Notes on cross-AI work.
Short essays on memory, prompts, sources, and the tools that turn scattered AI work into something worth returning to.
A calmer way to work with AI
Most AI software is loud. Endless notifications, infinite chat history, twenty open tabs of half-finished sessions. There's a quieter alternative. one where the work compounds instead of accumulates.
Read post →- Jun 7, 2026 6 min read
Why teams need shared memory for AI-assisted work
Five teammates with five private AI subscriptions equals zero institutional memory. The leverage AI tools offer doesn't survive turnover. Shared memory does.
Read post → - May 31, 2026 4 min read
When old prompts become new leverage
The prompt that produced great work last quarter is leverage you can use again. Most of us throw it away because we never learned to save the right one.
Read post → - May 24, 2026 5 min read
The difference between saving chats and remembering work
Saving is a storage problem. Remembering is a retrieval problem. If you can't find what you saved when you need it, you didn't really save it.
Read post → - May 17, 2026 5 min read
Decisions are the missing layer in AI work
We've gotten good at saving prompts and outputs. We've barely started saving the decisions those outputs informed. The gap is where six-month-old projects get re-litigated.
Read post → - May 10, 2026 6 min read
Source quality matters: building trust in your AI memory
When the answer is right but the citation is shaky, the answer doesn't ship. Treating sources as first-class objects. not afterthoughts. is what makes AI work credible.
Read post → - May 3, 2026 5 min read
The cost of context switching, and how to win it back
Every time you re-load context into a new AI session, you pay a tax. The tax is invisible until you measure it. and most of it can be reclaimed without changing tools.
Read post → - Apr 26, 2026 6 min read
ChatGPT for ideation, Claude for writing, Perplexity for sources
There is no single best AI tool. There's a best tool for each step. Here's what a serious cross-AI workflow looks like. and what breaks if you don't carry context between the steps.
Read post → - Apr 19, 2026 5 min read
Save the prompts that actually worked
Most prompt libraries are graveyards of speculative one-liners. The valuable ones are the prompts you've already pressure-tested with real work. Here's how Shelvia surfaces them automatically.
Read post → - Apr 12, 2026 4 min read
Why useful AI work disappears into old chats
You used Claude to draft a brief, Perplexity to fact-check it, Cursor to ship the code, ChatGPT to write the launch copy. Three weeks later, half of it is gone. The fix isn't another folder app.
Read post →
