My 7 principles for replacing paid apps with free skills
At 8 am this morning, I’m in the kitchen with a coffee, asking Claude what to work on first.
I haven’t opened Notion, Linear, LACRM, Things 3, Obsidian or any of the half-dozen to-do apps I’ve made my home over the years. I type three words into a terminal and Claude tells me what to do. It’s already read my projects, my journal from yesterday, and today’s calendar.
This is, I think, the most aggressive digital minimalism I’ve ever done. Not deleting apps, but categories of apps.
Let me back up.
A couple of weeks ago, I told you I’d replaced my CRM with a chat bar. What I didn’t say is that the CRM was the easy one. The CRM was where I started because the CRM was the most embarrassing—£20 a month for a dashboard that was, on a generous reading, hostile to actually getting anything done. Replacing it took a weekend.
The next thing to fall was project management. I had a Notion database with statuses and tags and a gallery view I’d lovingly arranged, mashed up with some projects in Things 3.
Now I have a folder of markdown files. Each one has a few lines of YAML at the top: status, area, next action, last updated. I don’t touch them; Claude does.
In the morning, I type /do what should I work on today? and Claude reads my values, projects, goals, progress, and then tells me where to point myself. When I finish something, I describe what happened, and the frontmatter rewrites itself. There’s no kanban, no tags.
Then the to-do app. I had a Things 3 list two hundred items deep, organised into projects and areas like an exhibition I was curating for an audience of one—me, in three months, who unfortunately would no longer care.
Currently, I’m using the same approach to rework my personal finances. I will not miss the three separate apps, nor the 40-tab spreadsheet.
Each rebuild looks the same. The starting impulse is to build a system, with a UI, and make it good. The correction is that the UI isn’t the point; the data and the conversation are.
Three rebuilds in, I have a solid set of principles I keep returning to. I hope you’ll find them useful in your own experiments:
The principles
The AI is the interface. Not under the interface. Not powering the interface. The interface. Don’t build a dashboard and bolt a chatbot to the side of it. Let the conversation do the work. Anything you want, you ask for. For most use cases, this will take the form of a skill (/morning, /crm, /triage, etc).
Build the data layer. Skip the presentation layer. AI needs structure—a database, a schema. It doesn’t need HTML. The minimum viable infrastructure is a place to store state and a way to query it. A folder of markdown files is often enough. I use SQLite where I need something more structured and searchable (e.g. for CRM contacts).
Do the work. Don’t build tools to help you do the work. The engineer’s instinct is to build a tool, then use the tool. Resist it. Have AI do the work now, with the thinnest possible scaffolding. Every hour spent on infrastructure is an hour not spent on the actual goal.
Ship in days. Let friction tell you what to build next. A database and one conversational skill is day one. Use it and start giving it feedback. These things are better built as you’re using them.
Automate the judgement, not the plumbing. Old automation moves data between systems. New automation makes calls: who needs attention, what to say, what’s falling behind. The value isn’t in the pipeline. It’s in the judgement.
Humans do human work. Connecting with people, sending the message, making the final call—that stays with me. Prioritising, tracking, remembering, summarising, nagging—that isn’t work I want to do. Let the robots do what they do best, and focus on the human interactions.
Don’t rebuild the old tool. Rebuild the workflow. The old tool was designed for a world without AI. A CRM replacement isn’t a better CRM. A Notion replacement isn’t a better Notion. A YNAB replacement isn’t a better YNAB. They all reduce, in this new world, to the same thing: a structured datastore and a conversation powered by a very smart agent.
The engineer in me still wants to build the dashboard. I still occasionally sketch out a beautiful kanban with smooth animations and a dark mode and, on one slightly unhinged morning, a custom font. I close the file.
I used to curate these systems for a future version of me who, it turned out, never showed up.
The conversation doesn’t need curating. It just needs me, this morning, with a coffee, asking what to do first.
—Dan