Graveyard

Five years as the task app, then a database of my own, then that retired too. Linear is the queue.

Its abstractions were the thing I kept working around. Replaced by a provider layer and plain TypeScript.

It was another API to learn, and it is built for workflows where I want autonomy.

It got unwieldy past the simplest flows. The one automation that mattered saved my thoughts to a database and did not use AI at all.

It asked too many questions and I never trusted its timestamps. Replaced by a log command and a voice recorder.

Slow realtime and clunky access control for what I needed. A paid plan sat unused until 2026.

It was fast to move with, until I needed SQL and larger reads.

Left three times: no API in 2020, too heavy in 2021, and unable to hold an indexed version of anything in 2026. Replaced by markdown in a folder with a search index.

The best markdown editing I had found, and too tied to its app. The wiki I published from it is still up.

It needs a server, and I wanted a local object store.

Jira was slow on a phone, Todoist filled with items a bot extracted that I never wanted, and Trello sufficed.

Anki's activation cost was too high, and the Mochi bot I built collapsed under its own ambition. The shape survived as a design pattern.

It crashed too often, and Aside replaced it.

Evernote was hard to capture into and closed. Alfred lost to a launcher with a faster path to a model.

Sponsored credits expired and every call became real billing. Off behind one switch, with two exceptions that nothing else serves.

A whole product version. The idea may have been right; I did not have enough production experience with multi-agent systems to carry it, and once everything ran locally the actors were unnecessary.

Its use cases were narrow. Markdown became the high-level language instead. Partially revived as a hunch in August 2026.

Three months of work before pivoting. Generalizing a simple tool produced too many abstractions and nothing interesting beyond my own use case.

Retired once the goals report was computed from what actually happened, which made written tasks redundant.

Never used; essays were written elsewhere the whole time. The command now errors on purpose.

One log carrying every role at once was a poor version of each. Replaced by narrow views that do one thing. The lesson kept: start with freeform text and let the model interpret.

Adopted as an identity, and dropped once I could see I did not believe it would make money. Replaced by my own voice.

The vision did not change; the shape did. Replaced by things I operate rather than sell.

It was slipping within two weeks. I replaced it with building a tool only when a real friction demands it.

Businesses need glue and I was already building the glue I would otherwise rent.

Twenty-five minutes was too far away for the reward to matter. Replaced by smaller, earlier ones.

Thirty days of thirty minutes, completed once and restarted twice. It recurs, so it is not a graveyard item.

Fifteen hundred hours and counting.