I recently gave a talk about this at LocalFirstConf. You can watch it on YouTube.
There’s a curious paradox in software right now.
Every company seems to be cramming AI into their products, yet the best apps for AI are often the ones with zero AI features.
Does anyone actually click these?
Notion is the latest example of a company pouring millions of dollars into reshaping its core product around AI. But it’s clear that people don’t like having it forced on them.
Meanwhile, Obsidian—Notion’s strongest competitor—has no mention of AI on its website, yet it has somehow captured people’s imagination and become the default choice for agentic notetaking.
The reason is simple: Obsidian is plaintext-first software. Your notes are just Markdown files in a directory, so any agent can read and edit them without any bespoke integration, through some simple Unix commands.
For me, plaintext-first tools like Obsidian have enabled far more advanced workflow automations than dedicated AI features ever have. Because agents love text.
This simple principle can also be extended beyond notes. I’ve been applying it to the calendar with my latest projects: caldir, a plaintext-first calendar system, and renCal, a GUI built on top of it.
When my calendar is just a directory of text files, I can easily do things like:
- Ask questions, e.g. “How many days did I spend abroad in the last 3 years?”
- Generate new events from any source, like a screenshot of a conference schedule.
None of these workflows required me to build an MCP server, integrate an LLM API, or add a special AI button to my app. Agents are already amazingly good at combining simple tools and adapting them to weirdly specific workflows, so all I needed was to keep the data in a format that’s as simple and human-readable as possible.
To me, this feels like a much bigger opportunity than adding chat assistants to everything. Plaintext-first apps give us the benefits of agentic automation without forcing us into somebody else’s AI ecosystem.
Build apps that agents can use, not apps that try to be agents.