullet Journal blog captured it perfectly: "The more my system grew, the more I deferred the work of thought to a future self who would sort, tag, and extract the gold. But that self never arrived." AI is that future self.

Dave Killeen calls this the "bloat radar" problem. It's actually a stress-testing rule baked into the Dex system itself: "Check the bloat radar. Is this replacing something, or is it one more thing?" Every feature in Dex replaces a manual process. The daily plan replaces your morning scramble through calendar and task lists. Commitment tracking replaces the mental load of remembering what you promised in meetings. Relationship intelligence replaces the frantic pre-meeting LinkedIn stalk.

AI changes the equation entirely. Dave’s core philosophy is that "ideas die between having them and recording them, and deciding where things belong kills momentum." With Dex, you don't organise — the system organises for you. You talk to it naturally, and information routes automatically to the right place: people pages, projects, tasks. The friction of capture drops to near zero, and the system's ability to surface the right context at the right time makes retrieval effortless.

Key Takeaways:

Dex isn't a SaaS product with a login page. It's an open-source starter kit that turns Claude into a role-specific operating system, running entirely on your machine. Here's what that means in practice.

The research lab Ink & Switch calls this the broken promise of personal computing: "We were promised a new kind of clay, but instead, we got appliances: built far away, sealed, unchangeable." Dex is literally the clay they describe - software you reshape to match exactly how you work.

The Foundation: Markdown Files
Everything in Dex is a markdown file organised into directories: Inbox, Quarter Goals, Week Priorities, Tasks, Projects, Areas, Resources, and Archives. As Dave explains: "I never look at any of it. I just trust that the AI knows where to find things." These aren't static documents — they're "living files" that AI "dances with." Metadata compounds over time as the system adds meeting notes, stakeholder feedback, and project updates automatically. Granola meeting transcripts, for example, arrive automatically 30 minutes after each meeting ends.

The Brain: Claude Integration
Dex works through a variety of access methods, each with different strengths:

The key differentiator is hooks — deterministic automation that always executes. As Dave explains: "The Claude MD file is good guidance, but having session start hooks with Claude Code is super effective, because it means you start that session and it always adheres to whatever is in that session start hook, every single time." Session start hooks inject your strategic pillars, quarterly goals, working preferences, and critically — a "mistakes file" so "all of the mistakes we've had together also get injected so those mistakes don't happen again."

Worth noting: UX Planet's analysis of CLAUDE.md best practices warns that over-specified context files become counterproductive - "important rules get lost in the noise." Dave's hooks solve this elegantly by injecting context deterministically at session start, rather than relying on a single monolithic file.

The Connective Tissue: MCP Servers
Model Context Protocol servers provide guardrailed, deterministic connections to external tools. Dave draws a crucial distinction here: skills (now called "commands") are plain English job descriptions that sometimes fail to invoke or execute inconsistently — useful for one-off requests. MCP servers enforce deterministic behaviour with tight guardrails for external service interaction — essential for recurring, mission-critical processes. Dave even directs Claude to API documentation and asks it to create MCP servers for him, generating the guardrails that prevent probabilistic errors. Dex uses them for meeting transcription (Granola), task synchronisation, calendar integration, and more.

Key Takeaways:

This is where it gets exciting. Dex isn't a general-purpose chatbot — it performs eight specific jobs that cover the cognitive overhead of professional life.

Every morning, Dex generates three priorities organised as Must/Should/Could — not a flat list of 47 tasks. If your calendar is overbooked, it automatically reduces the list. As Dave puts it: "I can't be across 45 enterprise deals every single week and what's happening and the nuance of all those deals, but it can for me." The output includes account health analysis identifying deals needing attention, suggested Slack messages for team coordination, and market intelligence clustering novel and contrarian signals across 120+ newsletters. It even writes the Slack messages for him. All from a single voice command.

A necessary counterpoint here: a METR study found experienced developers were actually 19% slower with AI tools - while believing they were 20% faster. And Fortune reported that after AI adoption, email time doubled and focused work sessions dropped 9%. The question isn't whether Dex feels productive - it's whether it measurably is. Dave's answer is the bloat radar: every feature must replace a manual process, not add a new one. That discipline is what separates a genuine operating system from productivity theatre.

This is the game-changer nobody talks about. Dex extracts promises from meetings — things you said you'd do, things others committed to — and flags anything older than three days. No more "I thought you were going to send that?" moments. No more commitments falling through the cracks. I've been exploring this territory myself - in my Claude Code for PMs series, I documented how context engineering and commitment tracking were consistently the highest-ROI features across 59 documented sessions. Dave's implementation takes this further by making it fully automatic.

Before every call, Dex surfaces previous conversations, open items, and shared priorities for the person you're meeting. No cold starts. No "remind me what we discussed last time." You walk into every meeting with full context, which is transformative for building trust and driving outcomes.

Here's one that pays dividends at review time. A dedicated Career MCP server scans for evidence across Granola transcripts and calls, performs skills gap analysis, generates promotion readiness scores, and connects personal goals to quarterly/weekly planning. The weekly /weekly-plan command even identifies skill gaps and suggests course-corrections: "You're leaning far too much in over here, you need to course-correct." When promotion conversations come around, you're not scrambling to remember what you accomplished — it's all documented, with evidence.

Check off a task in a meeting note, and it updates everywhere — Tasks.md, person pages, project files. The Work MCP server assigns unique IDs and handles deduplication, so you never have the same task listed in three places with three different statuses.

Dex converts mistakes into rules and adapts to your preferences. If you snap "Hey Claude, you bloody idiot, why did you do that?" — it picks that up, writes it to a working preferences file, and adjusts. Dave even had the AI audit his entire system and it came back brutal: "Dave, you're building a system more for yourself than for others." His response? "You're probably right." The AI then injected "harsh truths for Dave" into the Claude MD file. Now every session enforces those guardrails. Each session genuinely makes the next one better.

Any project without an update in 12+ days gets flagged automatically. Stalled initiatives surface before they become crises, giving you time to intervene while you can still influence the outcome.

This one blew my mind. Dex monitors Claude releases daily and suggests relevant implementations for your specific workflow. It's like having a concierge who reads release notes you'll never read and tells you which ones matter to you. The system literally upgrades itself.

Key Takeaways:

For anyone who spends their day in meetings (so, all of us), this is where Dex transforms your workflow.

Dex processes meeting transcripts — from Granola MCP or simply pasted in — and automatically extracts structured notes with action items. But it goes further than any meeting notes tool because those action items are synced to the relevant person pages, project files, and task lists.

Here's the practical flow: You finish a meeting. Granola captures the transcript. Dex processes it into structured notes, extracts commitments (yours and theirs), updates the person pages for everyone in the meeting, flags action items with deadlines, and links everything to the relevant project.

The compounding effect is powerful. After three months of use, every person page contains a rich history of interactions: what you discussed, what was promised, what was delivered, what's still outstanding. When you prep for a meeting with someone you haven't spoken to in weeks, the full context is right there.

Dave describes it perfectly: "It never forgets anything. It never needs to be brought up to speed, and it operates at the speed of our conversation."

Even Tiago Forte - the godfather of the Second Brain methodology - is pivoting BASB to be AI-first in 2026: "The context you bring to any interaction with AI matters more than ever." Dex proves his thesis - the value isn't in storing knowledge, it's in the personalised context that makes AI outputs genuinely useful rather than generically competent.

Key Takeaways:

This feature alone makes Dex worth setting up for any product leader. Dave has configured Dex to aggregate intelligence from 60+ YouTube channels (including Aakash Gupta's), 50+ newsletters (including Chamath's and Lenny's), LinkedIn messages cross-referenced against Pendo CRM using Phantom Buster, and Twitter bookmarks.

But it doesn't just dump links on you. It clusters signals across sources to identify novel and contrarian insights relevant to your specific role. The output is organised into categories: YouTube Intelligence (novel and contrarian summaries), LinkedIn Intelligence (key connection opportunities surfaced from CRM cross-referencing), and Market Intelligence (why emerging signals matter plus contrarian angles). When three different sources mention the same emerging trend, Dex connects the dots and surfaces it as a pattern worth investigating.

For product leaders, this means you're never blindsided in a strategy meeting by a competitor move or market shift. For anyone in a commercial role, it means you spot opportunities and threats before they hit your inbox as urgent escalations.

Dave also built a GitHub Repository Radar — a skill he created live on the podcast by simply asking Claude: "Create me a skill that pulls trending GitHub repos, recommends the top 10 weekly, and explains why they fit DEX context." Within minutes, the complete skill file was generated and ready to invoke via /repo-radar. It scans daily for trending repos, scoring them by potential usefulness. The system flagged tools like ScreenPipe (screen activity monitoring) as potentially valuable but controversial. As Dave reflects on the pace: "I normally sleep very well. Over the last four weeks, everything's shifted so much with Anthropic and OpenAI that your mind is always racing with lots of other things you can be doing."

Key Takeaways:

One of the most impressive things about Dex is how fast you can get running. Dave has deliberately designed it for non-engineers — product managers, executives, designers, consultants.

What You Need:

Three Steps:

The easiest starting point is Cursor (a free IDE that works with Claude). For the full experience with self-learning hooks, use Claude Code Desktop or Terminal.

A word on the "vibe coding" concern. Critics are right that AI-generated code has real risks - Red Hat found 69 vulnerabilities across 15 tested apps, and Veracode reports 45% of AI-generated code fails security tests. But Dex's risk radius is one person. Your personal operating system doesn't need a security audit or production SLA - it needs to work for you. This is the ideal use case for non-engineers building with AI.

Dave emphasises that Cursor users can be productive immediately: "Don't be intimidated by it. Don't worry about that messy left-hand rail of the files. Just trust the AI to file things the way it should be and you're pretty much good to go." Claude Code users (he recommends Ghostty terminal on Mac for a cleaner experience) get the additional benefit of hooks that guarantee context injection at every session, which is where the compounding magic really kicks in.

One crucial tip from the transcript: push your Claude MD file to GitHub so you have version control. As Aakash warns: "There have been many times I've seen a regression in my Claude MD file performance and I want to revert back."

Key Takeaways:

Beyond daily operations, Dave uses Dex for strategic product work. The PRD generation workflow is particularly impressive.

You voice-prompt Dex with a backlog idea — Dave literally says: "From the DEX backlog, can you work on idea number three and come up with a PRD? Think about how you can make this absolutely brilliant. 10x it. Don't settle for anything mediocre. Think about something that will make this brilliantly serendipitous, massively delightful." The system considers dependencies, existing solutions, and strategic commercial value, then generates a structured PRD complete with non-goals (preventing scope creep), metrics and baselines, and integration considerations. It even flags overlapping capabilities — "The last thing you want to be doing as a CPO is having your team increase more bloat into the system."

When asked if he even edits the PRDs anymore, Dave laughs: "I'm vibe CPOing. Typically it just is on the money and the AI is able to work out the edge cases and build things for you."

Dave built a Kanban board UI in React (three hours of work, accessible via localhost) that organises parallel PRDs with "Go build" buttons to spawn parallel development tasks. The AI ranks what to build next based on impact, alignment, and token efficiency — there's even a scoring competition between Dave and the AI on idea ranking. As he puts it: "You're just orchestrating everything and orchestrating this sense of taste and delight, which is brilliant."

He reframes the CPO role beautifully: "It's a bit like we're all head chefs of Michelin-starred restaurants, and we're not in the kitchen doing all the gubbins. We literally just design the menu and we have the AI chefs in the kitchen doing it for us." The broader point? Dex removes the persuasion overhead that has traditionally slowed product work. As Dave reflects on 25 years in product: "When you've got lots of ideas, nearly half the battle is trying to convince others of them and get the investment to get them built. Now that's gone."

Key Takeaways:

Here's something that makes Dex uniquely valuable for teams. Dave built a /xray command that anyone can run after any interaction. It generates Mermaid diagrams showing exactly what happened behind the scenes: yesterday's review processing, intel digest compilation, missing data identification, and structured data collection from MCP servers. You see what information the AI pulled in, what tools it used, and why it made each decision.

Daily usage becomes daily learning. The concepts you absorb — how AI uses context, how it decides what information to pull in, how to give it memory — are becoming the new literacy for anyone who leads product teams.

This is what separates Dex from a chatbot. It's not a black box you throw questions at. It's a transparent system you can inspect, learn from, and improve. Every interaction teaches you something about how to work with AI more effectively.

For teams adopting AI tools, this is invaluable. Instead of sending people on an "AI course," you give them Dex. They learn by doing, with the X-Ray feature revealing the principles behind every interaction.

Key Takeaways:

Most productivity tools follow a depressing curve: high enthusiasm at setup, declining engagement, eventual abandonment. Dex inverts this.

But let's be honest about the headwinds. HBR warns of "workslop" - AI-generated outputs that feel productive but nobody reads, costing employees two hours per week in cleanup. And economists point to Jevons Paradox: when something becomes more efficient, we tend to do more of it, not less. Workers in AI-exposed roles now work roughly 3 extra hours per week. Dave's bloat radar is his direct answer to this - every Dex feature must replace a manual process, not enable a new one. It's a design constraint, not a slogan.

Every meeting processed adds context to person pages. Every task completed refines the system's understanding of your priorities. Every mistake logged becomes a rule that prevents recurrence. Every preference noted makes future outputs more aligned with how you work.

After three months, your Dex knows your strategic pillars, your key relationships and their history, your working preferences, your common mistakes and how to avoid them, and the patterns in your industry that matter to your role.

Dave explains the mechanism: "Every time the AI then later pulls on that information, that particular entity, it's going to have all that fresh context. So you have these living files which then just get better and better and more useful for you." Unlike ChatGPT's stateless conversations, Dex maintains persistent entity pages — Project Pages auto-updated with Granola meeting actions and stakeholder feedback, Stakeholder Pages enriched by every interaction, and Company Pages with linked entity relationships maintained across all files.

This is the real promise of PKM that previous tools couldn't deliver. Not just storing knowledge, but compounding it. Making connections across information that a human brain can't hold in working memory. Surfacing the right context at exactly the right moment.

Key Takeaways:

Dave shares three critical prompting techniques that apply whether you're using Dex or any AI tool:

1. Goal Clarity Over Methodology
"Be very clear what your goal is when you're talking to the AI. Don't be vague about what you want it to do. The kindest thing you can do to the AI is give it a very clear goal." Don't describe how to do it — describe what success looks like.

2. Emotional Prompting
"Always push and push and push on your prompt and make it really emotional, and it really responds, particularly Opus 4.6. It gets quite excited with you when you really push." Dave's live PRD prompt included phrases like "brilliantly serendipitous" and "massively delightful" — high-energy prompting that produces dramatically better results than flat, corporate requests.

3. Voice-First Workflow
Dave uses Whisperflow.AI (recently funded $80M, press-to-talk, platform-agnostic) or Super Whisper (one-person company with mode-based prompts for different contexts — Google Doc feedback, email, Slack). Voice reduces input friction to near zero, making the whole system feel conversational rather than transactional.

Key Takeaways:

Dave Killeen hasn't just built a productivity tool — he's demonstrated what happens when you take PKM seriously and give it an AI backbone. As he puts it: "There has never been a better time to be a product geek." Here's what to take away:

Cross-references and further reading:

That’s a wrap.

As always, the journey doesn't end here!

Please share and let us know what you liked or want changing! 🚀👋

Alastair 🍽️.