Episode 12: Building Your Own Chief of Staff

Most conversations about AI focus on what the technology can do. This one is about what a person can do once they stop being a passive user and start directing the tools themselves.

Holly Joint tells the story of her own leap: from running Claude and ChatGPT on a desktop to standing up a dedicated server and building software with nothing but natural language. Ewan MacLeod, the self-confessed geek of the pair, plays interviewer here, drawing out a journey that will feel both aspirational and reassuringly human to anyone who has hesitated at the edge of this. Notably, Holly never warms to the phrase "vibe coding." What she is describing is something more deliberate: using plain English to direct an AI to make things that work.

The honesty is what makes it useful. Holly does not present a frictionless success story. She walks through the dead ends, the most memorable being days lost trying to use Notion as a database while the AI tools cheerfully helped her dig the hole deeper rather than suggesting she change course. The breakthrough came from a human nudge to switch to a SQL database, a reminder that judgement and a willingness to step back still matter. She also hits the limits of the tools' own knowledge: they were out of date on Notion's interface and useless on Apple's Shortcuts app, where the answer eventually came from a stranger on YouTube.

Out of all this emerges a working chief of staff app, built and redesigned by hand, surfacing her email, calendar, tasks and health data in a single window she controls completely. The thrill, she explains, is seeing output instantly, changing a colour or a feature in seconds, and watching the thing simply exist.

The episode's quiet centre is a distinction Holly draws between agents and agency. We talk endlessly about agents doing things for us, but rarely about the agency individuals gain to reshape how they work. Her example of photographing her father's hospital letters and turning them into calendar invites lands harder than any abstract argument, and points to real questions about software business models when anyone can build the apps they used to pay for.

There is restraint too. Holly keeps her setup deliberately read-only, unwilling to let an agent touch her email, WhatsApp or contacts, mindful that reputation is part of what is at stake. For her son, meanwhile, AI is simply a tool that lets his imagination run wild, a counter to the worry that these tools dull children's creativity.

She closes with practical advice for anyone tempted to try: start locally, build a game, lean in, and do not be intimidated by a screen that merely looks like coding.

Key Topics

• Moving from using AI tools to directing them to build real software
• Why "vibe coding" undersells natural-language software creation
• Learning through dead ends, and when AI tools fail to course-correct you
• The limits of AI knowledge on fast-changing third-party interfaces
• Building a personal chief of staff app to replace paid subscriptions
• The difference between AI agents and human agency
• Drawing a deliberate read-only line for reputation and safety
• Children using AI to enhance creativity rather than replace it

Links & References

• Anthropic (Claude, Claude Code) — https://www.anthropic.com
• ChatGPT (OpenAI) — https://openai.com
• DigitalOcean — https://www.digitalocean.com
• Notion — https://www.notion.com
• Termius — https://termius.com