I had seventeen AI bookmarks by March. Half of them needed a new login every time OpenAI, Anthropic, or Google shipped a badge. None of them made Friday quieter.
Pick one default chat model, one coding helper, and one search/research tool. Run the same five work tasks for two weeks. Keep what lowers edit time. Delete the rest without a funeral.
The Monday FOMO tax
I'd open ChatGPT, then Claude, then whatever launched on Product Hunt. Same prompt. Slightly different tone. Same rewrite afterward.
The cost wasn't the subscription. It was context switching. I'd lose twenty minutes deciding which window "felt smarter."
What survived
- ChatGPT for messy first drafts and spreadsheet talk
- Claude when the doc is long and I need careful wording
- Cursor (or Copilot) only inside the repo — not as a second chat habit
That's it. No "AI meeting notes" tool until someone else owns the notes.
The test I use now
Five tasks. Same every week: contract summary, support reply, SOP rewrite, sheet explanation, code note.
I time how long until the output is shippable. If a new model doesn't beat the incumbent by a clear margin twice, it doesn't get a seat.
What I refuse
I refuse tools that want my whole Google Drive on day one. I refuse "agents" that email customers without a human gate. I refuse another dashboard that shows green scores and no unpaid invoices.
Close the extra tabs. Your stack can be boring and still pay rent.
